diff --git a/.github/VOUCHED.td b/.github/VOUCHED.td index 90825a15f..1eb19c2af 100644 --- a/.github/VOUCHED.td +++ b/.github/VOUCHED.td @@ -67,6 +67,7 @@ inprealpha insodimension itertea itzrnvr +jaaneek jagravnaik jasonw22 jchristman diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 7dbe06010..00df0a797 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -622,10 +622,11 @@ jobs: with: node-version: "24" registry-url: "https://registry.npmjs.org" - # Keep npm aligned with trusted publishing setup (>= 11.16.0). + # npm runs under Bun when invoked by the release script; npm 12 + # requires a newer emulated Node version than Bun 1.3 provides. - name: Ensure npm supports trusted publishing if: ${{ !inputs.skip_npm }} - run: npm install -g npm@latest + run: npm install -g npm@11.17.0 - name: Cache bun dependencies uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0 with: @@ -775,9 +776,10 @@ jobs: with: node-version: "24" registry-url: "https://registry.npmjs.org" - # Keep npm aligned with trusted publishing setup (>= 11.16.0). + # npm runs under Bun when invoked by the release script; npm 12 + # requires a newer emulated Node version than Bun 1.3 provides. - name: Ensure npm supports trusted publishing - run: npm install -g npm@latest + run: npm install -g npm@11.17.0 - name: Cache bun dependencies uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0 with: diff --git a/Cargo.lock b/Cargo.lock index 1b29c8dce..a115a7785 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -1850,9 +1850,9 @@ checksum = "b9e0384b61958566e926dc50660321d12159025e767c18e043daf26b70104c39" [[package]] name = "ignore" -version = "0.4.27" +version = "0.4.28" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "fe112b004901c62c2faa11f4f75e9864e0cc5af8da71c9115d184a3aa888749f" +checksum = "2adf14691c72bcfc1058740436a35bdd3ae9c07d1a941ef00b749e9ea16aefa7" dependencies = [ "crossbeam-deque", "globset", @@ -1951,9 +1951,9 @@ dependencies = [ [[package]] name = "inotify" -version = "0.11.3" +version = "0.11.4" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "dd854a95a4ac672fed8c054136039fd32c22cf039ff09ead7280afe920486483" +checksum = "153be1941a183ec9ccd095ddbe17a8b8d435ef6c76e9e02451b933c3999af2c8" dependencies = [ "bitflags 2.13.0", "inotify-sys", @@ -2026,9 +2026,9 @@ checksum = "8f42a60cbdf9a97f5d2305f08a87dc4e09308d1276d28c869c684d7777685682" [[package]] name = "jiff" -version = "0.2.31" +version = "0.2.32" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "ccfe6121cbe750cf81efa362d85c0bde7ea298ec43092d3a193baca59cdbd634" +checksum = "961d16382652bfdd8c6f68b223b26a8c93e0d475c672f414411db31c6c5c900e" dependencies = [ "defmt", "jiff-static", @@ -2053,9 +2053,9 @@ dependencies = [ [[package]] name = "jiff-static" -version = "0.2.31" +version = "0.2.32" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "e165e897f662d428f3cd3828a919dbe067c2d42bb1031eede74ef9d27ecdedd2" +checksum = "d0879bd39df99c4c5e2c6615ccc026391a423dde10532c573e6086eb94a802cc" dependencies = [ "proc-macro2", "quote", @@ -2064,9 +2064,9 @@ dependencies = [ [[package]] name = "jiff-tzdb" -version = "0.1.7" +version = "0.1.8" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "6142247df1a93c2b3587402a19710be3e6e942f1581a1702e76408f2c21d6590" +checksum = "142bd39932ad231f10513df9ab62661fead8719872150b7ad02a2df79f4e141e" [[package]] name = "jiff-tzdb-platform" @@ -2881,7 +2881,7 @@ dependencies = [ [[package]] name = "pi-ast" -version = "16.3.12" +version = "16.3.14" dependencies = [ "anyhow", "ast-grep-core", @@ -2950,7 +2950,7 @@ dependencies = [ [[package]] name = "pi-iso" -version = "16.3.12" +version = "16.3.14" dependencies = [ "async-trait", "libc", @@ -2962,13 +2962,14 @@ dependencies = [ [[package]] name = "pi-natives" -version = "16.3.12" +version = "16.3.14" dependencies = [ "anyhow", "arboard", "ast-grep-core", "base64", "clap", + "clipboard-win", "flume", "fontdue", "globset", @@ -3014,7 +3015,7 @@ dependencies = [ [[package]] name = "pi-shell" -version = "16.3.12" +version = "16.3.14" dependencies = [ "anyhow", "brush-builtins", @@ -3063,7 +3064,7 @@ dependencies = [ [[package]] name = "pi-walker" -version = "16.3.12" +version = "16.3.14" dependencies = [ "dashmap", "globset", @@ -3403,9 +3404,9 @@ dependencies = [ [[package]] name = "regex" -version = "1.12.4" +version = "1.13.0" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "f1292b7759ae1cb9ec195452d1390a074f0cd8541ab7a5a8c31cd6db45d4a6ba" +checksum = "2a0e75113e14dc5acb068cd0786884f214f1312650a3d36d269f5c4f3cdee8a2" dependencies = [ "aho-corasick", "memchr", @@ -3415,9 +3416,9 @@ dependencies = [ [[package]] name = "regex-automata" -version = "0.4.14" +version = "0.4.15" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "6e1dd4122fc1595e8162618945476892eefca7b88c52820e74af6262213cae8f" +checksum = "1f388202e4b80542a0921078cc23b6333bcf1409c1e3f86404cae4766a6131db" dependencies = [ "aho-corasick", "memchr", @@ -5765,18 +5766,18 @@ dependencies = [ [[package]] name = "zerocopy" -version = "0.8.53" +version = "0.8.54" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "75726053136156d419e285b9b7eddaaea9e3fea6ce32eed44a89901f0bd98de1" +checksum = "b7cbbc0a705a0fd05cc3676525980d2bf5a9bc4adac6d6475209a7887cf59d19" dependencies = [ "zerocopy-derive", ] [[package]] name = "zerocopy-derive" -version = "0.8.53" +version = "0.8.54" source = "registry+https://github.com/rust-lang/crates.io-index" -checksum = "4714fd92cf900833d49538023a9b3915155210801d1c1169eba513b2addefd71" +checksum = "e2e817b7b52d0c7358d3246da9d69935ebb18116b2b102b4230dac079b4862f5" dependencies = [ "proc-macro2", "quote", diff --git a/Cargo.toml b/Cargo.toml index 88c64391b..4689b4be4 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -4,7 +4,7 @@ exclude = ["crates/vendor/brush-core", "crates/vendor/brush-builtins"] resolver = "3" [workspace.package] -version = "16.3.12" +version = "16.3.14" edition = "2024" license = "MIT" authors = ["Can Boluk"] @@ -269,6 +269,7 @@ napi-derive = "3" # Terminal & PTY # ────────────────────────────────────────────────────────────────────────────── arboard = { version = "3.6.1", features = ["wayland-data-control"] } +clipboard-win = "5.4" icy_sixel = "0.5" portable-pty = "0.9" diff --git a/bun.lock b/bun.lock index f97a68579..0d17797d5 100644 --- a/bun.lock +++ b/bun.lock @@ -21,7 +21,7 @@ }, "packages/agent": { "name": "@oh-my-pi/pi-agent-core", - "version": "16.3.12", + "version": "16.3.14", "dependencies": { "@oh-my-pi/pi-ai": "catalog:", "@oh-my-pi/pi-catalog": "catalog:", @@ -39,7 +39,7 @@ }, "packages/ai": { "name": "@oh-my-pi/pi-ai", - "version": "16.3.12", + "version": "16.3.14", "dependencies": { "@bufbuild/protobuf": "catalog:", "@oh-my-pi/pi-catalog": "catalog:", @@ -55,7 +55,7 @@ }, "packages/catalog": { "name": "@oh-my-pi/pi-catalog", - "version": "16.3.12", + "version": "16.3.14", "dependencies": { "@bufbuild/protobuf": "catalog:", "@oh-my-pi/pi-utils": "catalog:", @@ -69,7 +69,7 @@ }, "packages/coding-agent": { "name": "@oh-my-pi/pi-coding-agent", - "version": "16.3.12", + "version": "16.3.14", "bin": { "omp": "src/cli.ts", }, @@ -137,7 +137,7 @@ }, "packages/hashline": { "name": "@oh-my-pi/hashline", - "version": "16.3.12", + "version": "16.3.14", "dependencies": { "diff": "catalog:", "lru-cache": "catalog:", @@ -148,7 +148,7 @@ }, "packages/mnemopi": { "name": "@oh-my-pi/pi-mnemopi", - "version": "16.3.12", + "version": "16.3.14", "bin": { "mnemopi": "src/cli.ts", }, @@ -174,7 +174,7 @@ }, "packages/natives": { "name": "@oh-my-pi/pi-natives", - "version": "16.3.12", + "version": "16.3.14", "devDependencies": { "@napi-rs/cli": "catalog:", "@types/bun": "catalog:", @@ -182,7 +182,7 @@ }, "packages/snapcompact": { "name": "@oh-my-pi/snapcompact", - "version": "16.3.12", + "version": "16.3.14", "dependencies": { "@oh-my-pi/pi-ai": "catalog:", "@oh-my-pi/pi-natives": "catalog:", @@ -195,7 +195,7 @@ }, "packages/stats": { "name": "@oh-my-pi/omp-stats", - "version": "16.3.12", + "version": "16.3.14", "bin": { "omp-stats": "./src/index.ts", }, @@ -221,7 +221,7 @@ }, "packages/swarm-extension": { "name": "@oh-my-pi/swarm-extension", - "version": "16.3.12", + "version": "16.3.14", "bin": { "omp-swarm": "src/cli.ts", }, @@ -247,7 +247,7 @@ }, "packages/tui": { "name": "@oh-my-pi/pi-tui", - "version": "16.3.12", + "version": "16.3.14", "dependencies": { "@oh-my-pi/pi-natives": "catalog:", "@oh-my-pi/pi-utils": "catalog:", @@ -288,7 +288,7 @@ }, "packages/utils": { "name": "@oh-my-pi/pi-utils", - "version": "16.3.12", + "version": "16.3.14", "dependencies": { "@oh-my-pi/pi-natives": "catalog:", "handlebars": "catalog:", @@ -301,7 +301,7 @@ }, "packages/wire": { "name": "@oh-my-pi/pi-wire", - "version": "16.3.12", + "version": "16.3.14", "devDependencies": { "@types/bun": "catalog:", }, @@ -338,18 +338,18 @@ "@huggingface/transformers": "^4.2.0", "@mozilla/readability": "^0.6.0", "@napi-rs/cli": "3.7.0", - "@oh-my-pi/hashline": "16.3.12", - "@oh-my-pi/omp-stats": "16.3.12", - "@oh-my-pi/pi-agent-core": "16.3.12", - "@oh-my-pi/pi-ai": "16.3.12", - "@oh-my-pi/pi-catalog": "16.3.12", - "@oh-my-pi/pi-coding-agent": "16.3.12", - "@oh-my-pi/pi-mnemopi": "16.3.12", - "@oh-my-pi/pi-natives": "16.3.12", - "@oh-my-pi/pi-tui": "16.3.12", - "@oh-my-pi/pi-utils": "16.3.12", - "@oh-my-pi/pi-wire": "16.3.12", - "@oh-my-pi/snapcompact": "16.3.12", + "@oh-my-pi/hashline": "16.3.14", + "@oh-my-pi/omp-stats": "16.3.14", + "@oh-my-pi/pi-agent-core": "16.3.14", + "@oh-my-pi/pi-ai": "16.3.14", + "@oh-my-pi/pi-catalog": "16.3.14", + "@oh-my-pi/pi-coding-agent": "16.3.14", + "@oh-my-pi/pi-mnemopi": "16.3.14", + "@oh-my-pi/pi-natives": "16.3.14", + "@oh-my-pi/pi-tui": "16.3.14", + "@oh-my-pi/pi-utils": "16.3.14", + "@oh-my-pi/pi-wire": "16.3.14", + "@oh-my-pi/snapcompact": "16.3.14", "@opentelemetry/api": "^1.9.1", "@opentelemetry/context-async-hooks": "^2.7.1", "@opentelemetry/exporter-trace-otlp-proto": "^0.218.0", @@ -811,7 +811,7 @@ "@protobufjs/pool": ["@protobufjs/pool@1.1.0", "", {}, "sha512-0kELaGSIDBKvcgS4zkjz1PeddatrjYcmMWOlAuAPwAeccUrPHdUqo/J6LiymHHEiJT5NrF1UVwxY14f+fy4WQw=="], - "@protobufjs/utf8": ["@protobufjs/utf8@1.1.1", "", {}, "sha512-oOAWABowe8EAbMyWKM0tYDKi8Yaox52D+HWZhAIJqQXbqe0xI/GV7FhLWqlEKreMkfDjshR5FKgi3mnle0h6Eg=="], + "@protobufjs/utf8": ["@protobufjs/utf8@1.1.2", "", {}, "sha512-b1UQwcEZ4yCnMCD8DAL1VlbvBJE9/IX4FTIp7BG1xYpf29SLazLSrqUkj4w7Y5y7cCVP6E5tcqqcI0xemPkHug=="], "@puppeteer/browsers": ["@puppeteer/browsers@3.0.6", "", { "dependencies": { "modern-tar": "^0.7.6", "yargs": "^18.0.0" }, "peerDependencies": { "proxy-agent": ">=8.0.1", "yauzl": "^2.10.0 || ^3.4.0" }, "optionalPeers": ["proxy-agent", "yauzl"], "bin": { "browsers": "lib/main-cli.js" } }, "sha512-B/gKoqlFkzhvzsI6jo9K1cZz9o5ypviVv/xu8CwA4grZzyVwN+XfkT+tu8T1zrauuEXv6VhS2oGX+6NL95WcKA=="], @@ -963,7 +963,7 @@ "brace-expansion": ["brace-expansion@5.0.7", "", { "dependencies": { "balanced-match": "^4.0.2" } }, "sha512-7oFy703dxfY3/NLxC1fh2SUCQ0H9rmAY+5EpDVfXjUTTs+HEwR2nYaqLv+GWcTsumwxPfiz6CzCNkwXwBUwqCA=="], - "browserslist": ["browserslist@4.28.4", "", { "dependencies": { "baseline-browser-mapping": "^2.10.38", "caniuse-lite": "^1.0.30001799", "electron-to-chromium": "^1.5.376", "node-releases": "^2.0.48", "update-browserslist-db": "^1.2.3" }, "bin": { "browserslist": "cli.js" } }, "sha512-MTc8i/x9jBQd1iMw2CFGS+rwMa07eYjLR0CCTLDACl9xhxy+nIs3KeML/biicXtk9JrZ6dnnTatmc7ErPXIxqw=="], + "browserslist": ["browserslist@4.28.5", "", { "dependencies": { "baseline-browser-mapping": "^2.10.42", "caniuse-lite": "^1.0.30001800", "electron-to-chromium": "^1.5.387", "node-releases": "^2.0.50", "update-browserslist-db": "^1.2.3" }, "bin": { "browserslist": "cli.js" } }, "sha512-Cu2E6QejHWzuDMTkuwgpABFgDfZrXLQq5V13YOACZx4mFAG4IwGTbTfHPMr4WtxlHoXSM8FIuRwYYCz5XiabaQ=="], "bun-types": ["bun-types@1.3.14", "", { "dependencies": { "@types/node": "*" } }, "sha512-4N0ig0fEomHt5R0KCFWjovxow98rIoRwKolrYdCcknNwMekCXRnWEUvgu5soYV8QXtVsrUD8B95MBOZGPvr6KQ=="], diff --git a/crates/pi-natives/Cargo.toml b/crates/pi-natives/Cargo.toml index 6495a02c8..0f230bcd6 100644 --- a/crates/pi-natives/Cargo.toml +++ b/crates/pi-natives/Cargo.toml @@ -26,7 +26,7 @@ grep-searcher.workspace = true html-to-markdown-rs.workspace = true icy_sixel.workspace = true ignore.workspace = true -image.workspace = true +image = { workspace = true, features = ["bmp"] } inferno.workspace = true memmap2.workspace = true napi.workspace = true @@ -61,6 +61,7 @@ libc.workspace = true [target.'cfg(windows)'.dependencies] windows-sys = { workspace = true, features = ["Wdk_Storage_FileSystem", "Win32_Security"] } +clipboard-win.workspace = true winreg.workspace = true [build-dependencies] diff --git a/crates/pi-natives/src/clipboard.rs b/crates/pi-natives/src/clipboard.rs index 060cae6c7..c7b2251b2 100644 --- a/crates/pi-natives/src/clipboard.rs +++ b/crates/pi-natives/src/clipboard.rs @@ -30,7 +30,14 @@ fn encode_png(image: ImageData<'_>) -> Result> { let bytes = image.bytes.into_owned(); let buffer = RgbaImage::from_raw(width, height, bytes) .ok_or_else(|| Error::from_reason("Clipboard image buffer size mismatch"))?; - let capacity = width.saturating_mul(height).saturating_mul(4) as usize; + rgba_to_png(buffer) +} + +fn rgba_to_png(buffer: RgbaImage) -> Result> { + let capacity = (buffer + .width() + .saturating_mul(buffer.height()) + .saturating_mul(4)) as usize; let mut output = Vec::with_capacity(capacity); DynamicImage::ImageRgba8(buffer) .write_to(&mut Cursor::new(&mut output), ImageFormat::Png) @@ -38,6 +45,88 @@ fn encode_png(image: ImageData<'_>) -> Result> { Ok(output) } +/// Decode a packed DIB clipboard payload (`CF_DIB`: a `BITMAPINFOHEADER`-family +/// header, optional bitfield masks and palette, then the pixel array) into PNG +/// bytes. +/// +/// The payload is wrapped in a synthesized `BITMAPFILEHEADER` and decoded +/// through the BMP *file* path so the explicit `bfOffBits` pins the pixel +/// offset. This matters: the header-less decode path arboard uses mis-places +/// the pixel offset for V4/V5 headers with `BI_BITFIELDS` compression (it +/// skips 12 trailing mask bytes that those headers embed instead), which is +/// why Qt-based screenshot tools (`PixPin`, `Snipaste`, ...) fail through +/// arboard in the first place (#3426). +#[cfg_attr( + not(windows), + allow( + dead_code, + reason = "reached only by the Windows clipboard fallback; kept target-independent so unit \ + tests cover it on every host" + ) +)] +fn dib_to_png(dib: &[u8]) -> Result> { + const FILE_HEADER_SIZE: u64 = 14; + const INFO_HEADER_SIZE: u64 = 40; + const BI_BITFIELDS: u32 = 3; + + if dib.len() < INFO_HEADER_SIZE as usize { + return Err(Error::from_reason("Clipboard DIB shorter than BITMAPINFOHEADER")); + } + let u32_at = + |at: usize| u32::from_le_bytes(dib[at..at + 4].try_into().expect("bounds checked above")); + let header_size = u64::from(u32_at(0)); + if header_size < INFO_HEADER_SIZE || header_size > dib.len() as u64 { + return Err(Error::from_reason("Clipboard DIB header size out of range")); + } + let bit_count = u16::from_le_bytes([dib[14], dib[15]]); + let compression = u32_at(16); + let colors_used = u64::from(u32_at(32)); + + // A plain BITMAPINFOHEADER with BI_BITFIELDS is trailed by three DWORD + // masks; larger (V2..V5) headers embed the masks in the header itself. + let mask_bytes: u64 = if header_size == INFO_HEADER_SIZE && compression == BI_BITFIELDS { + 12 + } else { + 0 + }; + let palette_entries: u64 = if colors_used != 0 { + colors_used + } else if bit_count <= 8 { + 1u64 << bit_count + } else { + 0 + }; + let pixel_offset = + u32::try_from(FILE_HEADER_SIZE + header_size + mask_bytes + palette_entries * 4) + .map_err(|_| Error::from_reason("Clipboard DIB layout overflow"))?; + let file_size = u32::try_from(FILE_HEADER_SIZE + dib.len() as u64) + .map_err(|_| Error::from_reason("Clipboard DIB too large"))?; + + let mut bmp = Vec::with_capacity(FILE_HEADER_SIZE as usize + dib.len()); + bmp.extend_from_slice(b"BM"); + bmp.extend_from_slice(&file_size.to_le_bytes()); + bmp.extend_from_slice(&0u32.to_le_bytes()); + bmp.extend_from_slice(&pixel_offset.to_le_bytes()); + bmp.extend_from_slice(dib); + + let decoded = image::load_from_memory_with_format(&bmp, ImageFormat::Bmp) + .map_err(|err| Error::from_reason(format!("Failed to decode clipboard DIB: {err}")))?; + rgba_to_png(decoded.into_rgba8()) +} + +/// Read the raw `CF_DIB` bytes from the Windows clipboard. +/// +/// Windows synthesizes `CF_DIB` from whatever bitmap formats are present, so +/// it is available whenever the clipboard holds any image at all. +#[cfg(windows)] +fn read_raw_cf_dib() -> Option> { + let clip = clipboard_win::Clipboard::new_attempts(10).ok()?; + let mut dib = Vec::new(); + clipboard_win::raw::get_vec(clipboard_win::formats::CF_DIB, &mut dib).ok()?; + drop(clip); + (!dib.is_empty()).then_some(dib) +} + /// Copy plain text to the system clipboard. /// /// # Parameters @@ -120,7 +209,160 @@ pub fn read_image_from_clipboard() -> task::Promise> { })) }, Err(ClipboardError::ContentNotAvailable) => Ok(None), - Err(err) => Err(Error::from_reason(format!("Failed to read clipboard image: {err}"))), + Err(err) => { + // arboard rejects the CF_DIBV5 payloads Qt-based screenshot + // tools (PixPin, Snipaste, ...) produce; decode the raw CF_DIB + // ourselves before surfacing the error (#3426). A fallback + // decode failure keeps the original arboard error. + #[cfg(windows)] + if let Some(bytes) = read_raw_cf_dib().and_then(|dib| dib_to_png(&dib).ok()) { + return Ok(Some(ClipboardImage { + data: Uint8Array::from(bytes), + mime_type: "image/png".to_string(), + })); + } + Err(Error::from_reason(format!("Failed to read clipboard image: {err}"))) + }, } }) } + +#[cfg(test)] +mod tests { + use super::dib_to_png; + + fn push32(v: u32, out: &mut Vec) { + out.extend_from_slice(&v.to_le_bytes()); + } + + fn push16(v: u16, out: &mut Vec) { + out.extend_from_slice(&v.to_le_bytes()); + } + + /// 2x2 bottom-up BGRA pixel array: memory rows are [red, green] (bottom) + /// then [blue, white] (top), all with alpha 0xff. + const PIXELS_2X2: [u8; 16] = [ + 0x00, 0x00, 0xff, 0xff, // (0,1) red + 0x00, 0xff, 0x00, 0xff, // (1,1) green + 0xff, 0x00, 0x00, 0xff, // (0,0) blue + 0xff, 0xff, 0xff, 0xff, // (1,0) white + ]; + + /// `CF_DIB` as Qt's clipboard writer emits it for 32-bit content: a plain + /// `BITMAPINFOHEADER` with `BI_BITFIELDS` compression and three DWORD + /// masks between header and pixels. + fn qt_cf_dib(width: u32, height: u32, pixels_bgra: &[u8], compression: u32) -> Vec { + let mut d = Vec::with_capacity(52 + pixels_bgra.len()); + push32(40, &mut d); // biSize + push32(width, &mut d); + push32(height, &mut d); // positive: bottom-up + push16(1, &mut d); // biPlanes + push16(32, &mut d); // biBitCount + push32(compression, &mut d); + push32(pixels_bgra.len() as u32, &mut d); // biSizeImage + push32(0, &mut d); // biXPelsPerMeter + push32(0, &mut d); // biYPelsPerMeter + push32(0, &mut d); // biClrUsed + push32(0, &mut d); // biClrImportant + if compression == 3 { + push32(0x00ff_0000, &mut d); // red mask + push32(0x0000_ff00, &mut d); // green mask + push32(0x0000_00ff, &mut d); // blue mask + } + d.extend_from_slice(pixels_bgra); + d + } + + /// `CF_DIBV5` as PixPin (Qt) places it, after arboard's + /// `maybe_tweak_header` rewrite: a 124-byte `BITMAPV5HEADER` carrying + /// `BI_BITFIELDS` compression with the BGRA masks embedded in the header + /// and pixels immediately after it. This is the exact buffer shape that + /// arboard's header-less BMP decode rejects with `ConversionFailure` + /// (issue #3426); the file-header wrap must decode it. + fn pixpin_dibv5_tweaked(width: u32, height: u32, pixels_bgra: &[u8]) -> Vec { + let mut d = Vec::with_capacity(124 + pixels_bgra.len()); + push32(124, &mut d); // bV5Size + push32(width, &mut d); + push32(height, &mut d); + push16(1, &mut d); // bV5Planes + push16(32, &mut d); // bV5BitCount + push32(3, &mut d); // bV5Compression = BI_BITFIELDS (arboard-tweaked) + push32(0, &mut d); // bV5SizeImage + push32(0, &mut d); // bV5XPelsPerMeter + push32(0, &mut d); // bV5YPelsPerMeter + push32(0, &mut d); // bV5ClrUsed + push32(0, &mut d); // bV5ClrImportant + push32(0x00ff_0000, &mut d); // bV5RedMask + push32(0x0000_ff00, &mut d); // bV5GreenMask + push32(0x0000_00ff, &mut d); // bV5BlueMask + push32(0xff00_0000, &mut d); // bV5AlphaMask + push32(0x7352_4742, &mut d); // bV5CSType = LCS_sRGB + d.extend_from_slice(&[0u8; 36]); // bV5Endpoints + push32(0, &mut d); // bV5GammaRed + push32(0, &mut d); // bV5GammaGreen + push32(0, &mut d); // bV5GammaBlue + push32(4, &mut d); // bV5Intent = LCS_GM_IMAGES + push32(0, &mut d); // bV5ProfileData + push32(0, &mut d); // bV5ProfileSize + push32(0, &mut d); // bV5Reserved + assert_eq!(d.len(), 124); + d.extend_from_slice(pixels_bgra); + d + } + + fn decode_pixels(png: &[u8]) -> (u32, u32, Vec<[u8; 4]>) { + let img = image::load_from_memory(png).expect("fallback output must be valid PNG"); + let rgba = img.into_rgba8(); + let (w, h) = rgba.dimensions(); + let px = rgba.pixels().map(|p| p.0).collect(); + (w, h, px) + } + + const RED: [u8; 4] = [255, 0, 0, 255]; + const GREEN: [u8; 4] = [0, 255, 0, 255]; + const BLUE: [u8; 4] = [0, 0, 255, 255]; + const WHITE: [u8; 4] = [255, 255, 255, 255]; + + #[test] + fn decodes_qt_cf_dib_with_bitfields_masks() { + let dib = qt_cf_dib(2, 2, &PIXELS_2X2, 3); + let png = dib_to_png(&dib).expect("BI_BITFIELDS CF_DIB must decode"); + let (w, h, px) = decode_pixels(&png); + assert_eq!((w, h), (2, 2)); + // Row order flipped versus the bottom-up pixel array; BGRA -> RGBA. + assert_eq!(px, vec![BLUE, WHITE, RED, GREEN]); + } + + #[test] + fn decodes_pixpin_dibv5_payload_that_arboard_rejects() { + let dib = pixpin_dibv5_tweaked(2, 2, &PIXELS_2X2); + let png = dib_to_png(&dib).expect("V5 BI_BITFIELDS DIB must decode"); + let (w, h, px) = decode_pixels(&png); + assert_eq!((w, h), (2, 2)); + assert_eq!(px, vec![BLUE, WHITE, RED, GREEN]); + } + + #[test] + fn decodes_plain_bi_rgb_dib() { + // The common "copy image" payload: BI_RGB, 32-bit, no masks. The + // fourth byte is unused per the DIB contract — zero it to prove the + // decode still yields opaque pixels. + let mut pixels = PIXELS_2X2; + for alpha in pixels.iter_mut().skip(3).step_by(4) { + *alpha = 0; + } + let dib = qt_cf_dib(2, 2, &pixels, 0); + let png = dib_to_png(&dib).expect("BI_RGB CF_DIB must decode"); + let (w, h, px) = decode_pixels(&png); + assert_eq!((w, h), (2, 2)); + assert_eq!(px, vec![BLUE, WHITE, RED, GREEN]); + } + + #[test] + fn rejects_malformed_dib() { + assert!(dib_to_png(&[0u8; 12]).is_err(), "short buffer must not decode"); + let mut oversized_header = qt_cf_dib(2, 2, &PIXELS_2X2, 3); + oversized_header[0..4].copy_from_slice(&0xffff_ffffu32.to_le_bytes()); + assert!(dib_to_png(&oversized_header).is_err(), "header size beyond buffer must not decode"); + } +} diff --git a/crates/pi-natives/src/lib.rs b/crates/pi-natives/src/lib.rs index 7699d1fb1..3a0964bf5 100644 --- a/crates/pi-natives/src/lib.rs +++ b/crates/pi-natives/src/lib.rs @@ -248,7 +248,7 @@ fn create_windows_napi_tokio_runtime() -> Option { /// MUST stay in sync with `VERSION_SENTINEL_EXPORT` in /// `packages/natives/native/index.js` (which derives the name from /// `package.json#version`). -#[napi(js_name = "__piNativesV16_3_12")] +#[napi(js_name = "__piNativesV16_3_14")] pub const fn pi_natives_version_sentinel() {} /// Native module entry point: install crash diagnostics before any tool can diff --git a/crates/pi-natives/src/shell.rs b/crates/pi-natives/src/shell.rs index 47402b5fc..e159ba6ed 100644 --- a/crates/pi-natives/src/shell.rs +++ b/crates/pi-natives/src/shell.rs @@ -5,7 +5,7 @@ use std::{collections::HashMap, sync::Arc}; use napi::{ Env, Result, bindgen_prelude::*, - threadsafe_function::{ThreadsafeFunction, ThreadsafeFunctionCallMode}, + threadsafe_function::{ThreadsafeFunction, UnknownReturnValue}, }; use napi_derive::napi; use pi_shell::{ @@ -216,7 +216,7 @@ impl Shell { env: &'env Env, options: ShellRunOptions<'env>, #[napi(ts_arg_type = "((error: Error | null, chunk: string) => void) | undefined | null")] - on_chunk: Option>, + on_chunk: Option>, ) -> Result> { let cancel_token = task::CancelToken::new(options.timeout_ms, options.signal); let inner = Arc::clone(&self.inner); @@ -269,7 +269,7 @@ pub fn execute_shell<'env>( env: &'env Env, options: ShellExecuteOptions<'env>, #[napi(ts_arg_type = "((error: Error | null, chunk: string) => void) | undefined | null")] - on_chunk: Option>, + on_chunk: Option>, ) -> Result> { let cancel_token = task::CancelToken::new(options.timeout_ms, options.signal); let exec_options = CoreShellExecuteOptions { @@ -294,42 +294,66 @@ pub fn execute_shell<'env>( }) } +/// Capacity (in chunks) of the queue between the pipe readers and the JS +/// forwarding pump. One queued chunk is at most one pipe read (≤64 KiB), so +/// the Rust side of the bridge holds ~4 MiB worst case before the readers' +/// `send_async` parks — which in turn parks the child on its stdout/stderr +/// pipe (ordinary pipe backpressure) instead of buffering the surplus in +/// process memory (#4078). +const BRIDGE_QUEUE_CHUNKS: usize = 64; + fn bridge_chunks( - on_chunk: Option>, + on_chunk: Option>, ) -> (Option>, Option>) { let Some(on_chunk) = on_chunk else { return (None, None); }; - let (tx, rx) = flume::unbounded::(); - let handle = napi::tokio::spawn(async move { - // Hard cap on one coalesced batch so the JS main thread never sees a - // multi-MB napi callback (a giant single string would stall sanitize + - // tail-buffer maintenance for the whole copy). - const MAX_BATCH_BYTES: usize = 64 * 1024; - // Initial capacity sized for typical bursty pipe output. Re-allocated - // each batch because `String` ownership is moved into the napi call. - const INITIAL_BATCH_CAP: usize = 8 * 1024; - let mut batch = String::with_capacity(INITIAL_BATCH_CAP); - while let Ok(first) = rx.recv_async().await { - batch.push_str(&first); - // Greedily drain everything already queued. Child processes that - // write byte-at-a-time (printf-style progress, llama-cli token - // streams) otherwise produce one napi callback per `write(2)`, - // saturating the JS main thread (~200% CPU observed) and leaving - // the queue draining long after the child exits. - while batch.len() < MAX_BATCH_BYTES { - match rx.try_recv() { - Ok(more) => batch.push_str(&more), - Err(_) => break, - } - } - let payload = std::mem::replace(&mut batch, String::with_capacity(INITIAL_BATCH_CAP)); - on_chunk.call(Ok(payload), ThreadsafeFunctionCallMode::NonBlocking); - } - }); + let (tx, rx) = flume::bounded::(BRIDGE_QUEUE_CHUNKS); + let handle = napi::tokio::spawn(pump_chunks(rx, async move |payload: String| { + // `call_async` resolves only after the JS callback ran, so at most + // one batch sits in the napi queue at a time and the JS event loop's + // actual consumption rate backpressures the whole pipeline. An error + // means the JS side is gone (env teardown) — stop forwarding. + on_chunk.call_async(Ok(payload)).await.is_ok() + })); (Some(tx), Some(handle)) } +/// Drain `rx`, greedily coalescing queued chunks into ≤64 KiB batches, and +/// feed each batch to `forward`, awaiting its completion before pulling more. +/// Returns when `rx` disconnects (all senders dropped) or `forward` reports +/// the consumer is gone; dropping `rx` then disconnects the channel so +/// parked/future senders fail fast and the pipe readers keep draining the +/// child instead of wedging it. +async fn pump_chunks(rx: flume::Receiver, mut forward: impl AsyncFnMut(String) -> bool) { + // Hard cap on one coalesced batch so the JS main thread never sees a + // multi-MB napi callback (a giant single string would stall sanitize + + // tail-buffer maintenance for the whole copy). + const MAX_BATCH_BYTES: usize = 64 * 1024; + // Initial capacity sized for typical bursty pipe output. Re-allocated + // each batch because `String` ownership is moved into the napi call. + const INITIAL_BATCH_CAP: usize = 8 * 1024; + let mut batch = String::with_capacity(INITIAL_BATCH_CAP); + while let Ok(first) = rx.recv_async().await { + batch.push_str(&first); + // Greedily drain everything already queued. Child processes that + // write byte-at-a-time (printf-style progress, llama-cli token + // streams) otherwise produce one napi callback per `write(2)`, + // saturating the JS main thread (~200% CPU observed) and leaving + // the queue draining long after the child exits. + while batch.len() < MAX_BATCH_BYTES { + match rx.try_recv() { + Ok(more) => batch.push_str(&more), + Err(_) => break, + } + } + let payload = std::mem::replace(&mut batch, String::with_capacity(INITIAL_BATCH_CAP)); + if !forward(payload).await { + return; + } + } +} + /// Result of [`apply_bash_fixups`]: a possibly-rewritten command plus the /// substrings that were removed (in source order). #[napi(object)] @@ -360,7 +384,6 @@ pub fn apply_bash_fixups(command: String) -> BashFixupResult { mod tests { use std::time::Duration; - #[cfg(unix)] use flume; use pi_shell::{ ShellRunOptions as CoreShellRunOptions, @@ -368,7 +391,83 @@ mod tests { }; use tokio::time; - use super::CoreShell; + use super::{BRIDGE_QUEUE_CHUNKS, CoreShell, pump_chunks}; + + /// Regression for #4078: the reader→JS bridge queue must stay bounded when + /// the JS side (here: a deliberately slow `forward`) cannot keep up with a + /// fast producer, and backpressure must never drop or reorder chunks. On + /// the pre-fix bridge (`flume::unbounded` + fire-and-forget + /// `ThreadsafeFunctionCallMode::NonBlocking`) the same harness accumulates + /// the producer's entire surplus in the queue (measured: a 32 MiB stream + /// queued all 33_554_432 bytes while the consumer stalled). + #[tokio::test(flavor = "multi_thread")] + async fn bridge_pump_bounds_queue_and_delivers_all_bytes() { + const CHUNKS: usize = 512; + const CHUNK_BYTES: usize = 4096; + let (tx, rx) = flume::bounded::(BRIDGE_QUEUE_CHUNKS); + let producer = tokio::spawn(async move { + let mut expected = String::with_capacity(CHUNKS * CHUNK_BYTES); + let mut max_queued = 0usize; + for i in 0..CHUNKS { + let chunk = format!("[{i:06}]{}", "x".repeat(CHUNK_BYTES - 8)); + expected.push_str(&chunk); + tx.send_async(chunk) + .await + .expect("pump should outlive the producer"); + max_queued = max_queued.max(tx.len()); + } + (expected, max_queued) + }); + + let mut received = String::with_capacity(CHUNKS * CHUNK_BYTES); + time::timeout( + Duration::from_secs(30), + pump_chunks(rx, async |payload: String| { + received.push_str(&payload); + // Emulate a busy JS event loop: each napi callback takes a while. + time::sleep(Duration::from_micros(500)).await; + true + }), + ) + .await + .expect("pump should finish once the producer hangs up"); + + let (expected, max_queued) = producer.await.expect("producer task"); + assert!( + max_queued <= BRIDGE_QUEUE_CHUNKS, + "bridge queue grew past its bound: {max_queued} chunks", + ); + assert_eq!(received.len(), expected.len(), "bytes were dropped or duplicated"); + assert_eq!(received, expected, "chunks must arrive losslessly and in order"); + } + + /// When the JS side dies (`forward` fails: threadsafe function aborted on + /// env teardown), the pump must drop its receiver so parked and future + /// sends fail fast — the pipe readers keep draining the child instead of + /// wedging it on a full bridge queue. + #[tokio::test(flavor = "multi_thread")] + async fn bridge_pump_death_disconnects_channel_without_blocking_senders() { + let (tx, rx) = flume::bounded::(4); + let pump = tokio::spawn(pump_chunks(rx, async |_payload: String| false)); + let producer = tokio::spawn(async move { + let mut disconnected = 0usize; + for _ in 0..64 { + if tx.send_async("x".repeat(1024)).await.is_err() { + disconnected += 1; + } + } + disconnected + }); + let disconnected = time::timeout(Duration::from_secs(5), producer) + .await + .expect("sends must not park once the consumer died") + .expect("producer task"); + assert!(disconnected > 0, "channel should disconnect after the pump stops"); + time::timeout(Duration::from_secs(5), pump) + .await + .expect("pump should exit after forward fails") + .expect("pump task"); + } mod child_session_action_tests { use pi_shell::{ChildSessionAction, child_session_action}; diff --git a/crates/pi-shell/src/shell.rs b/crates/pi-shell/src/shell.rs index 80634bf91..b3ebe92b8 100644 --- a/crates/pi-shell/src/shell.rs +++ b/crates/pi-shell/src/shell.rs @@ -1649,7 +1649,7 @@ async fn read_output( let pending = &buf[..it]; match str::from_utf8(pending) { Ok(text) => { - emit_chunk(text, on_chunk.as_ref()); + emit_chunk(text, on_chunk.as_ref()).await; it = 0; break; }, @@ -1658,7 +1658,7 @@ async fn read_output( if p > 0 { // SAFETY: [..p] is guaranteed valid UTF-8 by valid_up_to(). let text = unsafe { str::from_utf8_unchecked(&pending[..p]) }; - emit_chunk(text, on_chunk.as_ref()); + emit_chunk(text, on_chunk.as_ref()).await; // copy p..it to the beginning of the buffer buf.copy_within(p..it, 0); it -= p; @@ -1667,7 +1667,7 @@ async fn read_output( match err.error_len() { Some(p) => { // Invalid byte sequence: emit replacement and drop those bytes. - emit_chunk(REPLACEMENT, on_chunk.as_ref()); + emit_chunk(REPLACEMENT, on_chunk.as_ref()).await; // copy p..it to the beginning of the buffer buf.copy_within(p..it, 0); it -= p; @@ -1688,10 +1688,10 @@ async fn read_output( for chunk in buf[..it].utf8_chunks() { let valid = chunk.valid(); if !valid.is_empty() { - emit_chunk(valid, on_chunk.as_ref()); + emit_chunk(valid, on_chunk.as_ref()).await; } if !chunk.invalid().is_empty() { - emit_chunk(REPLACEMENT, on_chunk.as_ref()); + emit_chunk(REPLACEMENT, on_chunk.as_ref()).await; } } } @@ -1777,7 +1777,7 @@ async fn read_output_buffered( while !pending.is_empty() { match str::from_utf8(&pending) { Ok(text) => { - emit_chunk(text, Some(cb)); + emit_chunk(text, Some(cb)).await; pending.clear(); break; }, @@ -1786,12 +1786,12 @@ async fn read_output_buffered( if p > 0 { // SAFETY: [..p] is valid UTF-8 per valid_up_to(). let text = unsafe { str::from_utf8_unchecked(&pending[..p]) }; - emit_chunk(text, Some(cb)); + emit_chunk(text, Some(cb)).await; pending.drain(..p); } match err.error_len() { Some(skip) => { - emit_chunk(REPLACEMENT, Some(cb)); + emit_chunk(REPLACEMENT, Some(cb)).await; pending.drain(..skip); }, None => break, @@ -1807,10 +1807,10 @@ async fn read_output_buffered( for chunk in pending.utf8_chunks() { let valid = chunk.valid(); if !valid.is_empty() { - emit_chunk(valid, Some(cb)); + emit_chunk(valid, Some(cb)).await; } if !chunk.invalid().is_empty() { - emit_chunk(REPLACEMENT, Some(cb)); + emit_chunk(REPLACEMENT, Some(cb)).await; } } } @@ -1858,9 +1858,16 @@ fn read_nonblocking(file: &T, buf: &mut [u8]) -> io::Re } } -fn emit_chunk(text: &str, callback: Option<&Sender>) { +/// Forward one decoded chunk to the streaming callback, honouring channel +/// backpressure: on a bounded channel (the pi-natives JS bridge) the send +/// parks until the consumer frees a slot — which parks the pipe reader and, +/// transitively, the child on its stdout/stderr pipe — so a fast producer +/// can never buffer unbounded output in memory (#4078). A disconnected +/// receiver (consumer gone) fails immediately, so the pipe keeps draining +/// and the child never wedges on a full pipe. +async fn emit_chunk(text: &str, callback: Option<&Sender>) { if let Some(callback) = callback { - let _ = callback.send(text.to_string()); + let _ = callback.send_async(text.to_string()).await; } } @@ -4140,4 +4147,39 @@ replace = [{ pattern = "^.+$", replacement = "PWD" }] "builtin nohup masked SIGHUP like the external tool (output: {out:?})", ); } + + /// Regression for #4078: the JS bridge hands the pipe readers a *bounded* + /// chunk channel. With a consumer slower than the producer the readers + /// must park on `send_async` (backpressuring the child through its pipe) + /// rather than buffer unboundedly — and, unlike a drop-on-full design, + /// every produced byte must still reach the consumer. + #[cfg(unix)] + #[tokio::test(flavor = "multi_thread")] + async fn streaming_output_backpressures_on_bounded_channel_without_loss() { + const TOTAL_BYTES: usize = 1_048_576; + let (tx, rx) = flume::bounded::(4); + let options = ShellExecuteOptions { + command: format!("yes x | head -c {TOTAL_BYTES}"), + ..Default::default() + }; + let run = tokio::spawn(execute_shell(options, Some(tx), CancelToken::default())); + + let mut received = 0usize; + while let Ok(chunk) = rx.recv_async().await { + received += chunk.len(); + // Slow consumer: forces the bounded queue to fill and the readers + // to park between chunks. + time::sleep(Duration::from_micros(50)).await; + } + + let result = time::timeout(Duration::from_secs(30), run) + .await + .expect("command should finish despite backpressure") + .expect("run task should not panic") + .expect("execute should succeed"); + assert_eq!(result.exit_code, Some(0)); + assert!(!result.cancelled); + assert!(!result.timed_out); + assert_eq!(received, TOTAL_BYTES, "streamed bytes were dropped under backpressure"); + } } diff --git a/docs/config-usage.md b/docs/config-usage.md index 6606618c2..984c08ee7 100644 --- a/docs/config-usage.md +++ b/docs/config-usage.md @@ -79,6 +79,8 @@ A named profile (`omp --profile `, the `--alias` shortcut, or `OMP_PROFILE The relocation is uniform across the native provider (`builtin.ts`) and the generic `config.ts` helpers, so it covers slash commands, rules, prompts, instructions, hooks, tools, extensions, settings, skills, and MCP, plus the top-level `SYSTEM.md` / `RULES.md` / `AGENTS.md` files and runtime state (sessions, blobs, `agent.db`). A profile sees only its own OMP config, never the default profile's `~/.omp/agent`. +Keybindings are the one exception: a named profile merges the default profile's `~/.omp/agent/keybindings.*` under its own `~/.omp/profiles//agent/keybindings.*`, with the profile file overriding per binding ([#4867](https://github.com/can1357/oh-my-pi/issues/4867)). Keybindings describe the terminal/keyboard in front of the user, which doesn't change with the active profile, so user-level remaps keep working in every profile unless the profile explicitly overrides them. The inherited file is read-only for the profile process — legacy-format migration of the default profile's file only happens when the default profile itself runs. + The other source bases are not profile-scoped and load identically under every profile: the external-tool bases (`~/.claude`, `~/.codex`, `~/.gemini`) belong to those tools, and the project-level bases (`/.omp`, `/.claude`, ...) are keyed to the working directory. Throughout this document, read `~/.omp/agent` as shorthand for the active profile's agent directory. ## Important constraint diff --git a/docs/extensions.md b/docs/extensions.md index 337a1d65c..7702c1767 100644 --- a/docs/extensions.md +++ b/docs/extensions.md @@ -142,7 +142,7 @@ Also exposed: - `deliverAs: "nextTurn"` — stored and injected on the next user prompt - `triggerTurn: true` — starts a turn when idle (also honored with `deliverAs: "nextTurn"`: idle prompts immediately; while streaming the queued message schedules an internal continuation) -`pi.sendUserMessage(content, { deliverAs })` always goes through prompt flow; while streaming it queues as steer/follow-up. +`pi.sendUserMessage(content, { deliverAs })` always goes through prompt flow. Omit `deliverAs` to start a normal prompt when idle; while streaming, omitted `deliverAs` queues the message as a steer. Set `deliverAs: "followUp"` to wait until the current run finishes. ## 2) Handler context (`ExtensionContext`) @@ -311,6 +311,7 @@ Supported: - dialogs: `select`, `confirm`, `input`, `editor` - input editing: `setEditorText`, `getEditorText`, `pasteToEditor`, `editor` +- autocomplete stacking: `addAutocompleteProvider(factory)` wraps the built-in editor provider (factories apply in registration order and re-apply on every slash-command refresh) - terminal title and working message (`setTitle`, `setWorkingMessage`) - notifications/status/editor text/terminal input/custom overlays - theme listing/loading by name (`setTheme` supports string names) @@ -334,7 +335,7 @@ Unsupported/no-op in RPC implementation: - `onTerminalInput` - `custom` -- `setFooter`, `setHeader`, `setEditorComponent` +- `setFooter`, `setHeader`, `setEditorComponent`, `addAutocompleteProvider` - `setWorkingMessage` - theme switching/loading (`setTheme` returns failure) - tool expansion controls are inert @@ -345,7 +346,7 @@ When no UI context is supplied to runner init, `ctx.hasUI` is `false` and method ### ACP mode -ACP installs an elicitation-bridged UI context (`createAcpExtensionUiContext` in `acp-agent.ts`). `ctx.hasUI` is `true` while only `select`/`confirm`/`input` round-trip (as ACP elicitations; defaults are returned when the client lacks the `elicitation.form` capability). The non-elicitation surface (widgets, editor, theming, terminal input) is stubbed no-op. +ACP installs an elicitation-bridged UI context (`createAcpExtensionUiContext` in `acp-agent.ts`). `ctx.hasUI` is `true` while only `select`/`confirm`/`input` round-trip (as ACP elicitations; defaults are returned when the client lacks the `elicitation.form` capability). The non-elicitation surface (widgets, editor, theming, terminal input, autocomplete stacking) is stubbed no-op. ## Session and state patterns diff --git a/docs/porting-from-pi-mono.md b/docs/porting-from-pi-mono.md index 427d47394..6234a940d 100644 --- a/docs/porting-from-pi-mono.md +++ b/docs/porting-from-pi-mono.md @@ -307,6 +307,7 @@ Our fork has architectural decisions that differ from upstream. **Do not port th | `FooterDataProvider` class | `StatusLineComponent` | Simpler, integrated status line | | `ctx.ui.setHeader()` / `ctx.ui.setFooter()` | No-op stubs in current extension contexts | Not currently wired to replace the TUI status/header UI | | `ctx.ui.setEditorComponent()` | Wired in interactive mode; no-op stubs in ACP/RPC/headless contexts | Custom editor replacement works in the interactive TUI; non-TUI runtimes keep stubs | +| `ctx.ui.addAutocompleteProvider()` | Wired in interactive mode; no-op stubs in ACP/RPC/headless contexts | Factory wrapping matches upstream; omp's editor has no custom `triggerCharacters`, so wrapped providers surface at the built-in trigger points | | `InteractiveModeOptions` options object | Positional constructor args (options type still exported) | Keep constructor signature; update the type when upstream adds fields | ### Component Naming diff --git a/docs/sdk.md b/docs/sdk.md index a0ab6b403..c06dec3a4 100644 --- a/docs/sdk.md +++ b/docs/sdk.md @@ -215,8 +215,9 @@ Behavior: 1. optional command/template expansion (`/` commands, custom commands, file slash commands, prompt templates) 2. if currently streaming: - - requires `streamingBehavior: "steer" | "followUp"` - - queues instead of throwing work away + - `streamingBehavior: "steer" | "followUp"` chooses how `prompt()` queues + - extension `sendUserMessage(content)` defaults to steer when `deliverAs` is omitted + - queued messages are preserved instead of throwing work away 3. if idle: - validates model + API key - appends user message diff --git a/package.json b/package.json index aaee74002..415f8d5a6 100644 --- a/package.json +++ b/package.json @@ -25,18 +25,18 @@ "@huggingface/transformers": "^4.2.0", "@mozilla/readability": "^0.6.0", "@napi-rs/cli": "3.7.0", - "@oh-my-pi/hashline": "16.3.12", - "@oh-my-pi/omp-stats": "16.3.12", - "@oh-my-pi/pi-agent-core": "16.3.12", - "@oh-my-pi/pi-ai": "16.3.12", - "@oh-my-pi/pi-catalog": "16.3.12", - "@oh-my-pi/pi-coding-agent": "16.3.12", - "@oh-my-pi/pi-mnemopi": "16.3.12", - "@oh-my-pi/pi-natives": "16.3.12", - "@oh-my-pi/pi-tui": "16.3.12", - "@oh-my-pi/pi-utils": "16.3.12", - "@oh-my-pi/pi-wire": "16.3.12", - "@oh-my-pi/snapcompact": "16.3.12", + "@oh-my-pi/hashline": "16.3.14", + "@oh-my-pi/omp-stats": "16.3.14", + "@oh-my-pi/pi-agent-core": "16.3.14", + "@oh-my-pi/pi-ai": "16.3.14", + "@oh-my-pi/pi-catalog": "16.3.14", + "@oh-my-pi/pi-coding-agent": "16.3.14", + "@oh-my-pi/pi-mnemopi": "16.3.14", + "@oh-my-pi/pi-natives": "16.3.14", + "@oh-my-pi/pi-tui": "16.3.14", + "@oh-my-pi/pi-utils": "16.3.14", + "@oh-my-pi/pi-wire": "16.3.14", + "@oh-my-pi/snapcompact": "16.3.14", "@opentelemetry/api": "^1.9.1", "@opentelemetry/context-async-hooks": "^2.7.1", "@opentelemetry/exporter-trace-otlp-proto": "^0.218.0", diff --git a/packages/agent/package.json b/packages/agent/package.json index 5d945c29a..fd21f9d17 100644 --- a/packages/agent/package.json +++ b/packages/agent/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/pi-agent-core", - "version": "16.3.12", + "version": "16.3.14", "description": "General-purpose agent with transport abstraction, state management, and attachment support", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md index 2be4cc39b..e6d81932f 100644 --- a/packages/ai/CHANGELOG.md +++ b/packages/ai/CHANGELOG.md @@ -2,6 +2,29 @@ ## [Unreleased] +## [16.3.14] - 2026-07-09 + +### Changed + +- Updated Codex reasoning effort mapping to support shifted wire tiers for newer models + +### Fixed + +- Fixed the Codex Responses request transformer bypassing catalog/compat reasoning effort maps: the clamped user effort is now remapped to the provider wire tier (GPT-5.6's shifted five-tier scale sends `max` for user `xhigh` and `xhigh` for `high`), failing loudly if a map produces a value outside the Codex wire vocabulary. + +## [16.3.13] - 2026-07-09 + +### Changed + +- Changed the xAI Grok OAuth (`xai-oauth`) provider to use manual code-paste login by default. `/login` now accepts a pasted authorization code or full `http://127.0.0.1:56121/callback?code=...` redirect URL without starting a local callback listener ([#3277](https://github.com/can1357/oh-my-pi/pull/3277) by [@Jaaneek](https://github.com/Jaaneek)). +- Renamed the xAI Grok OAuth provider in login and credential prompts to "xAI Grok OAuth (SuperGrok or X Premium+)" ([#3277](https://github.com/can1357/oh-my-pi/pull/3277) by [@Jaaneek](https://github.com/Jaaneek)). + +### Fixed + +- Fixed the generic lazy-stream idle watchdog aborting healthy `cursor-agent` streams with "Provider stream stalled while waiting for the next event" while a Cursor exec-channel local tool (shell/read/grep/write/MCP/…) legitimately ran longer than the idle budget. Provider streams now advertise consumer-side local work in flight and the watchdog slides its deadline instead of aborting; genuinely silent streams still time out. ([#4593](https://github.com/can1357/oh-my-pi/issues/4593)) +- Fixed OpenAI Codex/Responses reasoning streams so streamed thinking content is preserved when the final `output_item.done` reconstructs to an empty summary ([#4918](https://github.com/can1357/oh-my-pi/issues/4918)). +- Fixed Anthropic streams hanging forever when generation wedges mid-stream (notably long `write` tool calls on Opus 4.8 high/xhigh) while the server keeps sending `ping` keepalives: pings now extend the idle watchdog only within a bounded window (3x the idle timeout) since the last real stream event, so a stalled tool-call stream times out and recovers instead of hanging with no retry path ([#4900](https://github.com/can1357/oh-my-pi/issues/4900)). + ## [16.3.12] - 2026-07-08 ### Added diff --git a/packages/ai/package.json b/packages/ai/package.json index 814352b52..438c2f1f1 100644 --- a/packages/ai/package.json +++ b/packages/ai/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/pi-ai", - "version": "16.3.12", + "version": "16.3.14", "description": "Unified LLM API with automatic model discovery and provider configuration", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/ai/src/auth-broker/discover.ts b/packages/ai/src/auth-broker/discover.ts index c66ccdd20..71ef2c5fe 100644 --- a/packages/ai/src/auth-broker/discover.ts +++ b/packages/ai/src/auth-broker/discover.ts @@ -1,6 +1,6 @@ /** * Broker-aware auth-storage discovery used by both the coding-agent runtime and - * the catalog model generator. Keeps the precedence logic (env → config.yml → + * the catalog model generator. Keeps the precedence logic (env → config.yml/config.yaml → * token file → local SQLite) in one place so build-time tooling sees the same * credentials as the TUI. */ @@ -12,6 +12,7 @@ import { getConfigRootDir, isEnoent, logger, + MAIN_CONFIG_FILENAMES, } from "@oh-my-pi/pi-utils"; import { YAML } from "bun"; import { AuthStorage } from "../auth-storage"; @@ -72,21 +73,24 @@ interface ConfigSnapshot { } async function readConfigYaml(agentDir: string): Promise { - const configPath = path.join(agentDir, "config.yml"); - try { - const raw = await Bun.file(configPath).text(); - const parsed = YAML.parse(raw); - if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) return {}; - const record = parsed as Record; - const url = typeof record["auth.broker.url"] === "string" ? (record["auth.broker.url"] as string) : undefined; - const token = - typeof record["auth.broker.token"] === "string" ? (record["auth.broker.token"] as string) : undefined; - return { url, token }; - } catch (err) { - if (isEnoent(err)) return {}; - logger.warn("auth-broker config.yml unreadable", { error: String(err) }); - return {}; + for (const filename of MAIN_CONFIG_FILENAMES) { + const configPath = path.join(agentDir, filename); + try { + const raw = await Bun.file(configPath).text(); + const parsed = YAML.parse(raw); + if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) return {}; + const record = parsed as Record; + const url = typeof record["auth.broker.url"] === "string" ? (record["auth.broker.url"] as string) : undefined; + const token = + typeof record["auth.broker.token"] === "string" ? (record["auth.broker.token"] as string) : undefined; + return { url, token }; + } catch (err) { + if (isEnoent(err)) continue; + logger.warn("auth-broker config unreadable", { path: configPath, error: String(err) }); + return {}; + } } + return {}; } function resolveSnapshotTtlMs(): number { @@ -104,7 +108,7 @@ function resolveSnapshotTtlMs(): number { * Resolve broker connection configuration using the same precedence as the TUI: * * 1. `OMP_AUTH_BROKER_URL` / `OMP_AUTH_BROKER_TOKEN` env vars. - * 2. `auth.broker.url` / `auth.broker.token` in `/config.yml`. + * 2. `auth.broker.url` / `auth.broker.token` in `/config.yml` or `/config.yaml`. * 3. `/auth-broker.token` file (paired with a URL from env/config). * * Returns `null` when no broker URL is configured — callers should fall back to diff --git a/packages/ai/src/providers/anthropic.ts b/packages/ai/src/providers/anthropic.ts index 6b3570c23..a22302fd2 100644 --- a/packages/ai/src/providers/anthropic.ts +++ b/packages/ai/src/providers/anthropic.ts @@ -1462,6 +1462,16 @@ async function* observeDecodedAnthropicSdkEvents( const PROVIDER_MAX_RETRIES = 10; +/** + * How long `ping` keepalives may keep extending the idle deadline without any + * semantic stream progress, as a multiple of the idle timeout. Anthropic pings + * across legitimate generation gaps, so pings count as liveness — but a wedged + * upstream that pings forever while producing no events must eventually trip + * the idle watchdog instead of hanging an active tool-call stream without a + * recovery path (#4900). + */ +const PING_PROGRESS_MAX_IDLE_MULTIPLIER = 3; + /** * Log a malformed-stream-envelope anomaly without aborting the turn. The strict * parser would `throw new AnthropicStreamEnvelopeError(...)` here; we instead @@ -2007,11 +2017,20 @@ const streamAnthropicOnce = ( } >(); - // Pings keep the idle deadline alive once content is flowing, but a - // ping before message_start must not consume the first-event watchdog: - // it would flip the (retryable) pre-content stall classification into - // a terminal mid-stream idle timeout. + // Pings keep the idle deadline alive once content is flowing (Anthropic + // bridges legitimate generation gaps with keepalives), but only within a + // bounded window: a wedged upstream that pings forever while the model + // produces nothing must still trip the idle watchdog, otherwise an + // active tool-call stream hangs unrecoverably with no retry (#4900). + // A ping before message_start must not consume the first-event watchdog + // either: it would flip the (retryable) pre-content stall classification + // into a terminal mid-stream idle timeout. let sawNonPingEvent = false; + let lastNonPingProgressAtMs = 0; + const pingProgressCapMs = + idleTimeoutMs !== undefined && idleTimeoutMs > 0 + ? idleTimeoutMs * PING_PROGRESS_MAX_IDLE_MULTIPLIER + : undefined; const timedAnthropicStream = iterateWithIdleTimeout(anthropicStream, { idleTimeoutMs, firstItemTimeoutMs: firstEventTimeoutMs, @@ -2021,8 +2040,13 @@ const streamAnthropicOnce = ( onFirstItemTimeout: () => activeAbortTracker.abortLocally(firstEventTimeoutAbortError), abortSignal: options?.signal, isProgressItem: item => { - if ((item as AnthropicStreamEvent).type === "ping") return sawNonPingEvent; + if ((item as AnthropicStreamEvent).type === "ping") { + if (!sawNonPingEvent) return false; + if (pingProgressCapMs === undefined) return true; + return Date.now() - lastNonPingProgressAtMs < pingProgressCapMs; + } sawNonPingEvent = true; + lastNonPingProgressAtMs = Date.now(); return true; }, }); diff --git a/packages/ai/src/providers/cursor.ts b/packages/ai/src/providers/cursor.ts index b7cb3e3de..d2034de31 100644 --- a/packages/ai/src/providers/cursor.ts +++ b/packages/ai/src/providers/cursor.ts @@ -653,7 +653,8 @@ export interface UsageState { sawTokenDelta: boolean; } -async function handleServerMessage( +/** Exported for tests: drives one Cursor server message through the stream (exec waits mark the stream busy). */ +export async function handleServerMessage( msg: AgentServerMessage, output: AssistantMessage, stream: AssistantMessageEventStream, @@ -675,15 +676,21 @@ async function handleServerMessage( } else if (msgCase === "kvServerMessage") { handleKvServerMessage(msg.message.value as KvServerMessage, blobStore, h2Request); } else if (msgCase === "execServerMessage") { - await handleExecServerMessage( - msg.message.value as ExecServerMessage, - h2Request, - execHandlers, - onToolResult, - requestContextTools, - output, - stream, - state, + // The server is waiting on OUR local tool result during this window — no + // AssistantMessageEvent flows until the handler finishes. Mark the wait + // as local work so the lazy stream idle watchdog attributes the silence + // to the tool run instead of aborting a healthy stream (issue #4593). + await stream.trackLocalWork( + handleExecServerMessage( + msg.message.value as ExecServerMessage, + h2Request, + execHandlers, + onToolResult, + requestContextTools, + output, + stream, + state, + ), ); } else if (msgCase === "conversationCheckpointUpdate") { handleConversationCheckpointUpdate(msg.message.value, output, usageState, onConversationCheckpoint); diff --git a/packages/ai/src/providers/openai-codex-responses.ts b/packages/ai/src/providers/openai-codex-responses.ts index 80baabe8d..0bde541d0 100644 --- a/packages/ai/src/providers/openai-codex-responses.ts +++ b/packages/ai/src/providers/openai-codex-responses.ts @@ -95,6 +95,7 @@ import { encodeTextSignatureV1, finalizeCustomToolCallInputDone, finalizePendingResponsesToolCalls, + finalizeReasoningThinking, finalizeToolCallArgumentsDone, isOpenAIResponsesProgressEvent, mapOpenAIResponsesStopReason, @@ -1407,6 +1408,21 @@ class CodexStreamProcessor { return firstTokenTime; } + if (eventType === "response.reasoning_text.delta") { + const entry = this.runtime.openItemForEvent(rawEvent); + const delta = typeof rawEvent.delta === "string" ? rawEvent.delta : ""; + if (entry?.item.type === "reasoning" && entry.block?.type === "thinking") { + entry.block.thinking += delta; + stream.push({ + type: "thinking_delta", + contentIndex: entry.contentIndex, + delta, + partial: output, + }); + } + return firstTokenTime; + } + if (eventType === "response.reasoning_summary_part.done") { if (this.runtime.currentItem?.type === "reasoning" && this.runtime.currentBlock?.type === "thinking") { appendReasoningSummaryPartDone( @@ -1522,13 +1538,13 @@ class CodexStreamProcessor { // most-recently-added block may belong to a sibling (#2619). Some Codex // function/custom tool items omit `id`; in that case `output_index` still // routes `output_item.done` to the block that received `output_item.added`. - const itemId = typeof (item as { id?: string }).id === "string" ? (item as { id: string }).id : ""; + const itemId = "id" in item && typeof item.id === "string" ? item.id : ""; const entry = (itemId ? runtime.openItems.get(itemId) : null) ?? runtime.openItemForEvent(rawEvent); const block = entry?.block ?? null; const contentIndex = entry?.contentIndex ?? output.content.length - 1; if (item.type === "reasoning" && block?.type === "thinking") { - block.thinking = item.summary?.map(summary => summary.text).join("\n\n") || ""; + block.thinking = finalizeReasoningThinking(item, block.thinking); block.thinkingSignature = JSON.stringify(item); stream.push({ type: "thinking_end", diff --git a/packages/ai/src/providers/openai-codex/request-transformer.ts b/packages/ai/src/providers/openai-codex/request-transformer.ts index d603130c3..67af8f232 100644 --- a/packages/ai/src/providers/openai-codex/request-transformer.ts +++ b/packages/ai/src/providers/openai-codex/request-transformer.ts @@ -1,19 +1,33 @@ -import type { Effort } from "@oh-my-pi/pi-catalog/effort"; +import { Effort } from "@oh-my-pi/pi-catalog/effort"; import { supportsAllTurnsReasoningContext, supportsCodexReasoningSummary } from "@oh-my-pi/pi-catalog/identity"; import { requireSupportedEffort } from "@oh-my-pi/pi-catalog/model-thinking"; -import type { Api, Model } from "../../types"; +import type { Model } from "../../types"; +import { mapOpenAIReasoningEffort } from "../openai-shared"; /** Reasoning replay scope for the Codex Responses API (`reasoning.context`). */ export type CodexReasoningContext = "auto" | "current_turn" | "all_turns"; +/** User-facing effort levels accepted by Codex request options. */ +type CodexCallerEffort = "minimal" | "low" | "medium" | "high" | "xhigh"; + +/** Caller literal → catalog `Effort` bridge (the enum is nominal). */ +const EFFORT_BY_NAME: Record = { + minimal: Effort.Minimal, + low: Effort.Low, + medium: Effort.Medium, + high: Effort.High, + xhigh: Effort.XHigh, +}; + export interface ReasoningConfig { - effort: "none" | "minimal" | "low" | "medium" | "high" | "xhigh"; + effort: "none" | "minimal" | "low" | "medium" | "high" | "xhigh" | "max"; summary?: "auto" | "concise" | "detailed"; context?: CodexReasoningContext; } export interface CodexRequestOptions { - reasoningEffort?: ReasoningConfig["effort"]; + /** User-facing effort; the wire-only `max` tier is reached via the model's effort map. */ + reasoningEffort?: CodexCallerEffort | "none"; reasoningSummary?: ReasoningConfig["summary"] | null; /** Explicit `reasoning.context` override; defaults to `all_turns` when unset. The `all_turns` value is gated to gpt-5.4+ Codex models — older ids reject it, so it is suppressed and `context` omitted. */ reasoningContext?: CodexReasoningContext; @@ -80,10 +94,40 @@ export function shouldUseCodexResponsesLite(body: RequestBody, requested: boolea return requested === true && !containsInputImage(body.input); } -function getReasoningConfig(model: Model, options: CodexRequestOptions): ReasoningConfig { +/** + * Clamp a user-facing effort to the model's ladder, then remap to the wire + * tier (e.g. GPT-5.6's shifted five-tier scale sends `max` for user `xhigh`). + * A mapped value outside the Codex wire vocabulary is a broken compat/model + * effort map — fail loudly rather than silently sending a different tier. + */ +function mapCodexWireEffort( + model: Model<"openai-codex-responses">, + effort: CodexCallerEffort, +): ReasoningConfig["effort"] { + const mapped = mapOpenAIReasoningEffort(model, model.compat, requireSupportedEffort(model, EFFORT_BY_NAME[effort])); + switch (mapped) { + case "none": + case "minimal": + case "low": + case "medium": + case "high": + case "xhigh": + case "max": + return mapped; + default: + throw new Error( + `Effort map for ${model.provider}/${model.id} produced invalid Codex reasoning effort "${mapped}"`, + ); + } +} + +function getReasoningConfig( + model: Model<"openai-codex-responses">, + effort: NonNullable, + options: CodexRequestOptions, +): ReasoningConfig { const config: ReasoningConfig = { - effort: - options.reasoningEffort === "none" ? "none" : requireSupportedEffort(model, options.reasoningEffort as Effort), + effort: effort === "none" ? "none" : mapCodexWireEffort(model, effort), }; // `reasoning.summary` is accepted only from gpt-5.4 onward; earlier Codex ids // (gpt-5.1-codex, gpt-5.3-codex, gpt-5.3-codex-spark) reject it with @@ -216,7 +260,7 @@ function stripImageDetails(input: InputItem[]): void { export async function transformRequestBody( body: RequestBody, - model: Model, + model: Model<"openai-codex-responses">, options: CodexRequestOptions = {}, prompt?: { developerMessages: string[] }, ): Promise { @@ -300,7 +344,7 @@ export async function transformRequestBody( } if (options.reasoningEffort !== undefined) { - const reasoningConfig = getReasoningConfig(model, options); + const reasoningConfig = getReasoningConfig(model, options.reasoningEffort, options); body.reasoning = { ...body.reasoning, ...reasoningConfig, diff --git a/packages/ai/src/providers/openai-shared.ts b/packages/ai/src/providers/openai-shared.ts index 1d40b0ac5..bacf6d8af 100644 --- a/packages/ai/src/providers/openai-shared.ts +++ b/packages/ai/src/providers/openai-shared.ts @@ -695,13 +695,19 @@ export interface OpenAICompatPolicy { }; } -function mapOpenAIReasoningEffort( +/** + * Map a user-facing effort to the provider wire value: explicit compat + * override first, then the model's baked `thinking.effortMap`, else identity. + * Shared by the chat-completions/Responses policy resolver and the Codex + * request transformer. + */ +export function mapOpenAIReasoningEffort( model: Pick, - compat: OpenAICompatPolicyCompat, + compat: { reasoningEffortMap?: Partial> } | undefined, effort: string, ): string { const level = effort as Effort; - return compat.reasoningEffortMap?.[level] ?? model.thinking?.effortMap?.[level] ?? effort; + return compat?.reasoningEffortMap?.[level] ?? model.thinking?.effortMap?.[level] ?? effort; } function isImplicitDisableWhenNotRequested(disableMode: OpenAIReasoningDisableMode): boolean { @@ -1684,6 +1690,14 @@ export function appendReasoningSummaryPart( item.summary.push(part); } +/** Chooses the final reasoning text without discarding content already streamed into the block. */ +export function finalizeReasoningThinking(item: ResponseReasoningItem, streamedThinking: string): string { + const summaryThinking = item.summary?.map(part => part.text).join("\n\n") ?? ""; + if (summaryThinking) return summaryThinking; + const contentThinking = item.content?.[0]?.type === "reasoning_text" ? (item.content[0].text ?? "") : ""; + return contentThinking || streamedThinking || ""; +} + export function appendReasoningSummaryTextDelta( item: ResponseReasoningItem, block: ThinkingContent, @@ -2208,12 +2222,6 @@ export async function processResponsesStream( ? lookupOpenItem({ output_index: event.output_index, item_id: item.id ?? item.call_id }) : lookupOpenItem({ output_index: event.output_index, item_id: item.id }); if (item.type === "reasoning") { - const thinking = - item.summary?.length > 0 - ? item.summary.map(part => part.text).join("\n\n") - : item.content?.[0]?.type === "reasoning_text" - ? (item.content[0].text ?? "") - : ""; // Prefer the routed entry; the bare itemId find misroutes when ids are // absent (`undefined === undefined` matches the FIRST thinking block) and // misses entirely when the done-event id drifts from the added-event id. @@ -2224,12 +2232,12 @@ export async function processResponsesStream( | ThinkingContent | undefined); if (reasoningBlock) { - reasoningBlock.thinking = thinking; + reasoningBlock.thinking = finalizeReasoningThinking(item, reasoningBlock.thinking); reasoningBlock.thinkingSignature = JSON.stringify(item); stream.push({ type: "thinking_end", contentIndex: contentIndexOf(reasoningBlock), - content: thinking, + content: reasoningBlock.thinking, partial: output, }); } diff --git a/packages/ai/src/providers/register-builtins.ts b/packages/ai/src/providers/register-builtins.ts index efd6b8dd6..8bb3a20f0 100644 --- a/packages/ai/src/providers/register-builtins.ts +++ b/packages/ai/src/providers/register-builtins.ts @@ -157,6 +157,7 @@ let openAICompletionsProviderModulePromise: Promise> | undefined; let ollamaProviderModulePromise: Promise> | undefined; let cursorProviderModulePromise: Promise> | undefined; +let cursorProviderModuleOverride: LazyProviderModule<"cursor-agent"> | undefined; let devinProviderModulePromise: Promise> | undefined; let bedrockProviderModuleOverride: LazyProviderModule<"bedrock-converse-stream"> | undefined; let bedrockProviderModulePromise: Promise> | undefined; @@ -167,6 +168,12 @@ export function setBedrockProviderModule(module: BedrockProviderModule): void { }; } +export function setCursorProviderModule(module: CursorProviderModule): void { + cursorProviderModuleOverride = { + stream: module.streamCursor, + }; +} + // --------------------------------------------------------------------------- // Stream forwarding / error helpers // --------------------------------------------------------------------------- @@ -245,6 +252,10 @@ function forwardStream( (limits?.openAIIdleEnvFloorsFirstEvent ? getOpenAIStreamFirstEventTimeoutMs(idleTimeoutMs, limits.defaultFirstEventTimeoutMs) : getStreamFirstEventTimeoutMs(idleTimeoutMs, limits?.defaultFirstEventTimeoutMs))); + // Providers with a server-driven local tool bridge (e.g. the Cursor + // exec channel) mark their stream busy while a local tool runs; the + // watchdog must not read that silence as a provider stall (#4593). + const localWorkSource = source instanceof EventStreamImpl ? source : undefined; const watchedSource = iterateWithIdleTimeout(source, { idleTimeoutMs, firstItemTimeoutMs, @@ -260,6 +271,7 @@ function forwardStream( // `idleTimeoutMs` while we're still legitimately waiting on the model's // first response (slow first-token from reasoning models, cold proxies, etc.). isProgressItem: event => (event as AssistantMessageEvent).type !== "start", + hasPendingLocalWork: localWorkSource ? () => localWorkSource.hasPendingLocalWork : undefined, }); for await (const event of watchedSource) { @@ -411,6 +423,9 @@ function loadOllamaProviderModule(): Promise> } function loadCursorProviderModule(): Promise> { + if (cursorProviderModuleOverride) { + return Promise.resolve(cursorProviderModuleOverride); + } cursorProviderModulePromise ||= import("./cursor").then(module => { const provider = module as CursorProviderModule; return { stream: provider.streamCursor }; diff --git a/packages/ai/src/registry/oauth/__tests__/xai-oauth.test.ts b/packages/ai/src/registry/oauth/__tests__/xai-oauth.test.ts index 190e06aa7..e80a6a544 100644 --- a/packages/ai/src/registry/oauth/__tests__/xai-oauth.test.ts +++ b/packages/ai/src/registry/oauth/__tests__/xai-oauth.test.ts @@ -69,6 +69,61 @@ describe("XAIOAuthFlow", () => { expect(flow.redirectUri).toBe("http://127.0.0.1:56121/callback"); }); + + it("uses pasted-code login without starting a callback server", async () => { + const serveSpy = vi.spyOn(Bun, "serve").mockImplementation(() => { + throw new Error("callback server should not start"); + }); + let authUrl = ""; + let tokenRequestBody = ""; + const progress: string[] = []; + const fetchMock = vi.fn(async (input: string | URL | Request, init?: RequestInit) => { + const url = typeof input === "string" ? input : input instanceof Request ? input.url : input.toString(); + if (url.includes("/.well-known/openid-configuration")) { + return new Response( + JSON.stringify({ + authorization_endpoint: "https://auth.x.ai/oauth/authorize", + token_endpoint: "https://auth.x.ai/oauth/token", + }), + { status: 200, headers: { "Content-Type": "application/json" } }, + ); + } + tokenRequestBody = init?.body instanceof URLSearchParams ? init.body.toString() : String(init?.body ?? ""); + return new Response( + JSON.stringify({ + access_token: "access-token", + refresh_token: "refresh-token", + expires_in: 3600, + }), + { status: 200, headers: { "Content-Type": "application/json" } }, + ); + }); + + const flow = new XAIOAuthFlow({ + fetch: fetchMock as unknown as typeof fetch, + onAuth: info => { + authUrl = info.url; + }, + onManualCodeInput: async () => { + const parsed = new URL(authUrl); + const redirectUri = parsed.searchParams.get("redirect_uri") ?? ""; + const state = parsed.searchParams.get("state") ?? ""; + return `${redirectUri}?code=code-xyz&state=${encodeURIComponent(state)}`; + }, + onProgress: message => progress.push(message), + }); + + const credentials = await flow.login(); + const authorizeUrl = new URL(authUrl); + const tokenParams = new URLSearchParams(tokenRequestBody); + + expect(serveSpy).not.toHaveBeenCalled(); + expect(authorizeUrl.searchParams.get("redirect_uri")).toBe("http://127.0.0.1:56121/callback"); + expect(progress).toContain("Waiting for pasted authorization code..."); + expect(tokenParams.get("code")).toBe("code-xyz"); + expect(credentials.access).toBe("access-token"); + expect(credentials.refresh).toBe("refresh-token"); + }); }); describe("XAIOAuthFlow.exchangeToken", () => { diff --git a/packages/ai/src/registry/oauth/callback-server.ts b/packages/ai/src/registry/oauth/callback-server.ts index c3a45dc97..4342d2472 100644 --- a/packages/ai/src/registry/oauth/callback-server.ts +++ b/packages/ai/src/registry/oauth/callback-server.ts @@ -48,6 +48,8 @@ export interface OAuthCallbackFlowOptions { * an actionable message before opening the browser. */ allowPortFallback?: boolean; + /** Skip the local callback server entirely; the user pastes the code or redirect URL back. */ + manualInputOnly?: boolean; } /** @@ -60,6 +62,7 @@ export abstract class OAuthCallbackFlow { callbackHostname: string; redirectUri?: string; allowPortFallback: boolean; + #manualInputOnly: boolean; #callbackResolve?: (result: CallbackResult) => void; #callbackReject?: (error: string) => void; /** @@ -82,6 +85,7 @@ export abstract class OAuthCallbackFlow { this.callbackPath = callbackPath; this.callbackHostname = DEFAULT_HOSTNAME; this.allowPortFallback = true; + this.#manualInputOnly = false; return; } @@ -90,6 +94,7 @@ export abstract class OAuthCallbackFlow { this.callbackHostname = preferredPortOrOptions.callbackHostname ?? DEFAULT_HOSTNAME; this.redirectUri = preferredPortOrOptions.redirectUri; this.allowPortFallback = preferredPortOrOptions.allowPortFallback ?? true; + this.#manualInputOnly = preferredPortOrOptions.manualInputOnly ?? false; } /** @@ -135,8 +140,12 @@ export abstract class OAuthCallbackFlow { const state = this.generateState(); this.#throwIfCancelled(); - // Start callback server first to get actual redirect URI - const { server, redirectUri, launchUrl } = await this.#startCallbackServer(state); + // Start callback server first to get actual redirect URI. Manual-only + // flows never bind a server — the advertised redirect URI is fixed and + // the user pastes the code/redirect URL back instead. + const { server, redirectUri, launchUrl } = this.#manualInputOnly + ? { server: undefined, redirectUri: this.#buildRedirectUri(), launchUrl: undefined } + : await this.#startCallbackServer(state); try { this.#throwIfCancelled(); @@ -152,9 +161,12 @@ export abstract class OAuthCallbackFlow { // Notify controller that auth is ready this.ctrl.onAuth?.({ url: authUrl, launchUrl, instructions }); - this.ctrl.onProgress?.("Waiting for browser authentication..."); + this.ctrl.onProgress?.( + this.#manualInputOnly + ? "Waiting for pasted authorization code..." + : "Waiting for browser authentication...", + ); - // Wait for callback or manual input const { code } = await this.#waitForCallback(state); this.#throwIfCancelled(); @@ -163,10 +175,14 @@ export abstract class OAuthCallbackFlow { return await this.exchangeToken(code, state, redirectUri); } finally { this.#pendingAuthUrl = undefined; - server.stop(); + server?.stop(); } } + #buildRedirectUri(): string { + return this.redirectUri ?? `http://${this.callbackHostname}:${this.preferredPort}${this.callbackPath}`; + } + /** * Start callback server, trying preferred port first, falling back to random. * `launchUrl` is `undefined` when the caller configured `callbackPath` to diff --git a/packages/ai/src/registry/oauth/xai-oauth.ts b/packages/ai/src/registry/oauth/xai-oauth.ts index 9830847db..8c55de347 100644 --- a/packages/ai/src/registry/oauth/xai-oauth.ts +++ b/packages/ai/src/registry/oauth/xai-oauth.ts @@ -1,9 +1,10 @@ // Ported from NousResearch/hermes-agent (MIT) — hermes_cli/auth.py xAI sections (L93-111, L2979-3160, L5286-5469). /** - * xAI Grok (SuperGrok Subscription) OAuth flow. + * xAI Grok (SuperGrok or X Premium+) OAuth flow. * - * Loopback PKCE flow on `127.0.0.1:56121/callback`. One token unlocks Grok-4.x + * Manual-code PKCE flow using `127.0.0.1:56121/callback` as the allowlisted + * redirect URI. One token unlocks Grok-4.x * chat, Grok Imagine image generation, and Grok Voice TTS via subsequent * commits. Endpoint discovery is hardened against MITM via * {@link validateXAIEndpoint}: any non-HTTPS or non-`x.ai`/`*.x.ai` host is @@ -196,10 +197,7 @@ function buildXAIAuthorizeUrl(opts: BuildXAIAuthorizeUrlOptions): string { } /** - * xAI Grok OAuth loopback flow (Hermes `_xai_oauth_loopback_login` L5315-5469). - * - * Uses a fixed redirect URI so the callback server fails fast instead of - * falling back to a random port that xAI's redirect_uri allowlist rejects. + * xAI Grok OAuth code flow (Hermes `_xai_oauth_loopback_login` L5315-5469). */ export class XAIOAuthFlow extends OAuthCallbackFlow { #verifier: string = ""; @@ -211,6 +209,7 @@ export class XAIOAuthFlow extends OAuthCallbackFlow { callbackPath: XAI_OAUTH_REDIRECT_PATH, callbackHostname: XAI_OAUTH_REDIRECT_HOST, redirectUri: `http://${XAI_OAUTH_REDIRECT_HOST}:${XAI_OAUTH_REDIRECT_PORT}${XAI_OAUTH_REDIRECT_PATH}`, + manualInputOnly: true, } satisfies OAuthCallbackFlowOptions); this.#fetch = ctrl.fetch ?? fetch; } @@ -231,7 +230,7 @@ export class XAIOAuthFlow extends OAuthCallbackFlow { return { url, - instructions: `Complete login in your browser for xAI Grok (SuperGrok). Docs: ${XAI_OAUTH_DOCS_URL}`, + instructions: `Complete login in your browser for xAI Grok (SuperGrok or X Premium+). Docs: ${XAI_OAUTH_DOCS_URL}`, }; } @@ -308,9 +307,6 @@ export class XAIOAuthFlow extends OAuthCallbackFlow { } } -/** - * Login with xAI Grok OAuth (SuperGrok Subscription). - */ export async function loginXAIOAuth(ctrl: OAuthController): Promise { return new XAIOAuthFlow(ctrl).login(); } diff --git a/packages/ai/src/registry/xai-oauth.ts b/packages/ai/src/registry/xai-oauth.ts index bec1212d7..ed1a22bd4 100644 --- a/packages/ai/src/registry/xai-oauth.ts +++ b/packages/ai/src/registry/xai-oauth.ts @@ -3,7 +3,7 @@ import type { ProviderDefinition } from "./types"; export const xaiOauthProvider = { id: "xai-oauth", - name: "xAI Grok OAuth (SuperGrok Subscription)", + name: "xAI Grok OAuth (SuperGrok or X Premium+)", login: async (cb: OAuthLoginCallbacks) => { // Lazy import: keep heavy OAuth flow modules out of the eager registry graph. const { loginXAIOAuth } = await import("./oauth/xai-oauth"); @@ -14,4 +14,5 @@ export const xaiOauthProvider = { const { refreshXAIOAuthToken } = await import("./oauth/xai-oauth"); return refreshXAIOAuthToken(credentials.refresh); }, + pasteCodeFlow: true, } as const satisfies ProviderDefinition; diff --git a/packages/ai/src/utils/event-stream.ts b/packages/ai/src/utils/event-stream.ts index d7b75e863..6d1255e39 100644 --- a/packages/ai/src/utils/event-stream.ts +++ b/packages/ai/src/utils/event-stream.ts @@ -10,6 +10,14 @@ export class EventStream implements AsyncIterable { resultSettled = false; #failed = false; #error: unknown = undefined; + /** + * Consumer-side local operations currently in flight for this stream — a + * provider transport waiting on a server-requested local tool bridge + * (e.g. the Cursor exec channel) before it can send the result upstream. + * While non-zero, event silence is attributable to our own pending work, + * not a provider stall; idle watchdogs consult {@link hasPendingLocalWork}. + */ + #pendingLocalWork = 0; finalResultPromise: Promise; resolveFinalResult!: (result: R) => void; rejectFinalResult!: (err: unknown) => void; @@ -116,6 +124,24 @@ export class EventStream implements AsyncIterable { result(): Promise { return this.finalResultPromise; } + + /** True while local work tracked via {@link trackLocalWork} is pending. */ + get hasPendingLocalWork(): boolean { + return this.#pendingLocalWork > 0; + } + + /** + * Track a local-work promise so idle watchdogs on this stream do not treat + * the event silence while it is pending as a provider stall. + */ + async trackLocalWork(work: Promise): Promise { + this.#pendingLocalWork++; + try { + return await work; + } finally { + this.#pendingLocalWork--; + } + } } export class AssistantMessageEventStream extends EventStream { diff --git a/packages/ai/src/utils/idle-iterator.ts b/packages/ai/src/utils/idle-iterator.ts index 7cebaf64e..3accaf3c2 100644 --- a/packages/ai/src/utils/idle-iterator.ts +++ b/packages/ai/src/utils/idle-iterator.ts @@ -135,6 +135,16 @@ export interface IdleTimeoutIteratorOptions { * keepalive/no-op events from keeping a stalled tool call alive forever. */ isProgressItem?: (item: unknown) => boolean; + /** + * Reports consumer-side local work in flight for the stream: the provider + * transport is waiting on a server-requested local tool bridge (e.g. the + * Cursor exec channel) before anything can flow upstream again. While it + * returns true, an expired idle / first-item deadline slides forward + * instead of aborting — the silence is ours, not a provider stall. The + * watchdog re-arms with a full budget once the local work completes, so a + * provider that stalls afterwards is still caught. + */ + hasPendingLocalWork?: () => boolean; /** * Cancel iteration as soon as this signal aborts. Required for caller-driven * cancellation (ESC) when the underlying transport does not surface signal @@ -157,7 +167,7 @@ export async function* iterateWithIdleTimeout( options: IdleTimeoutIteratorOptions, ): AsyncGenerator { const firstItemTimeoutMs = options.firstItemTimeoutMs ?? options.idleTimeoutMs; - const firstItemDeadlineMs = + let firstItemDeadlineMs = firstItemTimeoutMs !== undefined && firstItemTimeoutMs > 0 ? Date.now() + firstItemTimeoutMs : undefined; const abortSignal = options.abortSignal; const iterator = iterable[Symbol.asyncIterator](); @@ -197,6 +207,28 @@ export async function* iterateWithIdleTimeout( }; let lastProgressAt = Date.now(); + const hasPendingLocalWork = (): boolean => { + if (!options.hasPendingLocalWork) return false; + try { + return options.hasPendingLocalWork(); + } catch { + return false; + } + }; + // Local work means the current gap is attributable to the consumer side, + // not the provider: slide the active deadline a full budget past now + // instead of aborting. Once the work completes the watchdog resumes from + // the last extension, so a provider that stalls afterwards is still caught. + const extendDeadlineForLocalWork = (): void => { + if (awaitingFirstItem) { + if (firstItemDeadlineMs !== undefined && firstItemTimeoutMs !== undefined) { + firstItemDeadlineMs = Date.now() + firstItemTimeoutMs; + } + } else { + lastProgressAt = Date.now(); + } + }; + const noTimeoutEnforced = (firstItemTimeoutMs === undefined || firstItemTimeoutMs <= 0) && (options.idleTimeoutMs === undefined || options.idleTimeoutMs <= 0); @@ -271,6 +303,12 @@ export async function* iterateWithIdleTimeout( timer = setTimeout(onTimerFire, Math.max(0, deadlineMs - Date.now())); }; + // The in-flight iterator.next() promise, persisted across loop iterations: + // a deadline extension for pending local work loops without consuming it, + // and issuing a second next() while one is outstanding would drop an item. + let pendingNext: + | Promise<{ kind: "next"; result: IteratorResult } | { kind: "error"; error: unknown }> + | undefined; try { let raceCount = 0; while (true) { @@ -291,21 +329,29 @@ export async function* iterateWithIdleTimeout( if (firstItemDeadlineMs !== undefined) { activeTimeoutMs = firstItemDeadlineMs - Date.now(); if (activeTimeoutMs <= 0) { - options.onFirstItemTimeout?.(); - closeIterator(); - throw new AIError.StreamTimeoutError(options.firstItemErrorMessage ?? options.errorMessage); + if (!hasPendingLocalWork()) { + options.onFirstItemTimeout?.(); + closeIterator(); + throw new AIError.StreamTimeoutError(options.firstItemErrorMessage ?? options.errorMessage); + } + extendDeadlineForLocalWork(); + activeTimeoutMs = firstItemDeadlineMs! - Date.now(); } } } else if (options.idleTimeoutMs !== undefined && options.idleTimeoutMs > 0) { activeTimeoutMs = options.idleTimeoutMs - (Date.now() - lastProgressAt); if (activeTimeoutMs <= 0) { - options.onIdle?.(); - closeIterator(); - throw new AIError.StreamTimeoutError(options.errorMessage); + if (!hasPendingLocalWork()) { + options.onIdle?.(); + closeIterator(); + throw new AIError.StreamTimeoutError(options.errorMessage); + } + extendDeadlineForLocalWork(); + activeTimeoutMs = options.idleTimeoutMs; } } - const nextResultPromise = withRacy(iterator.next()); + pendingNext ??= withRacy(iterator.next()); const racers: Array< Promise< @@ -314,7 +360,7 @@ export async function* iterateWithIdleTimeout( | { kind: "timeout" } | { kind: "abort" } > - > = [nextResultPromise]; + > = [pendingNext]; const enforceTimeout = !noTimeoutEnforced && activeTimeoutMs !== undefined && activeTimeoutMs > 0; if (enforceTimeout) { @@ -333,11 +379,21 @@ export async function* iterateWithIdleTimeout( let continuing = false; try { const outcome = await Promise.race(racers); + if (outcome.kind === "next" || outcome.kind === "error") { + pendingNext = undefined; + } if (outcome.kind === "abort") { closeIterator(); throw abortReason(abortSignal!); } if (outcome.kind === "timeout") { + if (hasPendingLocalWork()) { + // A local tool is still running; the provider cannot make + // progress until we hand its result back. Keep waiting. + extendDeadlineForLocalWork(); + continuing = true; + continue; + } if (!awaitingFirstItem) { options.onIdle?.(); } else { diff --git a/packages/ai/test/anthropic-ping-keepalive.test.ts b/packages/ai/test/anthropic-ping-keepalive.test.ts new file mode 100644 index 000000000..a7e309b9e --- /dev/null +++ b/packages/ai/test/anthropic-ping-keepalive.test.ts @@ -0,0 +1,233 @@ +import { afterEach, describe, expect, it, vi } from "bun:test"; +import { buildModel } from "@oh-my-pi/pi-catalog/build"; +import { streamAnthropic } from "../src/providers/anthropic"; +import type { AnthropicMessagesClientLike } from "../src/providers/anthropic-client"; +import type { Context, Model } from "../src/types"; +import { waitForDelayOrAbort } from "./helpers"; + +const model: Model<"anthropic-messages"> = buildModel({ + id: "claude-opus-4-8", + name: "Claude Opus 4.8", + api: "anthropic-messages", + provider: "anthropic", + baseUrl: "https://api.anthropic.com", + reasoning: true, + input: ["text"], + cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, + contextWindow: 200_000, + maxTokens: 8_192, +}); + +const context: Context = { + messages: [{ role: "user", content: "write a file", timestamp: Date.now() }], +}; + +type MockAnthropicEvent = Record; + +/** `{ waitMs, event }` script step; `waitMs` elapses (fake clock) before the event is yielded. */ +type ScriptStep = { waitMs: number; event: MockAnthropicEvent | "hang-with-pings" }; + +const writeToolCallOpening: MockAnthropicEvent[] = [ + { + type: "message_start", + message: { + id: "msg_ping_keepalive", + usage: { input_tokens: 10, output_tokens: 0, cache_read_input_tokens: 0, cache_creation_input_tokens: 0 }, + }, + }, + { + type: "content_block_start", + index: 0, + content_block: { type: "tool_use", id: "toolu_ping_keepalive", name: "write", input: {} }, + }, + { + type: "content_block_delta", + index: 0, + delta: { type: "input_json_delta", partial_json: '{"path":"notes.md",' }, + }, +]; + +const writeToolCallClosing: MockAnthropicEvent[] = [ + { + type: "content_block_delta", + index: 0, + delta: { type: "input_json_delta", partial_json: '"content":"hello world"}' }, + }, + { type: "content_block_stop", index: 0 }, + { + type: "message_delta", + delta: { stop_reason: "tool_use" }, + usage: { input_tokens: 10, output_tokens: 6, cache_read_input_tokens: 0, cache_creation_input_tokens: 0 }, + }, + { type: "message_stop" }, +]; + +function createScriptedClient( + script: ScriptStep[], + counters: { pings: number }, + onIteratorStart: () => void, +): AnthropicMessagesClientLike { + const create = ((_body: unknown, requestOptions?: { signal?: AbortSignal }) => { + const signal = requestOptions?.signal; + const response = new Response(null, { status: 200, headers: { "request-id": "req_ping_keepalive" } }); + const stream = { + async *[Symbol.asyncIterator]() { + onIteratorStart(); + for (const step of script) { + if (step.event === "hang-with-pings") { + // Wedged upstream: no semantic events ever again, but the edge + // keeps the SSE connection alive with keepalive pings. + while (true) { + await waitForDelayOrAbort(step.waitMs, signal); + counters.pings += 1; + yield { type: "ping" }; + } + } + if (step.waitMs > 0) { + await waitForDelayOrAbort(step.waitMs, signal); + } + if (step.event.type === "ping") counters.pings += 1; + yield step.event; + } + }, + }; + return { + async withResponse() { + return { data: stream, response, request_id: "req_ping_keepalive" }; + }, + } as never; + }) as unknown as AnthropicMessagesClientLike["messages"]["create"]; + return { messages: { create } } as AnthropicMessagesClientLike; +} + +async function drainMicrotasks(count: number): Promise { + for (let i = 0; i < count; i++) { + await Promise.resolve(); + } +} + +async function drainMicrotasksUntil(predicate: () => boolean, errorMessage: string): Promise { + for (let i = 0; i < 1000; i++) { + if (predicate()) return; + await Promise.resolve(); + } + throw new Error(errorMessage); +} + +afterEach(() => { + vi.useRealTimers(); + vi.restoreAllMocks(); +}); + +describe("anthropic ping keepalive idle cap", () => { + it("times out a stalled tool-call stream instead of letting pings extend it forever", async () => { + vi.useFakeTimers(); + const counters = { pings: 0 }; + let iteratorStarted = false; + const script: ScriptStep[] = [ + ...writeToolCallOpening.map(event => ({ waitMs: 0, event })), + { waitMs: 500, event: "hang-with-pings" as const }, + ]; + const client = createScriptedClient(script, counters, () => { + iteratorStarted = true; + }); + const providerRetryWait = vi.fn(async () => {}); + + let settled = false; + const resultPromise = streamAnthropic(model, context, { + client, + streamFirstEventTimeoutMs: 1_000, + streamIdleTimeoutMs: 1_000, + providerRetryWait, + }) + .result() + .then(message => { + settled = true; + return message; + }); + + await drainMicrotasksUntil(() => iteratorStarted, "Anthropic mock stream never started"); + await drainMicrotasks(30); + + // Pings arrive every 500 fake-ms while generation is wedged. Drive far + // past the bounded keepalive window (3x idle = 3_000ms) plus one idle + // budget; without the cap the idle deadline is reset by every ping and + // this loop ends with the result still pending (issue #4900's hang). + let stepsRun = 0; + for (let step = 0; step < 40 && !settled; step++) { + vi.advanceTimersByTime(500); + await drainMicrotasks(30); + stepsRun = step + 1; + } + + expect(settled).toBe(true); + // Cap (3_000ms) + idle budget (1_000ms) = fires at 3_500-4_000 fake ms. + expect(stepsRun).toBeLessThanOrEqual(9); + // Keepalives within the window were honored before the watchdog fired. + expect(counters.pings).toBeGreaterThanOrEqual(5); + + const result = await resultPromise; + expect(result.stopReason).toBe("error"); + expect(result.errorMessage).toBe("Anthropic stream stalled while waiting for the next event"); + // Mid-stream idle stalls are terminal for the provider loop (session-level + // auto-retry owns recovery); the provider must not silently re-request. + expect(providerRetryWait).not.toHaveBeenCalled(); + }); + + it("keeps a slow-but-alive stream open across ping-bridged gaps within the cap", async () => { + vi.useFakeTimers(); + const counters = { pings: 0 }; + let iteratorStarted = false; + // Silent generation gap of 1_800ms (> 1_000ms idle budget) bridged by + // pings at t=600 and t=1200, then semantic progress resumes and the + // tool call completes. Pings within the cap must count as liveness. + const script: ScriptStep[] = [ + ...writeToolCallOpening.map(event => ({ waitMs: 0, event })), + { waitMs: 600, event: { type: "ping" } }, + { waitMs: 600, event: { type: "ping" } }, + { waitMs: 600, event: writeToolCallClosing[0]! }, + ...writeToolCallClosing.slice(1).map(event => ({ waitMs: 0, event })), + ]; + const client = createScriptedClient(script, counters, () => { + iteratorStarted = true; + }); + const providerRetryWait = vi.fn(async () => {}); + + let settled = false; + const resultPromise = streamAnthropic(model, context, { + client, + streamFirstEventTimeoutMs: 1_000, + streamIdleTimeoutMs: 1_000, + providerRetryWait, + }) + .result() + .then(message => { + settled = true; + return message; + }); + + await drainMicrotasksUntil(() => iteratorStarted, "Anthropic mock stream never started"); + await drainMicrotasks(30); + + for (let step = 0; step < 30 && !settled; step++) { + vi.advanceTimersByTime(200); + await drainMicrotasks(30); + } + + expect(settled).toBe(true); + expect(counters.pings).toBe(2); + + const result = await resultPromise; + expect(result.errorMessage).toBeUndefined(); + expect(result.stopReason).toBe("toolUse"); + expect(providerRetryWait).not.toHaveBeenCalled(); + expect(JSON.parse(JSON.stringify(result.content))).toEqual([ + { + type: "toolCall", + id: "toolu_ping_keepalive", + name: "write", + arguments: { path: "notes.md", content: "hello world" }, + }, + ]); + }); +}); diff --git a/packages/ai/test/auth-broker-config-discovery.test.ts b/packages/ai/test/auth-broker-config-discovery.test.ts new file mode 100644 index 000000000..67d4033aa --- /dev/null +++ b/packages/ai/test/auth-broker-config-discovery.test.ts @@ -0,0 +1,59 @@ +import { afterEach, beforeEach, describe, expect, test } from "bun:test"; +import * as fs from "node:fs/promises"; +import * as os from "node:os"; +import * as path from "node:path"; +import { resolveAuthBrokerConfig } from "@oh-my-pi/pi-ai/auth-broker"; +import { removeWithRetries } from "../../utils/src/temp"; +import { withEnv } from "./helpers"; + +const SUPPRESS_AUTH_BROKER_ENV = { + OMP_AUTH_BROKER_URL: undefined, + OMP_AUTH_BROKER_TOKEN: undefined, +} as const; + +describe("resolveAuthBrokerConfig config discovery", () => { + let agentDir = ""; + + beforeEach(async () => { + agentDir = await fs.mkdtemp(path.join(os.tmpdir(), "pi-ai-auth-broker-config-")); + }); + + afterEach(async () => { + if (agentDir) { + await removeWithRetries(agentDir); + agentDir = ""; + } + }); + + test("resolves broker URL and token from config.yaml when config.yml is absent", async () => { + await Bun.write( + path.join(agentDir, "config.yaml"), + "auth.broker.url: https://yaml-broker.example/v1\nauth.broker.token: yaml-token\n", + ); + + await withEnv(SUPPRESS_AUTH_BROKER_ENV, async () => { + await expect(resolveAuthBrokerConfig({ agentDir })).resolves.toEqual({ + url: "https://yaml-broker.example/v1", + token: "yaml-token", + }); + }); + }); + + test("prefers config.yml over config.yaml when both exist", async () => { + await Bun.write( + path.join(agentDir, "config.yaml"), + "auth.broker.url: https://yaml-broker.example/v1\nauth.broker.token: yaml-token\n", + ); + await Bun.write( + path.join(agentDir, "config.yml"), + "auth.broker.url: https://yml-broker.example/v1\nauth.broker.token: yml-token\n", + ); + + await withEnv(SUPPRESS_AUTH_BROKER_ENV, async () => { + await expect(resolveAuthBrokerConfig({ agentDir })).resolves.toEqual({ + url: "https://yml-broker.example/v1", + token: "yml-token", + }); + }); + }); +}); diff --git a/packages/ai/test/cursor-exec-handlers.test.ts b/packages/ai/test/cursor-exec-handlers.test.ts index e77bc484f..5063e81ec 100644 --- a/packages/ai/test/cursor-exec-handlers.test.ts +++ b/packages/ai/test/cursor-exec-handlers.test.ts @@ -1,14 +1,25 @@ import { describe, expect, it } from "bun:test"; +import { create } from "@bufbuild/protobuf"; import { + type BlockState, buildCursorHistoryForTest, buildCursorSystemPromptJsons, emptyGrepPatternRejection, + handleServerMessage, resolveExecHandler, streamCursor, + type ToolCallState, } from "@oh-my-pi/pi-ai/providers/cursor"; -import type { Context, Model } from "@oh-my-pi/pi-ai/types"; +import { streamCursor as lazyStreamCursor, setCursorProviderModule } from "@oh-my-pi/pi-ai/providers/register-builtins"; +import type { AssistantMessage, Context, CursorExecHandlers, Model, ToolResultMessage } from "@oh-my-pi/pi-ai/types"; +import { AssistantMessageEventStream } from "@oh-my-pi/pi-ai/utils/event-stream"; import { buildModel } from "@oh-my-pi/pi-catalog/build"; -import type { AgentRunRequest } from "@oh-my-pi/pi-catalog/discovery/cursor-gen/agent_pb"; +import { + type AgentRunRequest, + AgentServerMessageSchema, + ExecServerMessageSchema, + ReadArgsSchema, +} from "@oh-my-pi/pi-catalog/discovery/cursor-gen/agent_pb"; const cursorModel: Model<"cursor-agent"> = buildModel({ id: "cursor-composer-2.5", @@ -361,3 +372,188 @@ describe("Cursor grepArgs empty-pattern guard (issue #4574)", () => { expect(emptyGrepPatternRejection("\t\n", "src/**/*.ts")).toContain('"src/**/*.ts"'); }); }); + +function cursorAssistantMessage(): AssistantMessage { + return { + role: "assistant", + content: [], + api: "cursor-agent", + provider: "cursor", + model: "cursor-composer-2.5", + usage: { + input: 0, + output: 0, + cacheRead: 0, + cacheWrite: 0, + totalTokens: 0, + cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 }, + }, + stopReason: "stop", + timestamp: 0, + }; +} + +function newBlockState(): BlockState { + let textBlock: BlockState["currentTextBlock"] = null; + let thinkingBlock: BlockState["currentThinkingBlock"] = null; + let toolCall: ToolCallState | null = null; + return { + get currentTextBlock() { + return textBlock; + }, + get currentThinkingBlock() { + return thinkingBlock; + }, + get currentToolCall() { + return toolCall; + }, + firstTokenTime: undefined, + setTextBlock: b => { + textBlock = b; + }, + setThinkingBlock: b => { + thinkingBlock = b; + }, + setToolCall: t => { + toolCall = t; + }, + setFirstTokenTime: () => {}, + }; +} + +describe("Cursor exec local-work tracking (issue #4593)", () => { + it("marks the stream busy for the duration of a local exec handler", async () => { + const output = cursorAssistantMessage(); + const stream = new AssistantMessageEventStream(); + const state = newBlockState(); + const written: unknown[] = []; + const h2Request = { + write: (chunk: unknown) => { + written.push(chunk); + return true; + }, + } as unknown as Parameters[5]; + const handlerGate = Promise.withResolvers(); + const execHandlers: CursorExecHandlers = { + async read(args) { + await handlerGate.promise; + return { + role: "toolResult", + toolCallId: args.toolCallId, + toolName: "read", + content: [{ type: "text", text: "file contents" }], + isError: false, + timestamp: 1, + } satisfies ToolResultMessage; + }, + }; + const serverMsg = create(AgentServerMessageSchema, { + message: { + case: "execServerMessage", + value: create(ExecServerMessageSchema, { + id: 1, + execId: "exec-1", + message: { + case: "readArgs", + value: create(ReadArgsSchema, { path: "/tmp/slow-file", toolCallId: "call-read-1" }), + }, + }), + }, + }); + + expect(stream.hasPendingLocalWork).toBe(false); + const dispatch = handleServerMessage( + serverMsg, + output, + stream, + state, + new Map(), + h2Request, + execHandlers, + undefined, + { sawTokenDelta: false }, + [], + ); + + // The exec round-trip is in flight: the stream must advertise local + // work so the lazy idle watchdog defers instead of aborting. + expect(stream.hasPendingLocalWork).toBe(true); + + handlerGate.resolve(); + await dispatch; + + expect(stream.hasPendingLocalWork).toBe(false); + // The read result went back out on the exec channel. + expect(written.length).toBe(1); + }); + + it("survives a local exec tool outliving the lazy idle budget end to end", async () => { + const workDone = Promise.withResolvers(); + // The tracked work completes only once the lazy watchdog has consulted + // the stream's local-work state at two expired deadlines, proving the + // idle budget was truly exceeded while the exec tool ran. + class ProbedStream extends AssistantMessageEventStream { + probeCalls = 0; + override get hasPendingLocalWork(): boolean { + this.probeCalls++; + if (this.probeCalls >= 2) workDone.resolve(); + return super.hasPendingLocalWork; + } + } + const source = new ProbedStream(); + let providerSignal: AbortSignal | undefined; + setCursorProviderModule({ + streamCursor: (_model, _context, options) => { + providerSignal = options.signal; + void (async () => { + const partial = cursorAssistantMessage(); + source.push({ type: "start", partial }); + source.push({ type: "text_delta", contentIndex: 0, delta: "spawning local tool", partial }); + await source.trackLocalWork(workDone.promise); + const message = cursorAssistantMessage(); + source.push({ type: "done", reason: "stop", message }); + })(); + return source; + }, + }); + + const stream = lazyStreamCursor(cursorModel, { messages: [] }, { apiKey: "test", streamIdleTimeoutMs: 5 }); + const result = await stream.result(); + + expect(providerSignal?.aborted).toBe(false); + expect(source.probeCalls).toBeGreaterThanOrEqual(2); + expect(result.stopReason).toBe("stop"); + }); + + it("still aborts a silent cursor stream with no local work in flight", async () => { + const partial = cursorAssistantMessage(); + let providerSignal: AbortSignal | undefined; + const source = { + async *[Symbol.asyncIterator]() { + yield { type: "start", partial } as const; + yield { type: "text_delta", contentIndex: 0, delta: "hello", partial } as const; + const stalled = Promise.withResolvers(); + if (providerSignal?.aborted) { + stalled.reject(new Error("Request was aborted")); + } + providerSignal?.addEventListener("abort", () => stalled.reject(new Error("Request was aborted")), { + once: true, + }); + await stalled.promise; + }, + } as unknown as AssistantMessageEventStream; + setCursorProviderModule({ + streamCursor: (_model, _context, options) => { + providerSignal = options.signal; + return source; + }, + }); + + const stream = lazyStreamCursor(cursorModel, { messages: [] }, { apiKey: "test", streamIdleTimeoutMs: 10 }); + const result = await stream.result(); + + expect(providerSignal?.aborted).toBe(true); + expect(result.stopReason).toBe("error"); + expect(result.errorMessage).toBe("Provider stream stalled while waiting for the next event"); + }); +}); diff --git a/packages/ai/test/issue-4593-repro.test.ts b/packages/ai/test/issue-4593-repro.test.ts new file mode 100644 index 000000000..e79f4f1e1 --- /dev/null +++ b/packages/ai/test/issue-4593-repro.test.ts @@ -0,0 +1,190 @@ +import { describe, expect, it } from "bun:test"; +import { setBedrockProviderModule, streamBedrock } from "@oh-my-pi/pi-ai/providers/register-builtins"; +import type { AssistantMessage, Context, Model } from "@oh-my-pi/pi-ai/types"; +import { AssistantMessageEventStream } from "@oh-my-pi/pi-ai/utils/event-stream"; +import { iterateWithIdleTimeout } from "@oh-my-pi/pi-ai/utils/idle-iterator"; +import { buildModel } from "@oh-my-pi/pi-catalog/build"; + +// Issue #4593: the generic lazy stream watchdog treats "no AssistantMessageEvent" +// as "provider stalled". During a Cursor exec-channel round-trip the server is +// waiting on OUR local tool result and legitimately sends nothing, so a local +// tool outliving the idle budget aborted a healthy stream with "Provider stream +// stalled while waiting for the next event". Provider streams now advertise +// pending local work and the watchdog slides its deadline instead of aborting. +// +// These tests exercise the real watchdog timer against the platform clock (that +// timer IS the unit under test), but never guess durations: the simulated local +// work completes only once the watchdog has demonstrably reached an expired +// deadline and consulted the local-work probe, so the tests stay causal on a +// loaded machine. Budgets are a few milliseconds. + +function createModel(): Model<"bedrock-converse-stream"> { + return buildModel({ + id: "mock-bedrock", + name: "Mock Bedrock", + api: "bedrock-converse-stream", + provider: "amazon-bedrock", + baseUrl: "https://example.invalid", + reasoning: false, + input: ["text"], + cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, + contextWindow: 8192, + maxTokens: 2048, + }); +} + +function createAssistantMessage(): AssistantMessage { + return { + role: "assistant", + content: [{ type: "text", text: "ok" }], + api: "bedrock-converse-stream", + provider: "amazon-bedrock", + model: "mock-bedrock", + usage: { + input: 0, + output: 0, + cacheRead: 0, + cacheWrite: 0, + totalTokens: 0, + cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 }, + }, + stopReason: "stop", + timestamp: Date.now(), + }; +} + +const baseContext: Context = { messages: [] }; + +describe("idle watchdog local-work deferral (issue #4593)", () => { + it("slides the idle deadline while consumer-side local work is pending", async () => { + const workDone = Promise.withResolvers(); + let probeCalls = 0; + let busy = true; + async function* source() { + yield "first"; + // The "local tool": finishes only after the watchdog has hit an + // expired deadline twice and deferred both times. + await workDone.promise; + busy = false; + yield "second"; + } + let idleFired = false; + const items: string[] = []; + for await (const item of iterateWithIdleTimeout(source(), { + idleTimeoutMs: 5, + errorMessage: "stalled", + onIdle: () => { + idleFired = true; + }, + hasPendingLocalWork: () => { + probeCalls++; + if (probeCalls >= 2) workDone.resolve(); + return busy; + }, + })) { + items.push(item); + } + expect(items).toEqual(["first", "second"]); + expect(probeCalls).toBeGreaterThanOrEqual(2); + expect(idleFired).toBe(false); + }); + + it("still aborts a silent stream once local work has finished", async () => { + const workDone = Promise.withResolvers(); + let busy = true; + async function* source() { + yield "first"; + await workDone.promise; + busy = false; + // The provider genuinely stalls after the local work completed. + await new Promise(() => {}); + yield "never"; + } + const items: string[] = []; + let error: Error | undefined; + try { + for await (const item of iterateWithIdleTimeout(source(), { + idleTimeoutMs: 5, + errorMessage: "stalled", + hasPendingLocalWork: () => { + workDone.resolve(); + return busy; + }, + })) { + items.push(item); + } + } catch (err) { + error = err as Error; + } + expect(items).toEqual(["first"]); + expect(error?.message).toBe("stalled"); + }); + + it("slides the first-event deadline while local work is pending", async () => { + const workDone = Promise.withResolvers(); + let probeCalls = 0; + let busy = true; + async function* source() { + // Local bridge work before the model has produced any event. + await workDone.promise; + yield "first"; + } + const items: string[] = []; + for await (const item of iterateWithIdleTimeout(source(), { + idleTimeoutMs: 5, + firstItemTimeoutMs: 5, + errorMessage: "stalled", + firstItemErrorMessage: "first event timed out", + hasPendingLocalWork: () => { + probeCalls++; + if (probeCalls >= 2) workDone.resolve(); + return busy; + }, + })) { + items.push(item); + busy = false; + } + expect(items).toEqual(["first"]); + expect(probeCalls).toBeGreaterThanOrEqual(2); + }); + + it("does not abort a lazy provider stream while tracked local work outlives the idle budget", async () => { + const workDone = Promise.withResolvers(); + // Counts how often the lazy wrapper's watchdog consults the stream's + // local-work state at an expired deadline; the tracked work completes + // only after two deferrals, proving the budget was truly exceeded. + class ProbedStream extends AssistantMessageEventStream { + probeCalls = 0; + override get hasPendingLocalWork(): boolean { + this.probeCalls++; + if (this.probeCalls >= 2) workDone.resolve(); + return super.hasPendingLocalWork; + } + } + const source = new ProbedStream(); + let providerSignal: AbortSignal | undefined; + setBedrockProviderModule({ + streamBedrock: (_model, _context, options) => { + providerSignal = options.signal; + void (async () => { + const partial = createAssistantMessage(); + source.push({ type: "start", partial }); + source.push({ type: "text_delta", contentIndex: 0, delta: "running a local tool", partial }); + // Server-driven local tool run: no events flow while the + // tracked work is pending. + await source.trackLocalWork(workDone.promise); + source.push({ type: "done", reason: "stop", message: createAssistantMessage() }); + })(); + return source; + }, + }); + + const stream = streamBedrock(createModel(), baseContext, { streamIdleTimeoutMs: 5 }); + const result = await stream.result(); + + expect(providerSignal?.aborted).toBe(false); + expect(source.probeCalls).toBeGreaterThanOrEqual(2); + expect(result.stopReason).toBe("stop"); + expect(result.errorMessage).toBeUndefined(); + }); +}); diff --git a/packages/ai/test/openai-codex-stream.test.ts b/packages/ai/test/openai-codex-stream.test.ts index e7fa92c2e..505039c5c 100644 --- a/packages/ai/test/openai-codex-stream.test.ts +++ b/packages/ai/test/openai-codex-stream.test.ts @@ -442,6 +442,129 @@ describe("openai-codex streaming", () => { expect(capturedText).toEqual({ verbosity: "low" }); }); + it("preserves streamed reasoning when the done item has no summary text", async () => { + const token = createCodexTestToken(); + const model = { ...createCodexTestModel("https://chatgpt.com/backend-api"), preferWebsockets: false }; + const events = [ + { + type: "response.output_item.added", + output_index: 0, + item: { type: "reasoning", id: "rs_1", summary: [] }, + }, + { + type: "response.reasoning_summary_part.added", + output_index: 0, + item_id: "rs_1", + summary_index: 0, + part: { type: "summary_text", text: "" }, + }, + { + type: "response.reasoning_summary_text.delta", + output_index: 0, + item_id: "rs_1", + summary_index: 0, + delta: "streamed thinking", + }, + { + type: "response.output_item.done", + output_index: 0, + item: { type: "reasoning", id: "rs_1", summary: [] }, + }, + { + type: "response.output_item.added", + output_index: 1, + item: { type: "message", id: "msg_1", role: "assistant", status: "in_progress", content: [] }, + }, + { + type: "response.content_part.added", + output_index: 1, + item_id: "msg_1", + part: { type: "output_text", text: "" }, + }, + { type: "response.output_text.delta", output_index: 1, item_id: "msg_1", delta: "done" }, + { + type: "response.output_item.done", + output_index: 1, + item: { + type: "message", + id: "msg_1", + role: "assistant", + status: "completed", + content: [{ type: "output_text", text: "done" }], + }, + }, + { + type: "response.completed", + response: { + id: "resp_1", + status: "completed", + usage: { + input_tokens: 5, + output_tokens: 3, + total_tokens: 8, + input_tokens_details: { cached_tokens: 0 }, + }, + }, + }, + ]; + const sse = `${events.map(event => `data: ${JSON.stringify(event)}`).join("\n\n")}\n\n`; + const fetchMock: FetchImpl = async () => + new Response(sse, { status: 200, headers: { "content-type": "text/event-stream" } }); + + const result = await streamOpenAICodexResponses(model, createCodexTestContext(), { + apiKey: token, + fetch: fetchMock, + }).result(); + + expect(result.content.find(block => block.type === "thinking")?.thinking).toBe("streamed thinking"); + }); + + it("streams raw reasoning text deltas into the final thinking block", async () => { + const token = createCodexTestToken(); + const model = { ...createCodexTestModel("https://chatgpt.com/backend-api"), preferWebsockets: false }; + const events = [ + { + type: "response.output_item.added", + output_index: 0, + item: { type: "reasoning", id: "rs_raw", summary: [] }, + }, + { + type: "response.reasoning_text.delta", + output_index: 0, + item_id: "rs_raw", + delta: "raw streamed thinking", + }, + { + type: "response.output_item.done", + output_index: 0, + item: { type: "reasoning", id: "rs_raw", summary: [] }, + }, + { + type: "response.completed", + response: { + id: "resp_raw", + status: "completed", + usage: { + input_tokens: 5, + output_tokens: 3, + total_tokens: 8, + input_tokens_details: { cached_tokens: 0 }, + }, + }, + }, + ]; + const sse = `${events.map(event => `data: ${JSON.stringify(event)}`).join("\n\n")}\n\n`; + const fetchMock: FetchImpl = async () => + new Response(sse, { status: 200, headers: { "content-type": "text/event-stream" } }); + + const result = await streamOpenAICodexResponses(model, createCodexTestContext(), { + apiKey: token, + fetch: fetchMock, + }).result(); + + expect(result.content.find(block => block.type === "thinking")?.thinking).toBe("raw streamed thinking"); + }); + it("maps end_turn=false on the terminal event to a pause_turn stop", async () => { const tempDir = TempDir.createSync("@pi-codex-stream-"); setAgentDir(tempDir.path()); diff --git a/packages/ai/test/openai-codex.test.ts b/packages/ai/test/openai-codex.test.ts index e28633fd1..1c7492afc 100644 --- a/packages/ai/test/openai-codex.test.ts +++ b/packages/ai/test/openai-codex.test.ts @@ -314,6 +314,36 @@ describe("openai-codex reasoning effort validation", () => { }); }); +describe("openai-codex reasoning effort wire mapping", () => { + it("shifts gpt-5.6 user efforts one wire tier up via the baked effort map", async () => { + const model = createCodexModel("gpt-5.6-sol"); + const shifted = [ + ["minimal", "low"], + ["low", "medium"], + ["medium", "high"], + ["high", "xhigh"], + ["xhigh", "max"], + ] as const; + + for (const [requested, wire] of shifted) { + const transformed = await transformRequestBody({ model: model.id }, model, { + reasoningEffort: requested, + }); + expect(transformed.reasoning?.effort).toBe(wire); + } + }); + + it("keeps pre-5.6 efforts unshifted and passes none through unmapped", async () => { + const gpt55 = createCodexModel("gpt-5.5"); + const unshifted = await transformRequestBody({ model: gpt55.id }, gpt55, { reasoningEffort: "xhigh" }); + expect(unshifted.reasoning?.effort).toBe("xhigh"); + + const gpt56 = createCodexModel("gpt-5.6-sol"); + const none = await transformRequestBody({ model: gpt56.id }, gpt56, { reasoningEffort: "none" }); + expect(none.reasoning?.effort).toBe("none"); + }); +}); + describe("openai-codex error parsing", () => { it("produces friendly usage-limit messages and rate limits", async () => { const resetAt = Math.floor(Date.now() / 1000) + 600; diff --git a/packages/ai/test/openai-responses-stream-terminal.test.ts b/packages/ai/test/openai-responses-stream-terminal.test.ts index 95c501834..63bc5f331 100644 --- a/packages/ai/test/openai-responses-stream-terminal.test.ts +++ b/packages/ai/test/openai-responses-stream-terminal.test.ts @@ -311,6 +311,49 @@ describe("processResponsesStream: lost output_item.added recovery", () => { expect(second.thinkingSignature).toBeDefined(); }); + test("preserves streamed reasoning when the done item has no summary text", async () => { + const output = makeOutput(); + const stream = { push: () => {}, end: () => {} } as never; + + await processResponsesStream( + makeStream([ + { + type: "response.output_item.added", + output_index: 0, + item: { type: "reasoning", id: "rs_1", summary: [] }, + }, + { + type: "response.reasoning_summary_part.added", + output_index: 0, + item_id: "rs_1", + summary_index: 0, + part: { type: "summary_text", text: "" }, + }, + { + type: "response.reasoning_summary_text.delta", + output_index: 0, + item_id: "rs_1", + summary_index: 0, + delta: "streamed thinking", + }, + { + type: "response.output_item.done", + output_index: 0, + item: { type: "reasoning", id: "rs_1", summary: [] }, + }, + { type: "response.completed", response: { id: "resp_reasoning", status: "completed" } }, + ]), + output, + stream, + makeModel(), + ); + + const block = output.content[0]; + if (block?.type !== "thinking") throw new Error("expected a thinking block"); + expect(block.thinking).toBe("streamed thinking"); + expect(block.thinkingSignature).toBeDefined(); + }); + test("treats content_filter incomplete responses as errors, not length", async () => { const output = makeOutput(); const stream = { push: () => {}, end: () => {} } as never; diff --git a/packages/ai/test/provider-registry.test.ts b/packages/ai/test/provider-registry.test.ts index 2ed3d88bd..a72eb59c7 100644 --- a/packages/ai/test/provider-registry.test.ts +++ b/packages/ai/test/provider-registry.test.ts @@ -81,6 +81,7 @@ describe("provider registry auth surface", () => { "google-antigravity", "google-gemini-cli", "openai-codex", + "xai-oauth", ].sort(), ); expect(PASTE_CODE_LOGIN_PROVIDERS.has("zenmux")).toBe(false); diff --git a/packages/catalog/CHANGELOG.md b/packages/catalog/CHANGELOG.md index 8bf50c3a7..bece2a3cd 100644 --- a/packages/catalog/CHANGELOG.md +++ b/packages/catalog/CHANGELOG.md @@ -2,6 +2,30 @@ ## [Unreleased] +## [16.3.14] - 2026-07-09 + +### Added + +- Added support for GPT-5.6 (Luna, Sol, Terra) model variants +- Enabled expanded five-tier reasoning effort scale (minimal to xhigh) for GPT-5.6 models +- Added GPT-5.6 (Terra/Luna/Sol) support for the new `max` reasoning tier: on wire-effort APIs (OpenAI Responses, Codex, Azure, openai-compat/OpenRouter models that advertise reasoning) user efforts shift up one notch — `xhigh` sends `max`, `high` sends `xhigh` — mirroring the Claude Fable/Opus 4.7+ five-tier mapping, and the exposed ladder becomes `minimal..xhigh` with `minimal` reaching the native `low` tier. Devin's per-tier GPT-5.6 sibling rows now collapse into `gpt-5-6-{luna,sol,terra}` logical models with the same shifted routing (`xhigh` → `-max`), plus `-fast` families that keep the direct `low..xhigh` `-priority` scale since Devin serves no `-max-priority` tier. + +## [16.3.13] - 2026-07-09 + +### Added + +- Added support for Grok 4.5 across multiple providers +- Added support for GPT-5.6 series models (Luna, Sol, Terra) +- Added Aion 3.0 and 3.0 Mini models +- Added Kuaishou KAT-Coder v2.5 models +- Added Nex-N2-Mini and SWE-1.7 series models +- Added Hy3 models and free variants + +### Changed + +- Updated cost and token configurations for various models across providers +- Renamed several models for consistency (e.g., MiniMax M3, Gemma 4 31B, Qwen variants) + ## [16.3.12] - 2026-07-08 ### Fixed diff --git a/packages/catalog/package.json b/packages/catalog/package.json index e2272c3d0..dc1388a19 100644 --- a/packages/catalog/package.json +++ b/packages/catalog/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/pi-catalog", - "version": "16.3.12", + "version": "16.3.14", "description": "Model catalog for omp: bundled model database, provider discovery descriptors, model identity, classification, and equivalence", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/catalog/src/discovery/cursor.ts b/packages/catalog/src/discovery/cursor.ts index 880aefec5..65d6ca645 100644 --- a/packages/catalog/src/discovery/cursor.ts +++ b/packages/catalog/src/discovery/cursor.ts @@ -13,6 +13,13 @@ const CURSOR_GET_USABLE_MODELS_PATH = "/agent.v1.AgentService/GetUsableModels"; const DEFAULT_CONTEXT_WINDOW = 200_000; const DEFAULT_MAX_TOKENS = 64_000; +/** + * Model-id families whose native catalogs (anthropic, openai/openai-codex, + * google) are multimodal. Cursor-only or text-only families (`composer-*`, + * `grok-code-*`) intentionally stay outside this pattern. + */ +const CURSOR_MULTIMODAL_ID_PATTERN = /claude|gemini|gpt-|codex/; + const OptionalDisplayNameSchema = type("unknown").pipe(raw => (typeof raw === "string" ? raw : undefined)); const CursorAliasesSchema = type("unknown").pipe(raw => { if (Array.isArray(raw)) { @@ -292,7 +299,7 @@ function normalizeCursorModel( provider: "cursor", baseUrl: baseUrlOverride ?? CURSOR_DEFAULT_BASE_URL, reasoning, - input: ["text"], + input: inferInputFromCursorId(id), cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, contextWindow: DEFAULT_CONTEXT_WINDOW, maxTokens: DEFAULT_MAX_TOKENS, @@ -312,3 +319,18 @@ function pickModelDisplayName(model: CursorModelDetailsValue, fallbackId: string } return fallbackId; } + +/** + * Infers input modalities for Cursor models without a bundled reference. + * + * `GetUsableModels` carries no per-model modality metadata, so classification + * falls back to the model family: families that are multimodal in OMP's own + * native catalogs accept images, everything else stays text-only. Mirrors + * `inferInputFromGeminiId` in ./gemini.ts. + */ +function inferInputFromCursorId(id: string): ("text" | "image")[] { + if (CURSOR_MULTIMODAL_ID_PATTERN.test(id.toLowerCase())) { + return ["text", "image"]; + } + return ["text"]; +} diff --git a/packages/catalog/src/model-manager.ts b/packages/catalog/src/model-manager.ts index eccec451f..33fc8cf88 100644 --- a/packages/catalog/src/model-manager.ts +++ b/packages/catalog/src/model-manager.ts @@ -35,7 +35,7 @@ export interface ModelManagerOptions = { }; /** - * Effort → wire-value map for the 5-tier adaptive scale (Opus 4.7+ and - * Fable/Mythos 5 on the Messages API). User-facing efforts shift up one notch - * so the top tier reaches the genuine "max" and "high" lands on Anthropic's - * recommended "xhigh" coding/agentic default. + * Effort → wire-value map for a shifted five-tier scale (`low..max`): + * user-facing efforts shift up one notch so the top tier reaches the genuine + * "max" and "high" lands on the recommended "xhigh" coding/agentic default. + * Used by Anthropic adaptive models with a real xhigh tier (Opus 4.7+ and + * Fable/Mythos 5 on the Messages API) and by GPT-5.6+ wire-effort models, + * which expose the same genuine `max` tier above `xhigh`. */ -export const ANTHROPIC_ADAPTIVE_EFFORT_MAP_5_TIER: Readonly>> = { +export const SHIFTED_FIVE_TIER_EFFORT_MAP: Readonly>> = { [Effort.Minimal]: "low", [Effort.Low]: "medium", [Effort.Medium]: "high", @@ -295,6 +298,27 @@ function isOpenAICompatReasoningApi(api: Api): boolean { return api === "openai-completions" || api === "openrouter"; } +/** + * GPT-5.6+ addressed through a wire `reasoning.effort`/`reasoning_effort` + * field, where the shifted five-tier map applies. Devin (`devin-agent`) + * selects effort by routing to per-tier sibling model ids instead and must + * stay unmapped. + */ +function isGpt56PlusWireEffortModel(spec: ModelSpec): boolean { + switch (spec.api) { + case "openai-responses": + case "openai-codex-responses": + case "azure-openai-responses": + case "openai-completions": + case "openrouter": + break; + default: + return false; + } + const parsed = parseOpenAIModel(bareModelId(spec.id)); + return parsed !== null && semverGte(parsed.version, "5.6"); +} + function getModelDefinedEfforts( spec: ModelSpec, compat: CompatOf, @@ -313,6 +337,12 @@ function getModelDefinedEfforts( if (isSakanaFuguReasoningModel(spec)) { return FUGU_REASONING_EFFORTS; } + if (isGpt56PlusWireEffortModel(spec)) { + // Normalize stale baked/discovered `low..xhigh` surfaces to the full + // five-tier ladder so the shifted map keeps the native `low` tier + // reachable (user `minimal`). + return DEFAULT_REASONING_EFFORTS_WITH_XHIGH; + } return isOpenAICompatReasoningApi(spec.api) && (isMinimaxM2FamilyModelId(spec.id) || isOpenAIGptOssModelId(spec.id) || @@ -373,7 +403,7 @@ function inferDetectedEffortMap( return MINIMAX_ANTHROPIC_ADAPTIVE_EFFORT_MAP; } return anthropicModelHasRealXHighEffort(spec, parsedModel) - ? ANTHROPIC_ADAPTIVE_EFFORT_MAP_5_TIER + ? SHIFTED_FIVE_TIER_EFFORT_MAP : ANTHROPIC_ADAPTIVE_EFFORT_MAP_4_TIER; } // GLM-5.2 coding SKUs accept `reasoning_effort`, but the effort dialect is @@ -397,6 +427,9 @@ function inferDetectedEffortMap( if (isSakanaFuguReasoningModel(spec)) { return FUGU_REASONING_EFFORT_MAP; } + if (isGpt56PlusWireEffortModel(spec)) { + return SHIFTED_FIVE_TIER_EFFORT_MAP; + } if (!isOpenAICompatReasoningApi(spec.api)) { return undefined; } @@ -446,7 +479,7 @@ function getOpenRouterAnthropicReasoningEffortMap(modelId: string): EffortMap | if (!isAnthropicAdaptiveGenAtLeast(parsed, "4.6")) return undefined; const hasRealXHigh = isAnthropicAdaptiveGenAtLeast(parsed, "4.7"); - return hasRealXHigh ? ANTHROPIC_ADAPTIVE_EFFORT_MAP_5_TIER : ANTHROPIC_ADAPTIVE_EFFORT_MAP_4_TIER; + return hasRealXHigh ? SHIFTED_FIVE_TIER_EFFORT_MAP : ANTHROPIC_ADAPTIVE_EFFORT_MAP_4_TIER; } function inferSupportedEfforts( @@ -474,6 +507,11 @@ function inferOpenAISupportedEfforts(model: OpenAIModel): readonly Effort[] { if (model.variant === "codex-mini" && semverEqual(model.version, "5.1")) { return GPT_5_1_CODEX_MINI_EFFORTS; } + // 5.6+ exposes the full five-tier ladder: the shifted wire map spans + // low..max, with user `minimal` reaching the native `low` tier. + if (semverGte(model.version, "5.6")) { + return DEFAULT_REASONING_EFFORTS_WITH_XHIGH; + } if (semverGte(model.version, "5.2")) { return GPT_5_2_PLUS_EFFORTS; } diff --git a/packages/catalog/src/models.json b/packages/catalog/src/models.json index 6493e482b..55d87779c 100644 --- a/packages/catalog/src/models.json +++ b/packages/catalog/src/models.json @@ -4972,7 +4972,7 @@ }, "minimax/minimax-m3": { "id": "minimax/minimax-m3", - "name": "MiniMax-M3", + "name": "MiniMax M3", "api": "openai-completions", "provider": "aimlapi", "baseUrl": "https://api.aimlapi.com/v1", @@ -14269,7 +14269,7 @@ "cost": { "input": 0.55, "output": 1.65, - "cacheRead": 0, + "cacheRead": 0.55, "cacheWrite": 0 }, "contextWindow": 161000, @@ -14286,13 +14286,13 @@ "text" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.14, + "output": 0.28, + "cacheRead": 0.07, "cacheWrite": 0 }, "contextWindow": 1048576, - "maxTokens": 384000, + "maxTokens": 1048576, "thinking": { "mode": "effort", "efforts": [ @@ -14322,13 +14322,13 @@ "text" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 1.74, + "output": 3.48, + "cacheRead": 0.14, "cacheWrite": 0 }, "contextWindow": 1048576, - "maxTokens": 393216, + "maxTokens": 1048576, "thinking": { "mode": "effort", "efforts": [ @@ -14349,7 +14349,7 @@ }, "google/gemma-4-31B-it": { "id": "google/gemma-4-31B-it", - "name": "Gemma 4 31B Instruct", + "name": "Gemma 4 31B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14359,13 +14359,13 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.12, + "output": 0.35, + "cacheRead": 0.09, "cacheWrite": 0 }, "contextWindow": 262144, - "maxTokens": 131072, + "maxTokens": 262144, "thinking": { "mode": "effort", "efforts": [ @@ -14388,9 +14388,9 @@ "text" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.05, + "output": 0.1, + "cacheRead": 0.05, "cacheWrite": 0 }, "contextWindow": 131072, @@ -14398,7 +14398,7 @@ }, "JetBrains/Mellum2-12B-A2.5B-Instruct": { "id": "JetBrains/Mellum2-12B-A2.5B-Instruct", - "name": "JetBrains/Mellum2-12B-A2.5B-Instruct", + "name": "Mellum2 12B A2.5B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14407,13 +14407,13 @@ "text" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.05, + "output": 0.1, + "cacheRead": 0.05, "cacheWrite": 0 }, - "contextWindow": null, - "maxTokens": null + "contextWindow": 131072, + "maxTokens": 131072 }, "meta-llama/Llama-3.1-70B-Instruct": { "id": "meta-llama/Llama-3.1-70B-Instruct", @@ -14428,7 +14428,7 @@ "cost": { "input": 0.8, "output": 0.8, - "cacheRead": 0, + "cacheRead": 0.8, "cacheWrite": 0 }, "contextWindow": 128000, @@ -14436,7 +14436,7 @@ }, "meta-llama/Llama-3.1-8B-Instruct": { "id": "meta-llama/Llama-3.1-8B-Instruct", - "name": "Meta-Llama-3.1-8B-Instruct", + "name": "Llama 3.1 8B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14447,7 +14447,7 @@ "cost": { "input": 0.22, "output": 0.22, - "cacheRead": 0, + "cacheRead": 0.22, "cacheWrite": 0 }, "contextWindow": 128000, @@ -14455,7 +14455,7 @@ }, "meta-llama/Llama-3.3-70B-Instruct": { "id": "meta-llama/Llama-3.3-70B-Instruct", - "name": "Llama-3.3-70B-Instruct", + "name": "Llama 3.3 70B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14466,7 +14466,7 @@ "cost": { "input": 0.71, "output": 0.71, - "cacheRead": 0, + "cacheRead": 0.71, "cacheWrite": 0 }, "contextWindow": 128000, @@ -14494,7 +14494,7 @@ }, "microsoft/Phi-4-mini-instruct": { "id": "microsoft/Phi-4-mini-instruct", - "name": "Phi-4-mini-instruct", + "name": "Phi 4 Mini 3.8B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14505,7 +14505,7 @@ "cost": { "input": 0.08, "output": 0.35, - "cacheRead": 0, + "cacheRead": 0.08, "cacheWrite": 0 }, "contextWindow": 128000, @@ -14524,7 +14524,7 @@ "cost": { "input": 0.3, "output": 1.2, - "cacheRead": 0, + "cacheRead": 0.3, "cacheWrite": 0 }, "contextWindow": 196608, @@ -14551,9 +14551,9 @@ "image" ], "cost": { - "input": 0.5, - "output": 2.85, - "cacheRead": 0, + "input": 0.6, + "output": 3, + "cacheRead": 0.1, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14571,7 +14571,7 @@ }, "moonshotai/Kimi-K2.6": { "id": "moonshotai/Kimi-K2.6", - "name": "Kimi-K2.6", + "name": "Kimi K2.6", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14581,9 +14581,9 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.95, + "output": 4, + "cacheRead": 0.16, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14611,9 +14611,9 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.94, + "output": 4, + "cacheRead": 0.19, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14631,7 +14631,7 @@ }, "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8": { "id": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8", - "name": "NVIDIA Nemotron 3 Super 120B", + "name": "Nemotron 3 Super", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14642,7 +14642,7 @@ "cost": { "input": 0.2, "output": 0.8, - "cacheRead": 0, + "cacheRead": 0.2, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14660,22 +14660,32 @@ }, "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B": { "id": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", - "name": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", + "name": "Nemotron 3 Ultra", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", - "reasoning": false, + "reasoning": true, "input": [ "text" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.75, + "output": 2.75, + "cacheRead": 0.15, "cacheWrite": 0 }, - "contextWindow": null, - "maxTokens": null + "contextWindow": 262144, + "maxTokens": 262144, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } }, "openai/gpt-oss-120b": { "id": "openai/gpt-oss-120b", @@ -14688,9 +14698,9 @@ "text" ], "cost": { - "input": 0.15, - "output": 0.6, - "cacheRead": 0, + "input": 0.04, + "output": 0.14, + "cacheRead": 0.04, "cacheWrite": 0 }, "contextWindow": 131072, @@ -14715,9 +14725,9 @@ "text" ], "cost": { - "input": 0.05, - "output": 0.2, - "cacheRead": 0, + "input": 0.03, + "output": 0.13, + "cacheRead": 0.03, "cacheWrite": 0 }, "contextWindow": 131072, @@ -14733,7 +14743,7 @@ }, "OpenPipe/Qwen3-14B-Instruct": { "id": "OpenPipe/Qwen3-14B-Instruct", - "name": "OpenPipe Qwen3 14B Instruct", + "name": "Qwen3 14B Instruct", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14744,7 +14754,7 @@ "cost": { "input": 0.05, "output": 0.22, - "cacheRead": 0, + "cacheRead": 0.05, "cacheWrite": 0 }, "contextWindow": 32768, @@ -14752,7 +14762,7 @@ }, "Qwen/Qwen3-235B-A22B-Instruct-2507": { "id": "Qwen/Qwen3-235B-A22B-Instruct-2507", - "name": "Qwen3 235B A22B Instruct 2507", + "name": "Qwen3 235B A22B-2507", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14763,7 +14773,7 @@ "cost": { "input": 0.1, "output": 0.1, - "cacheRead": 0, + "cacheRead": 0.1, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14771,7 +14781,7 @@ }, "Qwen/Qwen3-235B-A22B-Thinking-2507": { "id": "Qwen/Qwen3-235B-A22B-Thinking-2507", - "name": "Qwen3-235B-A22B-Thinking-2507", + "name": "Qwen3 235B A22B Thinking-2507", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14782,7 +14792,7 @@ "cost": { "input": 0.1, "output": 0.1, - "cacheRead": 0, + "cacheRead": 0.1, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14811,7 +14821,7 @@ "cost": { "input": 0.1, "output": 0.3, - "cacheRead": 0, + "cacheRead": 0.1, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14819,7 +14829,7 @@ }, "Qwen/Qwen3-Coder-480B-A35B-Instruct": { "id": "Qwen/Qwen3-Coder-480B-A35B-Instruct", - "name": "Qwen3-Coder-480B-A35B-Instruct", + "name": "Qwen3 Coder 480B A35B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14830,7 +14840,7 @@ "cost": { "input": 1, "output": 1.5, - "cacheRead": 0, + "cacheRead": 1, "cacheWrite": 0 }, "contextWindow": 262144, @@ -14838,7 +14848,7 @@ }, "Qwen/Qwen3.5-27B": { "id": "Qwen/Qwen3.5-27B", - "name": "Qwen3.5 27B", + "name": "Qwen3.5-27B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14848,13 +14858,13 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.39, + "output": 3.12, + "cacheRead": 0.08, "cacheWrite": 0 }, "contextWindow": 262144, - "maxTokens": 65536, + "maxTokens": 262144, "thinking": { "mode": "effort", "efforts": [ @@ -14867,7 +14877,7 @@ }, "Qwen/Qwen3.5-35B-A3B": { "id": "Qwen/Qwen3.5-35B-A3B", - "name": "Qwen3.5 35B-A3B", + "name": "Qwen3.5-35B-A3B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14877,13 +14887,13 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.25, + "output": 1.25, + "cacheRead": 0.25, "cacheWrite": 0 }, "contextWindow": 262144, - "maxTokens": 65536, + "maxTokens": 262144, "thinking": { "mode": "effort", "efforts": [ @@ -14906,13 +14916,13 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.6, + "output": 3.6, + "cacheRead": 0.12, "cacheWrite": 0 }, "contextWindow": 262144, - "maxTokens": 65536, + "maxTokens": 262144, "thinking": { "mode": "effort", "efforts": [ @@ -14925,7 +14935,7 @@ }, "Qwen/Qwen3.6-35B-A3B": { "id": "Qwen/Qwen3.6-35B-A3B", - "name": "Qwen3.6 35B-A3B", + "name": "Qwen3.6 35B A3B", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14935,13 +14945,13 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.25, + "output": 1.25, + "cacheRead": 0.25, "cacheWrite": 0 }, "contextWindow": 262144, - "maxTokens": 65536, + "maxTokens": 262144, "thinking": { "mode": "effort", "efforts": [ @@ -14973,7 +14983,7 @@ }, "zai-org/GLM-5.1": { "id": "zai-org/GLM-5.1", - "name": "GLM-5.1", + "name": "GLM 5.1", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -14987,8 +14997,8 @@ "cacheRead": 0.26, "cacheWrite": 0 }, - "contextWindow": 200000, - "maxTokens": 131072, + "contextWindow": 202752, + "maxTokens": 202752, "thinking": { "mode": "effort", "efforts": [ @@ -15002,7 +15012,7 @@ }, "zai-org/GLM-5.2": { "id": "zai-org/GLM-5.2", - "name": "GLM-5.2", + "name": "GLM 5.2", "api": "openai-completions", "provider": "coreweave", "baseUrl": "https://api.inference.wandb.ai/v1", @@ -15011,13 +15021,13 @@ "text" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 1.39, + "output": 4.4, + "cacheRead": 0.26, "cacheWrite": 0 }, "contextWindow": 262144, - "maxTokens": 164000, + "maxTokens": 262144, "thinking": { "mode": "effort", "efforts": [ @@ -16764,26 +16774,6 @@ }, "requestModelId": "MODEL_GOOGLE_GEMINI_3_0_FLASH_MINIMAL" }, - "glm-5-1": { - "id": "glm-5-1", - "name": "GLM-5.1", - "api": "devin-agent", - "provider": "devin", - "baseUrl": "https://server.codeium.com", - "reasoning": true, - "input": [ - "text" - ], - "supportsTools": true, - "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, - "cacheWrite": 0 - }, - "contextWindow": 200000, - "maxTokens": 64000 - }, "glm-5-2": { "id": "glm-5-2", "name": "GLM-5.2 High", @@ -17224,6 +17214,246 @@ }, "requestModelId": "gpt-5-5-none-priority" }, + "gpt-5-6-luna": { + "id": "gpt-5-6-luna", + "name": "GPT-5.6 Luna", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text", + "image" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1000000, + "maxTokens": 64000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortRouting": { + "off": "gpt-5-6-luna-none", + "minimal": "gpt-5-6-luna-low", + "low": "gpt-5-6-luna-medium", + "medium": "gpt-5-6-luna-high", + "high": "gpt-5-6-luna-xhigh", + "xhigh": "gpt-5-6-luna-max" + } + }, + "requestModelId": "gpt-5-6-luna-none" + }, + "gpt-5-6-luna-fast": { + "id": "gpt-5-6-luna-fast", + "name": "GPT-5.6 Luna Fast", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text", + "image" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1000000, + "maxTokens": 64000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortRouting": { + "off": "gpt-5-6-luna-none-priority", + "minimal": "gpt-5-6-luna-low-priority", + "low": "gpt-5-6-luna-low-priority", + "medium": "gpt-5-6-luna-medium-priority", + "high": "gpt-5-6-luna-high-priority", + "xhigh": "gpt-5-6-luna-xhigh-priority" + } + }, + "requestModelId": "gpt-5-6-luna-none-priority" + }, + "gpt-5-6-sol": { + "id": "gpt-5-6-sol", + "name": "GPT-5.6 Sol", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text", + "image" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1000000, + "maxTokens": 64000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortRouting": { + "off": "gpt-5-6-sol-none", + "minimal": "gpt-5-6-sol-low", + "low": "gpt-5-6-sol-medium", + "medium": "gpt-5-6-sol-high", + "high": "gpt-5-6-sol-xhigh", + "xhigh": "gpt-5-6-sol-max" + } + }, + "requestModelId": "gpt-5-6-sol-none" + }, + "gpt-5-6-sol-fast": { + "id": "gpt-5-6-sol-fast", + "name": "GPT-5.6 Sol Fast", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text", + "image" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1000000, + "maxTokens": 64000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortRouting": { + "off": "gpt-5-6-sol-none-priority", + "minimal": "gpt-5-6-sol-low-priority", + "low": "gpt-5-6-sol-low-priority", + "medium": "gpt-5-6-sol-medium-priority", + "high": "gpt-5-6-sol-high-priority", + "xhigh": "gpt-5-6-sol-xhigh-priority" + } + }, + "requestModelId": "gpt-5-6-sol-none-priority" + }, + "gpt-5-6-terra": { + "id": "gpt-5-6-terra", + "name": "GPT-5.6 Terra", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text", + "image" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1000000, + "maxTokens": 64000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortRouting": { + "off": "gpt-5-6-terra-none", + "minimal": "gpt-5-6-terra-low", + "low": "gpt-5-6-terra-medium", + "medium": "gpt-5-6-terra-high", + "high": "gpt-5-6-terra-xhigh", + "xhigh": "gpt-5-6-terra-max" + } + }, + "requestModelId": "gpt-5-6-terra-none" + }, + "gpt-5-6-terra-fast": { + "id": "gpt-5-6-terra-fast", + "name": "GPT-5.6 Terra Fast", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text", + "image" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1000000, + "maxTokens": 64000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortRouting": { + "off": "gpt-5-6-terra-none-priority", + "minimal": "gpt-5-6-terra-low-priority", + "low": "gpt-5-6-terra-low-priority", + "medium": "gpt-5-6-terra-medium-priority", + "high": "gpt-5-6-terra-high-priority", + "xhigh": "gpt-5-6-terra-xhigh-priority" + } + }, + "requestModelId": "gpt-5-6-terra-none-priority" + }, "kimi-k2-6": { "id": "kimi-k2-6", "name": "Kimi K2.6", @@ -17454,6 +17684,46 @@ }, "contextWindow": 200000, "maxTokens": 64000 + }, + "swe-1-7": { + "id": "swe-1-7", + "name": "SWE-1.7", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 262000, + "maxTokens": 64000 + }, + "swe-1-7-lightning": { + "id": "swe-1-7-lightning", + "name": "SWE-1.7 Lightning", + "api": "devin-agent", + "provider": "devin", + "baseUrl": "https://server.codeium.com", + "reasoning": true, + "input": [ + "text" + ], + "supportsTools": true, + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 202752, + "maxTokens": 64000 } }, "firepass": { @@ -23447,6 +23717,33 @@ ] } }, + "openai/gpt-oss-20b": { + "id": "openai/gpt-oss-20b", + "name": "GPT OSS 20B", + "api": "openai-completions", + "provider": "huggingface", + "baseUrl": "https://router.huggingface.co/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 0.1, + "output": 0.5, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 131072, + "maxTokens": 32768, + "thinking": { + "mode": "effort", + "efforts": [ + "low", + "medium", + "high" + ] + } + }, "Qwen/Qwen3-235B-A22B": { "id": "Qwen/Qwen3-235B-A22B", "name": "Qwen3 235B-A22B", @@ -24542,6 +24839,44 @@ "contextWindow": null, "maxTokens": null }, + "aion-labs/aion-3.0": { + "id": "aion-labs/aion-3.0", + "name": "Aion-3.0", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": null, + "maxTokens": null + }, + "aion-labs/aion-3.0-mini": { + "id": "aion-labs/aion-3.0-mini", + "name": "Aion-3.0-Mini", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": null, + "maxTokens": null + }, "aion-labs/aion-rp-llama-3.1-8b": { "id": "aion-labs/aion-rp-llama-3.1-8b", "name": "Aion-RP 1.0 (8B)", @@ -28532,7 +28867,7 @@ }, "minimax/minimax-m3": { "id": "minimax/minimax-m3", - "name": "MiniMax-M3", + "name": "MiniMax M3", "api": "openai-completions", "provider": "kilo", "baseUrl": "https://api.kilo.ai/api/gateway", @@ -28542,13 +28877,13 @@ "image" ], "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, + "input": 0.3, + "output": 1.2, + "cacheRead": 0.06, "cacheWrite": 0 }, - "contextWindow": 512000, - "maxTokens": 128000, + "contextWindow": 1048576, + "maxTokens": 512000, "thinking": { "mode": "effort", "efforts": [ @@ -29470,6 +29805,25 @@ "contextWindow": 131072, "maxTokens": 163840 }, + "nex-agi/nex-n2-mini": { + "id": "nex-agi/nex-n2-mini", + "name": "Nex-N2-Mini", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": 262144 + }, "nex-agi/nex-n2-pro": { "id": "nex-agi/nex-n2-pro", "name": "Nex-N2-Pro", @@ -29486,8 +29840,8 @@ "cacheRead": 0, "cacheWrite": 0 }, - "contextWindow": null, - "maxTokens": null + "contextWindow": 262144, + "maxTokens": 262144 }, "nex-agi/nex-n2-pro:free": { "id": "nex-agi/nex-n2-pro:free", @@ -29505,8 +29859,8 @@ "cacheRead": 0, "cacheWrite": 0 }, - "contextWindow": null, - "maxTokens": null + "contextWindow": 262144, + "maxTokens": 262144 }, "nousresearch/hermes-2-pro-llama-3-8b": { "id": "nousresearch/hermes-2-pro-llama-3-8b", @@ -31048,6 +31402,120 @@ }, "contextPromotionTarget": "kilo/openai/gpt-5.4" }, + "openai/gpt-5.6-luna": { + "id": "openai/gpt-5.6-luna", + "name": "GPT-5.6 Luna", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 372000, + "maxTokens": 128000 + }, + "openai/gpt-5.6-luna-pro": { + "id": "openai/gpt-5.6-luna-pro", + "name": "GPT-5.6 Luna Pro", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000 + }, + "openai/gpt-5.6-sol": { + "id": "openai/gpt-5.6-sol", + "name": "GPT-5.6 Sol", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 372000, + "maxTokens": 128000 + }, + "openai/gpt-5.6-sol-pro": { + "id": "openai/gpt-5.6-sol-pro", + "name": "GPT-5.6 Sol Pro", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000 + }, + "openai/gpt-5.6-terra": { + "id": "openai/gpt-5.6-terra", + "name": "GPT-5.6 Terra", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 372000, + "maxTokens": 128000 + }, + "openai/gpt-5.6-terra-pro": { + "id": "openai/gpt-5.6-terra-pro", + "name": "GPT-5.6 Terra Pro", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000 + }, "openai/gpt-audio": { "id": "openai/gpt-audio", "name": "GPT Audio", @@ -33859,6 +34327,25 @@ "contextWindow": null, "maxTokens": null }, + "tencent/hy3": { + "id": "tencent/hy3", + "name": "Hy3", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": null + }, "tencent/hy3-preview": { "id": "tencent/hy3-preview", "name": "Hy3 Preview", @@ -33907,6 +34394,25 @@ "contextWindow": 262144, "maxTokens": 64000 }, + "tencent/hy3:free": { + "id": "tencent/hy3:free", + "name": "Hy3 (free)", + "api": "openai-completions", + "provider": "kilo", + "baseUrl": "https://api.kilo.ai/api/gateway", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": null + }, "thedrummer/cydonia-24b-v4.1": { "id": "thedrummer/cydonia-24b-v4.1", "name": "Cydonia 24B V4.1", @@ -37363,6 +37869,44 @@ "contextWindow": null, "maxTokens": null }, + "aion-labs/aion-3.0": { + "id": "aion-labs/aion-3.0", + "name": "aion-labs/aion-3.0", + "api": "openai-completions", + "provider": "nanogpt", + "baseUrl": "https://nano-gpt.com/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": null, + "maxTokens": null + }, + "aion-labs/aion-3.0-mini": { + "id": "aion-labs/aion-3.0-mini", + "name": "aion-labs/aion-3.0-mini", + "api": "openai-completions", + "provider": "nanogpt", + "baseUrl": "https://nano-gpt.com/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": null, + "maxTokens": null + }, "aion-labs/aion-rp-llama-3.1-8b": { "id": "aion-labs/aion-rp-llama-3.1-8b", "name": "aion-labs/aion-rp-llama-3.1-8b", @@ -44952,6 +45496,25 @@ "contextWindow": null, "maxTokens": null }, + "mellum2-12b-a2-5b-instruct": { + "id": "mellum2-12b-a2-5b-instruct", + "name": "mellum2-12b-a2-5b-instruct", + "api": "openai-completions", + "provider": "nanogpt", + "baseUrl": "https://nano-gpt.com/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": null, + "maxTokens": null + }, "mercury-2": { "id": "mercury-2", "name": "Mercury 2", @@ -46617,6 +47180,25 @@ "contextWindow": 131072, "maxTokens": 163840 }, + "nex-agi/nex-n2-mini": { + "id": "nex-agi/nex-n2-mini", + "name": "nex-agi/nex-n2-mini", + "api": "openai-completions", + "provider": "nanogpt", + "baseUrl": "https://nano-gpt.com/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": 262144 + }, "nex-agi/nex-n2-pro": { "id": "nex-agi/nex-n2-pro", "name": "nex-agi/nex-n2-pro", @@ -46633,8 +47215,8 @@ "cacheRead": 0, "cacheWrite": 0 }, - "contextWindow": null, - "maxTokens": null + "contextWindow": 262144, + "maxTokens": 262144 }, "nothingiisreal/L3.1-70B-Celeste-V0.1-BF16": { "id": "nothingiisreal/L3.1-70B-Celeste-V0.1-BF16", @@ -48807,7 +49389,7 @@ }, "Qwen/Qwen3-235B-A22B-Instruct-2507": { "id": "Qwen/Qwen3-235B-A22B-Instruct-2507", - "name": "Qwen3 235B A22B Instruct 2507", + "name": "Qwen3 235B A22B-2507", "api": "openai-completions", "provider": "nanogpt", "baseUrl": "https://nano-gpt.com/api/v1", @@ -48864,7 +49446,7 @@ }, "Qwen/Qwen3-235B-A22B-Thinking-2507": { "id": "Qwen/Qwen3-235B-A22B-Thinking-2507", - "name": "Qwen3-235B-A22B-Thinking-2507", + "name": "Qwen3 235B A22B Thinking-2507", "api": "openai-completions", "provider": "nanogpt", "baseUrl": "https://nano-gpt.com/api/v1", @@ -49250,7 +49832,7 @@ }, "Qwen/Qwen3.6-35B-A3B": { "id": "Qwen/Qwen3.6-35B-A3B", - "name": "Qwen3.6 35B-A3B", + "name": "Qwen3.6 35B A3B", "api": "openai-completions", "provider": "nanogpt", "baseUrl": "https://nano-gpt.com/api/v1", @@ -52792,6 +53374,25 @@ ] } }, + "x-ai/grok-4.5": { + "id": "x-ai/grok-4.5", + "name": "x-ai/grok-4.5", + "api": "openai-completions", + "provider": "nanogpt", + "baseUrl": "https://nano-gpt.com/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 500000 + }, "x-ai/grok-build-0.1": { "id": "x-ai/grok-build-0.1", "name": "Grok Build 0.1", @@ -59495,6 +60096,120 @@ }, "contextPromotionTarget": "openai/gpt-5.4" }, + "gpt-5.6-luna": { + "id": "gpt-5.6-luna", + "name": "GPT-5.6 Luna", + "api": "openai-responses", + "provider": "openai", + "baseUrl": "https://api.openai.com/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 1, + "output": 6, + "cacheRead": 0.1, + "cacheWrite": 1.25 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "applyPatchToolType": "freeform", + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "gpt-5.6-sol": { + "id": "gpt-5.6-sol", + "name": "GPT-5.6 Sol", + "api": "openai-responses", + "provider": "openai", + "baseUrl": "https://api.openai.com/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 5, + "output": 30, + "cacheRead": 0.5, + "cacheWrite": 6.25 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "applyPatchToolType": "freeform", + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "gpt-5.6-terra": { + "id": "gpt-5.6-terra", + "name": "GPT-5.6 Terra", + "api": "openai-responses", + "provider": "openai", + "baseUrl": "https://api.openai.com/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2.5, + "output": 15, + "cacheRead": 0.25, + "cacheWrite": 3.125 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "applyPatchToolType": "freeform", + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, "o1": { "id": "o1", "name": "o1", @@ -60261,7 +60976,7 @@ "contextWindow": 272000, "maxTokens": 128000, "preferWebsockets": true, - "priority": 7, + "priority": 0, "applyPatchToolType": "freeform", "thinking": { "mode": "effort", @@ -60273,6 +60988,141 @@ ] }, "contextPromotionTarget": "openai-codex/gpt-5.4" + }, + "gpt-5.6-luna": { + "id": "gpt-5.6-luna", + "name": "GPT-5.6 Luna", + "api": "openai-codex-responses", + "provider": "openai-codex", + "baseUrl": "https://chatgpt.com/backend-api", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 1, + "output": 6, + "cacheRead": 0.1, + "cacheWrite": 1.25 + }, + "remoteCompaction": { + "enabled": true, + "api": "openai-codex-responses", + "v2StreamingEnabled": true + }, + "contextWindow": 372000, + "maxTokens": 128000, + "preferWebsockets": true, + "priority": 3, + "applyPatchToolType": "freeform", + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "gpt-5.6-sol": { + "id": "gpt-5.6-sol", + "name": "GPT-5.6 Sol", + "api": "openai-codex-responses", + "provider": "openai-codex", + "baseUrl": "https://chatgpt.com/backend-api", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 5, + "output": 30, + "cacheRead": 0.5, + "cacheWrite": 6.25 + }, + "remoteCompaction": { + "enabled": true, + "api": "openai-codex-responses", + "v2StreamingEnabled": true + }, + "contextWindow": 372000, + "maxTokens": 128000, + "preferWebsockets": true, + "priority": 1, + "applyPatchToolType": "freeform", + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "gpt-5.6-terra": { + "id": "gpt-5.6-terra", + "name": "GPT-5.6 Terra", + "api": "openai-codex-responses", + "provider": "openai-codex", + "baseUrl": "https://chatgpt.com/backend-api", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2.5, + "output": 15, + "cacheRead": 0.25, + "cacheWrite": 3.125 + }, + "remoteCompaction": { + "enabled": true, + "api": "openai-codex-responses", + "v2StreamingEnabled": true + }, + "contextWindow": 372000, + "maxTokens": 128000, + "preferWebsockets": true, + "priority": 2, + "applyPatchToolType": "freeform", + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } } }, "opencode": { @@ -62311,6 +63161,36 @@ }, "contextPromotionTarget": "opencode-zen/gpt-5.4" }, + "grok-4.5": { + "id": "grok-4.5", + "name": "Grok 4.5", + "api": "openai-completions", + "provider": "opencode-zen", + "baseUrl": "https://opencode.ai/zen/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2, + "output": 6, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 500000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, "grok-build-0.1": { "id": "grok-build-0.1", "name": "Grok Build 0.1", @@ -62341,6 +63221,35 @@ ] } }, + "hy3-free": { + "id": "hy3-free", + "name": "Hy3 Free", + "api": "openai-completions", + "provider": "opencode-zen", + "baseUrl": "https://opencode.ai/zen/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 256000, + "maxTokens": 64000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, "hy3-preview-free": { "id": "hy3-preview-free", "name": "Hy3 preview Free", @@ -63214,7 +64123,7 @@ "image" ], "cost": { - "input": 0.66, + "input": 0.65, "output": 3.41, "cacheRead": 0.14, "cacheWrite": 0 @@ -63289,6 +64198,35 @@ ] } }, + "~x-ai/grok-latest": { + "id": "~x-ai/grok-latest", + "name": "Grok Latest", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2, + "output": 6, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": null, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, "ai21/jamba-large-1.7": { "id": "ai21/jamba-large-1.7", "name": "Jamba Large 1.7", @@ -63308,6 +64246,90 @@ "contextWindow": 256000, "maxTokens": 4096 }, + "aion-labs/aion-2.0": { + "id": "aion-labs/aion-2.0", + "name": "Aion-2.0", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 0.7999999999999999, + "output": 1.5999999999999999, + "cacheRead": 0.19999999999999998, + "cacheWrite": 0 + }, + "contextWindow": 131072, + "maxTokens": 32768, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, + "aion-labs/aion-3.0": { + "id": "aion-labs/aion-3.0", + "name": "Aion-3.0", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 3, + "output": 6, + "cacheRead": 0.75, + "cacheWrite": 0 + }, + "contextWindow": 131072, + "maxTokens": 32768, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, + "aion-labs/aion-3.0-mini": { + "id": "aion-labs/aion-3.0-mini", + "name": "Aion-3.0-Mini", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 0.7, + "output": 1.4, + "cacheRead": 0.18, + "cacheWrite": 0 + }, + "contextWindow": 131072, + "maxTokens": 32768, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, "alibaba/tongyi-deepresearch-30b-a3b": { "id": "alibaba/tongyi-deepresearch-30b-a3b", "name": "Tongyi DeepResearch 30B A3B", @@ -64846,7 +65868,7 @@ "cacheWrite": 0 }, "contextWindow": 1048576, - "maxTokens": 16384, + "maxTokens": 65536, "thinking": { "mode": "effort", "efforts": [ @@ -66146,8 +67168,8 @@ "text" ], "cost": { - "input": 0.12, - "output": 0.48, + "input": 0.15, + "output": 0.8999999999999999, "cacheRead": 0.049999999999999996, "cacheWrite": 0 }, @@ -66205,8 +67227,8 @@ "text" ], "cost": { - "input": 0.18, - "output": 0.72, + "input": 0.24, + "output": 0.96, "cacheRead": 0.049999999999999996, "cacheWrite": 0 }, @@ -66224,7 +67246,7 @@ }, "minimax/minimax-m3": { "id": "minimax/minimax-m3", - "name": "MiniMax-M3", + "name": "MiniMax M3", "api": "openrouter", "provider": "openrouter", "baseUrl": "https://openrouter.ai/api/v1", @@ -66875,7 +67897,7 @@ "cost": { "input": 0.375, "output": 2.025, - "cacheRead": 0.09, + "cacheRead": 0.203, "cacheWrite": 0 }, "contextWindow": 262144, @@ -66902,7 +67924,7 @@ "image" ], "cost": { - "input": 0.66, + "input": 0.65, "output": 3.41, "cacheRead": 0.14, "cacheWrite": 0 @@ -66960,13 +67982,13 @@ "image" ], "cost": { - "input": 0.74, + "input": 0.72, "output": 3.5, "cacheRead": 0.15, "cacheWrite": 0 }, "contextWindow": 262144, - "maxTokens": 16384, + "maxTokens": 262144, "thinking": { "mode": "effort", "efforts": [ @@ -66996,6 +68018,64 @@ "contextWindow": 131072, "maxTokens": 163840 }, + "nex-agi/nex-n2-mini": { + "id": "nex-agi/nex-n2-mini", + "name": "Nex-N2-Mini", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 0.024999999999999998, + "output": 0.09999999999999999, + "cacheRead": 0.0025, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": 262144, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, + "nex-agi/nex-n2-pro": { + "id": "nex-agi/nex-n2-pro", + "name": "Nex-N2-Pro", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 0.25, + "output": 1, + "cacheRead": 0.024999999999999998, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": 262144, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, "nex-agi/nex-n2-pro:free": { "id": "nex-agi/nex-n2-pro:free", "name": "Nex-N2-Pro (free)", @@ -68452,6 +69532,228 @@ }, "contextPromotionTarget": "openrouter/openai/gpt-5.4" }, + "openai/gpt-5.6-luna": { + "id": "openai/gpt-5.6-luna", + "name": "GPT-5.6 Luna", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 1, + "output": 6, + "cacheRead": 0.09999999999999999, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "openai/gpt-5.6-luna-pro": { + "id": "openai/gpt-5.6-luna-pro", + "name": "GPT-5.6 Luna Pro", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 1, + "output": 6, + "cacheRead": 0.09999999999999999, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "openai/gpt-5.6-sol": { + "id": "openai/gpt-5.6-sol", + "name": "GPT-5.6 Sol", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 5, + "output": 30, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "openai/gpt-5.6-sol-pro": { + "id": "openai/gpt-5.6-sol-pro", + "name": "GPT-5.6 Sol Pro", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 5, + "output": 30, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "openai/gpt-5.6-terra": { + "id": "openai/gpt-5.6-terra", + "name": "GPT-5.6 Terra", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2.5, + "output": 15, + "cacheRead": 0.25, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, + "openai/gpt-5.6-terra-pro": { + "id": "openai/gpt-5.6-terra-pro", + "name": "GPT-5.6 Terra Pro", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2.5, + "output": 15, + "cacheRead": 0.25, + "cacheWrite": 0 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ], + "effortMap": { + "minimal": "low", + "low": "medium", + "medium": "high", + "high": "xhigh", + "xhigh": "max" + } + } + }, "openai/gpt-audio": { "id": "openai/gpt-audio", "name": "GPT Audio", @@ -68530,8 +69832,8 @@ "text" ], "cost": { - "input": 0.03, - "output": 0.15, + "input": 0.036, + "output": 0.18, "cacheRead": 0, "cacheWrite": 0 }, @@ -70301,7 +71603,7 @@ "cost": { "input": 0.385, "output": 2.4499999999999997, - "cacheRead": 0.195, + "cacheRead": 0.111, "cacheWrite": 0 }, "contextWindow": 256000, @@ -70938,6 +72240,34 @@ "supportsToolChoice": false } }, + "tencent/hy3": { + "id": "tencent/hy3", + "name": "Hy3", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 0.14, + "output": 0.58, + "cacheRead": 0.035, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": null, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, "tencent/hy3-preview": { "id": "tencent/hy3-preview", "name": "Hy3 Preview", @@ -70994,6 +72324,34 @@ ] } }, + "tencent/hy3:free": { + "id": "tencent/hy3:free", + "name": "Hy3 (free)", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 262144, + "maxTokens": 262144, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, "thedrummer/rocinante-12b": { "id": "thedrummer/rocinante-12b", "name": "Rocinante 12B", @@ -71418,6 +72776,35 @@ ] } }, + "x-ai/grok-4.5": { + "id": "x-ai/grok-4.5", + "name": "Grok 4.5", + "api": "openrouter", + "provider": "openrouter", + "baseUrl": "https://openrouter.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2, + "output": 6, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 500000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, "x-ai/grok-build-0.1": { "id": "x-ai/grok-build-0.1", "name": "Grok Build 0.1", @@ -71571,7 +72958,7 @@ "cost": { "input": 0.105, "output": 0.28, - "cacheRead": 0.0028, + "cacheRead": 0.028, "cacheWrite": 0 }, "contextWindow": 1048576, @@ -71971,13 +73358,13 @@ "text" ], "cost": { - "input": 0.9086, - "output": 2.8556, - "cacheRead": 0.16874, + "input": 0.54, + "output": 1.76, + "cacheRead": 0.09999999999999999, "cacheWrite": 0 }, "contextWindow": 1048576, - "maxTokens": 131072, + "maxTokens": 101376, "thinking": { "mode": "effort", "efforts": [ @@ -73401,38 +74788,6 @@ "escapeBuiltinToolNames": true } }, - "umans-glm-5.2-nvfp4": { - "id": "umans-glm-5.2-nvfp4", - "name": "Umans GLM 5.2 NVFP4 (experimental, short test from Jun 29)", - "api": "anthropic-messages", - "provider": "umans", - "baseUrl": "https://api.code.umans.ai", - "reasoning": true, - "thinking": { - "mode": "anthropic-budget-effort", - "efforts": [ - "high", - "xhigh" - ], - "effortMap": { - "xhigh": "max" - } - }, - "input": [ - "text" - ], - "cost": { - "input": 0, - "output": 0, - "cacheRead": 0, - "cacheWrite": 0 - }, - "contextWindow": 405504, - "maxTokens": 131071, - "compat": { - "escapeBuiltinToolNames": true - } - }, "umans-kimi-k2.7": { "id": "umans-kimi-k2.7", "name": "Umans Kimi K2.7 Code", @@ -73524,6 +74879,64 @@ "supportsUsageInStreaming": false } }, + "aion-labs-aion-3-0": { + "id": "aion-labs-aion-3-0", + "name": "Aion 3.0", + "api": "openai-completions", + "provider": "venice", + "baseUrl": "https://api.venice.ai/api/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 3.75, + "output": 7.5, + "cacheRead": 0.9375, + "cacheWrite": 0 + }, + "contextWindow": 128000, + "maxTokens": 32768, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, + "aion-labs-aion-3-0-mini": { + "id": "aion-labs-aion-3-0-mini", + "name": "Aion 3.0 Mini", + "api": "openai-completions", + "provider": "venice", + "baseUrl": "https://api.venice.ai/api/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 0.875, + "output": 1.75, + "cacheRead": 0.225, + "cacheWrite": 0 + }, + "contextWindow": 128000, + "maxTokens": 32768, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, "aion-labs.aion-2-0": { "id": "aion-labs.aion-2-0", "name": "aion-labs.aion-2-0", @@ -74080,6 +75493,28 @@ } } }, + "e2ee-deepseek-v4-flash": { + "id": "e2ee-deepseek-v4-flash", + "name": "e2ee-deepseek-v4-flash", + "api": "openai-completions", + "provider": "venice", + "baseUrl": "https://api.venice.ai/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 1000000, + "maxTokens": 1048576, + "compat": { + "supportsUsageInStreaming": false + } + }, "e2ee-gemma-3-27b-p": { "id": "e2ee-gemma-3-27b-p", "name": "e2ee-gemma-3-27b-p", @@ -74366,6 +75801,28 @@ "supportsUsageInStreaming": false } }, + "e2ee-qwen3-6-27b": { + "id": "e2ee-qwen3-6-27b", + "name": "e2ee-qwen3-6-27b", + "api": "openai-completions", + "provider": "venice", + "baseUrl": "https://api.venice.ai/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 256000, + "maxTokens": 65536, + "compat": { + "supportsUsageInStreaming": false + } + }, "e2ee-qwen3-6-35b-a3b": { "id": "e2ee-qwen3-6-35b-a3b", "name": "e2ee-qwen3-6-35b-a3b", @@ -74845,6 +76302,36 @@ ] } }, + "grok-4-5": { + "id": "grok-4-5", + "name": "Grok 4.5", + "api": "openai-completions", + "provider": "venice", + "baseUrl": "https://api.venice.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2.27, + "output": 6.8, + "cacheRead": 0.57, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 32000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, "grok-41-fast": { "id": "grok-41-fast", "name": "Grok 4.1 Fast", @@ -79099,7 +80586,7 @@ }, "minimax/minimax-m3": { "id": "minimax/minimax-m3", - "name": "MiniMax-M3", + "name": "MiniMax M3", "api": "anthropic-messages", "provider": "vercel-ai-gateway", "baseUrl": "https://ai-gateway.vercel.sh", @@ -80667,6 +82154,96 @@ }, "contextPromotionTarget": "vercel-ai-gateway/openai/gpt-5.4" }, + "openai/gpt-5.6-luna": { + "id": "openai/gpt-5.6-luna", + "name": "GPT 5.6 Luna", + "api": "anthropic-messages", + "provider": "vercel-ai-gateway", + "baseUrl": "https://ai-gateway.vercel.sh", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 1, + "output": 6, + "cacheRead": 0.09999999999999999, + "cacheWrite": 1.25 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "budget", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, + "openai/gpt-5.6-sol": { + "id": "openai/gpt-5.6-sol", + "name": "GPT 5.6 Sol", + "api": "anthropic-messages", + "provider": "vercel-ai-gateway", + "baseUrl": "https://ai-gateway.vercel.sh", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 5, + "output": 30, + "cacheRead": 0.5, + "cacheWrite": 6.25 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "budget", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, + "openai/gpt-5.6-terra": { + "id": "openai/gpt-5.6-terra", + "name": "GPT 5.6 Terra", + "api": "anthropic-messages", + "provider": "vercel-ai-gateway", + "baseUrl": "https://ai-gateway.vercel.sh", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2.5, + "output": 15, + "cacheRead": 0.25, + "cacheWrite": 3.125 + }, + "contextWindow": 1050000, + "maxTokens": 128000, + "thinking": { + "mode": "budget", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, "openai/gpt-oss-120b": { "id": "openai/gpt-oss-120b", "name": "GPT OSS 120B", @@ -81557,6 +83134,36 @@ ] } }, + "xai/grok-4.5": { + "id": "xai/grok-4.5", + "name": "Grok 4.5", + "api": "anthropic-messages", + "provider": "vercel-ai-gateway", + "baseUrl": "https://ai-gateway.vercel.sh", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2, + "output": 6, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 500000, + "thinking": { + "mode": "budget", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, "xai/grok-build-0.1": { "id": "xai/grok-build-0.1", "name": "Grok Build 0.1", @@ -82313,6 +83920,39 @@ "supportsDeveloperRole": false } }, + "GLM5.2-Turbo": { + "id": "GLM5.2-Turbo", + "name": "GLM5.2-Turbo", + "api": "openai-completions", + "provider": "wafer-serverless", + "baseUrl": "https://pass.wafer.ai/v1", + "reasoning": true, + "input": [ + "text" + ], + "cost": { + "input": 3.75, + "output": 12.8125, + "cacheRead": 0.625, + "cacheWrite": 0 + }, + "contextWindow": 1048576, + "maxTokens": 65536, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + }, + "compat": { + "reasoningContentField": "reasoning_content", + "supportsDeveloperRole": false + } + }, "Kimi-K2.6": { "id": "Kimi-K2.6", "name": "Kimi-K2.6", @@ -83094,6 +84734,35 @@ ] } }, + "grok-4.5": { + "id": "grok-4.5", + "name": "Grok 4.5", + "api": "openai-completions", + "provider": "xai", + "baseUrl": "https://api.x.ai/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2, + "output": 6, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 500000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high" + ] + } + }, "grok-beta": { "id": "grok-beta", "name": "Grok Beta", @@ -86308,6 +87977,25 @@ ] } }, + "kuaishou/kat-coder-air-v2.5": { + "id": "kuaishou/kat-coder-air-v2.5", + "name": "KAT-Coder-Air-V2.5", + "api": "openai-completions", + "provider": "zenmux", + "baseUrl": "https://zenmux.ai/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0.135, + "output": 0.54, + "cacheRead": 0.027, + "cacheWrite": 0 + }, + "contextWindow": 256000, + "maxTokens": null + }, "kuaishou/kat-coder-pro-v1": { "id": "kuaishou/kat-coder-pro-v1", "name": "KAT-Coder-Pro-V1", @@ -86365,6 +88053,25 @@ "contextWindow": 256000, "maxTokens": 80000 }, + "kuaishou/kat-coder-pro-v2.5": { + "id": "kuaishou/kat-coder-pro-v2.5", + "name": "KAT-Coder-Pro-V2.5", + "api": "openai-completions", + "provider": "zenmux", + "baseUrl": "https://zenmux.ai/api/v1", + "reasoning": false, + "input": [ + "text" + ], + "cost": { + "input": 0.444, + "output": 1.776, + "cacheRead": 0.09, + "cacheWrite": 0 + }, + "contextWindow": 256000, + "maxTokens": null + }, "meituan/longcat-2.0": { "id": "meituan/longcat-2.0", "name": "LongCat-2.0", @@ -88482,9 +90189,9 @@ "text" ], "cost": { - "input": 0.134561595, - "output": 0.539161765, - "cacheRead": 0.033869245, + "input": 0.1323, + "output": 0.5301, + "cacheRead": 0.0333, "cacheWrite": 0 }, "contextWindow": 262144, @@ -88938,6 +90645,66 @@ ] } }, + "x-ai/grok-4.5": { + "id": "x-ai/grok-4.5", + "name": "Grok 4.5", + "api": "openai-completions", + "provider": "zenmux", + "baseUrl": "https://zenmux.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 2, + "output": 6, + "cacheRead": 0.5, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 500000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, + "x-ai/grok-4.5-free": { + "id": "x-ai/grok-4.5-free", + "name": "Grok 4.5 (Free)", + "api": "openai-completions", + "provider": "zenmux", + "baseUrl": "https://zenmux.ai/api/v1", + "reasoning": true, + "input": [ + "text", + "image" + ], + "cost": { + "input": 0, + "output": 0, + "cacheRead": 0, + "cacheWrite": 0 + }, + "contextWindow": 500000, + "maxTokens": 500000, + "thinking": { + "mode": "effort", + "efforts": [ + "minimal", + "low", + "medium", + "high", + "xhigh" + ] + } + }, "x-ai/grok-build-0.1": { "id": "x-ai/grok-build-0.1", "name": "Grok Build 0.1", diff --git a/packages/catalog/src/variant-collapse.ts b/packages/catalog/src/variant-collapse.ts index b3042503d..46d1f9d52 100644 --- a/packages/catalog/src/variant-collapse.ts +++ b/packages/catalog/src/variant-collapse.ts @@ -113,6 +113,14 @@ function thinkingPair(baseId: string, name: string): EffortVariantFamily { type DevinTierRoutes = Partial>; +const DEVIN_FIVE_TIER_EFFORTS: readonly Effort[] = [ + Effort.Minimal, + Effort.Low, + Effort.Medium, + Effort.High, + Effort.XHigh, +]; + function devinTierFamily( id: string, name: string, @@ -160,6 +168,44 @@ function devinTierFamily( }; } +/** + * GPT-5.6 (Luna/Sol/Terra) adds a genuine `max` tier above `xhigh`, so the + * standard family shifts every user effort up one notch (`minimal` → `-low` + * … `xhigh` → `-max`), mirroring the Opus 4.7+ five-tier mapping. Devin + * serves no `-max-priority` sibling, so the fast family keeps the direct + * `low..xhigh` `-priority` scale. + */ +function devinGpt56Families(variant: "luna" | "sol" | "terra", name: string): readonly EffortVariantFamily[] { + const base = `gpt-5-6-${variant}`; + return [ + devinTierFamily( + base, + name, + { + off: `${base}-none`, + minimal: `${base}-low`, + low: `${base}-medium`, + medium: `${base}-high`, + high: `${base}-xhigh`, + xhigh: `${base}-max`, + }, + DEVIN_FIVE_TIER_EFFORTS, + ), + devinTierFamily( + `${base}-fast`, + `${name} Fast`, + { + off: `${base}-none-priority`, + low: `${base}-low-priority`, + medium: `${base}-medium-priority`, + high: `${base}-high-priority`, + xhigh: `${base}-xhigh-priority`, + }, + DEVIN_FIVE_TIER_EFFORTS, + ), + ]; +} + const GEMINI_3_FLASH_FAMILY_EFFORTS: readonly Effort[] = [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High]; const GEMINI_3_PRO_FAMILY_EFFORTS: readonly Effort[] = [Effort.Low, Effort.High]; @@ -330,7 +376,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { }, thinking: { mode: "effort", - efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + efforts: DEVIN_FIVE_TIER_EFFORTS, requiresEffort: true, }, }, @@ -353,7 +399,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { }, thinking: { mode: "effort", - efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + efforts: DEVIN_FIVE_TIER_EFFORTS, requiresEffort: true, }, }, @@ -376,7 +422,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { }, thinking: { mode: "effort", - efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + efforts: DEVIN_FIVE_TIER_EFFORTS, requiresEffort: true, }, }, @@ -399,7 +445,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { }, thinking: { mode: "effort", - efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + efforts: DEVIN_FIVE_TIER_EFFORTS, requiresEffort: true, }, }, @@ -413,7 +459,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "MODEL_GPT_5_2_HIGH", xhigh: "MODEL_GPT_5_2_XHIGH", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), devinTierFamily( "gpt-5-3-codex", @@ -424,7 +470,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "gpt-5-3-codex-high", xhigh: "gpt-5-3-codex-xhigh", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), devinTierFamily( "gpt-5-3-codex-fast", @@ -435,7 +481,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "gpt-5-3-codex-high-priority", xhigh: "gpt-5-3-codex-xhigh-priority", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), devinTierFamily( "gpt-5-4", @@ -447,7 +493,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "gpt-5-4-high", xhigh: "gpt-5-4-xhigh", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), devinTierFamily( "gpt-5-4-fast", @@ -459,7 +505,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "gpt-5-4-high-priority", xhigh: "gpt-5-4-xhigh-priority", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), devinTierFamily( "gpt-5-4-mini", @@ -470,7 +516,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "gpt-5-4-mini-high", xhigh: "gpt-5-4-mini-xhigh", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), devinTierFamily( "gpt-5-5", @@ -482,7 +528,7 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "gpt-5-5-high", xhigh: "gpt-5-5-xhigh", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), devinTierFamily( "gpt-5-5-fast", @@ -494,8 +540,11 @@ export const DEVIN_VARIANT_COLLAPSE_TABLE: VariantCollapseTable = { high: "gpt-5-5-high-priority", xhigh: "gpt-5-5-xhigh-priority", }, - [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + DEVIN_FIVE_TIER_EFFORTS, ), + ...devinGpt56Families("luna", "GPT-5.6 Luna"), + ...devinGpt56Families("sol", "GPT-5.6 Sol"), + ...devinGpt56Families("terra", "GPT-5.6 Terra"), devinTierFamily( "gemini-3-1-pro", "Gemini 3.1 Pro", diff --git a/packages/catalog/test/cursor-discovery.test.ts b/packages/catalog/test/cursor-discovery.test.ts new file mode 100644 index 000000000..f62b7c37c --- /dev/null +++ b/packages/catalog/test/cursor-discovery.test.ts @@ -0,0 +1,96 @@ +import { afterAll, beforeAll, describe, expect, it } from "bun:test"; +import * as http2 from "node:http2"; +import { create, toBinary } from "@bufbuild/protobuf"; +// Import from source, not the package specifier: the workspace `node_modules` +// copy resolves to the primary checkout, not this worktree. +import { fetchCursorUsableModels } from "../src/discovery/cursor"; +import { GetUsableModelsResponseSchema, ModelDetailsSchema } from "../src/discovery/cursor-gen/agent_pb"; +import type { ModelSpec } from "../src/types"; + +const FIXTURE_MODEL_IDS = [ + // Reference-less ids from families whose native catalogs are multimodal. + "claude-opus-4-8-99999999", + "gpt-5.5-codex-20991231", + "gemini-4-pro-exp", + // Reference-less ids from text-only families. + "composer-3", + "grok-code-fast-2", + // Bundled-reference ids: the reference stays authoritative. + "claude-4.5-opus-high", + "claude-4.6-opus-high", + "composer-1", +]; + +let server: http2.Http2Server; +let baseUrl: string; + +beforeAll(async () => { + const response = create(GetUsableModelsResponseSchema, { + models: FIXTURE_MODEL_IDS.map(modelId => create(ModelDetailsSchema, { modelId })), + }); + const payload = Buffer.from(toBinary(GetUsableModelsResponseSchema, response)); + + server = http2.createServer(); + server.on("stream", (stream: http2.ServerHttp2Stream, headers: http2.IncomingHttpHeaders) => { + stream.on("data", () => {}); + stream.on("end", () => { + if (headers[":path"] !== "/agent.v1.AgentService/GetUsableModels") { + stream.respond({ ":status": 404 }); + stream.end(); + return; + } + stream.respond({ ":status": 200, "content-type": "application/proto" }); + stream.end(payload); + }); + }); + await new Promise(resolve => server.listen(0, "127.0.0.1", resolve)); + const address = server.address(); + if (!address || typeof address === "string") { + throw new Error("expected http2 fixture server to bind a tcp port"); + } + baseUrl = `http://127.0.0.1:${address.port}`; +}); + +afterAll(() => { + server?.close(); +}); + +async function discover(): Promise>> { + const models = await fetchCursorUsableModels({ apiKey: "test-key", baseUrl }); + expect(models).not.toBeNull(); + return new Map((models ?? []).map(model => [model.id, model])); +} + +describe("cursor discovery input modalities (issue #4726)", () => { + it("classifies reference-less multimodal-family models as text+image", async () => { + const byId = await discover(); + expect(byId.get("claude-opus-4-8-99999999")?.input).toEqual(["text", "image"]); + expect(byId.get("gpt-5.5-codex-20991231")?.input).toEqual(["text", "image"]); + expect(byId.get("gemini-4-pro-exp")?.input).toEqual(["text", "image"]); + }); + + it("keeps reference-less text-only families text-only", async () => { + const byId = await discover(); + expect(byId.get("composer-3")?.input).toEqual(["text"]); + expect(byId.get("grok-code-fast-2")?.input).toEqual(["text"]); + }); + + it("keeps bundled references authoritative for input modalities", async () => { + const byId = await discover(); + // Bundled cursor references carry their own input classification; the + // id-based inference must not override it in either direction. + expect(byId.get("claude-4.5-opus-high")?.input).toEqual(["text", "image"]); + expect(byId.get("claude-4.6-opus-high")?.input).toEqual(["text"]); + expect(byId.get("composer-1")?.input).toEqual(["text"]); + }); + + it("preserves fallback defaults for reference-less models", async () => { + const byId = await discover(); + const spec = byId.get("claude-opus-4-8-99999999"); + expect(spec?.provider).toBe("cursor"); + expect(spec?.api).toBe("cursor-agent"); + expect(spec?.contextWindow).toBe(200_000); + expect(spec?.maxTokens).toBe(64_000); + expect(spec?.cost).toEqual({ input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }); + }); +}); diff --git a/packages/catalog/test/model-thinking.test.ts b/packages/catalog/test/model-thinking.test.ts index 6d6019fe1..a21a323fc 100644 --- a/packages/catalog/test/model-thinking.test.ts +++ b/packages/catalog/test/model-thinking.test.ts @@ -567,6 +567,90 @@ describe("model thinking derivation", () => { expect(getSupportedEfforts(model)).toEqual([]); expect(clampThinkingLevelForModel(model, Effort.High)).toBeUndefined(); }); + + it("bakes the GPT-5.6 shifted five-tier effort map on wire-effort APIs", () => { + const codex = createModel({ + id: "gpt-5.6-sol", + api: "openai-codex-responses", + provider: "openai-codex", + }); + + expect(codex.thinking).toEqual({ + mode: "effort", + efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + effortMap: { + minimal: "low", + low: "medium", + medium: "high", + high: "xhigh", + xhigh: "max", + }, + }); + + // Stale baked four-tier metadata (caches/discovery) normalizes back to + // the five-tier ladder with the map attached — the wire-defaults + // backfill path — and namespaced OpenRouter ids parse. + const staleOpenRouter = createModel({ + id: "openai/gpt-5.6-terra", + api: "openrouter", + provider: "openrouter", + baseUrl: "https://openrouter.ai/api/v1", + thinking: { + mode: "effort", + efforts: [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + }, + }); + + expect(staleOpenRouter.thinking).toEqual({ + mode: "effort", + efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + effortMap: { + minimal: "low", + low: "medium", + medium: "high", + high: "xhigh", + xhigh: "max", + }, + }); + }); + + it("keeps pre-5.6 and Devin-routed GPT models off the shifted effort map", () => { + const gpt55 = createModel({ + id: "gpt-5.5", + api: "openai-responses", + provider: "openai", + }); + + expect(gpt55.thinking).toEqual({ + mode: "effort", + efforts: [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + }); + expect(gpt55.thinking?.effortMap).toBeUndefined(); + + // Devin selects effort by routing to per-tier sibling model ids, never + // via a wire reasoning.effort field — the shifted map must not attach. + const devin = createModel({ + id: "gpt-5-6-sol", + api: "devin-agent", + provider: "devin", + baseUrl: "https://server.codeium.com", + thinking: { + mode: "effort", + efforts: [Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh], + effortRouting: { + off: "gpt-5-6-sol-none", + minimal: "gpt-5-6-sol-low", + low: "gpt-5-6-sol-medium", + medium: "gpt-5-6-sol-high", + high: "gpt-5-6-sol-xhigh", + xhigh: "gpt-5-6-sol-max", + }, + }, + }); + + expect(devin.thinking?.effortMap).toBeUndefined(); + expect(devin.thinking?.efforts).toEqual([Effort.Minimal, Effort.Low, Effort.Medium, Effort.High, Effort.XHigh]); + }); }); describe("model thinking runtime helpers", () => { diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md index d0c83a8e9..cc19ab747 100644 --- a/packages/coding-agent/CHANGELOG.md +++ b/packages/coding-agent/CHANGELOG.md @@ -2,6 +2,31 @@ ## [Unreleased] +## [16.3.14] - 2026-07-09 + +### Fixed + +- Fixed issue where unfinalized tool blocks could incorrectly pin the live-region scroll seam +- Improved rendering of raw thinking blocks by stripping empty HTML comment noise +- Fixed display of thinking blocks consisting entirely of hidden comment noise +- Fixed gpt-5.6 reasoning summaries rendering literal `` sentinel lines in thinking blocks; empty HTML comments (and the unterminated ``), streamed as a ``. Comments with actual content are left untouched. +const EMPTY_COMMENT_RE = /^$/; +const OPEN_COMMENT_RE = /^` + * sentinel lines outside code fences (see {@link isCommentNoise}); prose-only + * mode additionally elides fenced code down to a trailing ellipsis. + */ export function formatThinkingForDisplay(text: string, proseOnly: boolean): string { - if (!proseOnly || !text) return text; - if (text === formatCacheKey) return formatCacheValue; + if (!text) return text; + const hasComment = text.includes("` leaves no blank tail. + if (hasComment && isCommentNoise(line, i === lines.length - 1)) continue; + + const open = FENCE.exec(line); + if (open) { const marker = open[2]!; const ch = marker[0]!; // A backtick fence's info string may not contain a backtick. @@ -79,18 +116,25 @@ export function formatThinkingForDisplay(text: string, proseOnly: boolean): stri inFence = true; fenceChar = ch; fenceLen = marker.length; - appendEllipsis(); - } else { - resultLines.push(line); + if (proseOnly) { + appendEllipsis(); + } else { + resultLines.push(line); + } + continue; } - } else { - resultLines.push(line); } + resultLines.push(line); } const formatted = resultLines.join("\n"); - formatCacheKey = text; - formatCacheValue = formatted; + if (proseOnly) { + proseCacheKey = text; + proseCacheValue = formatted; + } else { + rawCacheKey = text; + rawCacheValue = formatted; + } return formatted; } @@ -99,9 +143,11 @@ export function hasDisplayableThinking( text: string | null | undefined, formattedText: string | null | undefined, ): boolean { - if (!text) return false; - if (!formattedText) return false; - return formattedText.length > 0 && canonicalizeMessage(text).length > 0; + if (!text || !formattedText) return false; + // Visibility keys off the formatted text: a block whose raw text is only + // comment noise (`\n`) formats to whitespace and stays hidden. The + // raw canonicalize check still hides dot/ellipsis-only placeholder blocks. + return formattedText.trim().length > 0 && canonicalizeMessage(text).length > 0; } /** Whether an assistant message contains thinking content the TUI can reveal. */ diff --git a/packages/coding-agent/test/acp-agent.test.ts b/packages/coding-agent/test/acp-agent.test.ts index 556e32813..7e9554119 100644 --- a/packages/coding-agent/test/acp-agent.test.ts +++ b/packages/coding-agent/test/acp-agent.test.ts @@ -124,6 +124,7 @@ class FakeAgentSession { } promptCalls: string[] = []; customMessages: Array<{ customType: string; content: string; details?: unknown }> = []; + customMessageOptions: Array<{ streamingBehavior?: "steer" | "followUp"; queueChipText?: string } | undefined> = []; skillsSettings = { enableSkillCommands: true }; skills: Array<{ name: string; description: string; filePath: string; baseDir: string; source: string }> = []; planModeState: PlanModeState | undefined; @@ -235,8 +236,12 @@ class FakeAgentSession { this.isStreaming = false; } - async promptCustomMessage(message: { customType: string; content: string; details?: unknown }): Promise { + async promptCustomMessage( + message: { customType: string; content: string; details?: unknown }, + options?: { streamingBehavior?: "steer" | "followUp"; queueChipText?: string }, + ): Promise { this.customMessages.push(message); + this.customMessageOptions.push(options); this.isStreaming = true; const assistantMessage = makeAssistantMessage("skill pong"); for (const listener of this.#listeners) { @@ -1013,6 +1018,115 @@ describe("ACP agent", () => { await Bun.sleep(0); }); + it("delivers the final visible answer when agent_end overtakes the assistant message_end (#4902)", async () => { + const harness = await createHarness(); + const created = await harness.agent.newSession({ cwd: harness.cwdA, mcpServers: [] }); + const session = harness.findSession(created.sessionId); + if (!session) throw new Error("session not registered"); + + // Live turn as observed through the prompt subscription when the + // fire-and-forget assistant message_end handler loses the race against + // the agent_end flush: thinking streams, then the turn ends. No + // text_delta and no message_end ever reach this subscriber — the final + // text exists only on the agent_end payload. + const assistantMessage = makeAssistantMessage("Final visible answer.", "Considering the greeting."); + session.prompt = async (text: string): Promise => { + session.promptCalls.push(text); + session.isStreaming = true; + for (const listener of session.listeners()) { + listener({ + type: "message_update", + message: assistantMessage, + assistantMessageEvent: { type: "thinking_delta", delta: "Considering the greeting." }, + } as AgentSessionEvent); + } + session.sessionManager.appendMessage(assistantMessage); + for (const listener of session.listeners()) { + listener({ type: "agent_end", messages: [assistantMessage] } as AgentSessionEvent); + } + session.isStreaming = false; + return true; + }; + + const response = await harness.agent.prompt({ + sessionId: created.sessionId, + prompt: [{ type: "text", text: "Say hello" }], + }); + expectAcpStructure(zPromptResponse, response); + expect(response.stopReason).toBe("end_turn"); + + const chunks = harness.updates.filter(update => update.sessionId === created.sessionId); + const thoughtChunks = chunks.filter(update => update.update.sessionUpdate === "agent_thought_chunk"); + const messageChunks = chunks.filter(update => update.update.sessionUpdate === "agent_message_chunk"); + expect(thoughtChunks).toHaveLength(1); + // The visible answer must reach the client exactly once even though the + // assistant message_end never arrived on this subscription. + expect(messageChunks).toHaveLength(1); + expect(messageChunks[0]?.update).toEqual( + expect.objectContaining({ + sessionUpdate: "agent_message_chunk", + content: { type: "text", text: "Final visible answer." }, + }), + ); + // Flushed answer belongs to the same live message as the thought chunk. + expect(getChunkMessageId(messageChunks[0]!)).toBe(getChunkMessageId(thoughtChunks[0]!)!); + expectAcpNotifications(harness.updates); + + harness.abortController.abort(); + await Bun.sleep(0); + }); + + it("does not duplicate the final answer when the assistant message_end arrives before agent_end", async () => { + // Companion to the #4902 regression: when message_end IS delivered, its + // fallback emission wins and the agent_end flush must stay silent. + const harness = await createHarness(); + const created = await harness.agent.newSession({ cwd: harness.cwdA, mcpServers: [] }); + const session = harness.findSession(created.sessionId); + if (!session) throw new Error("session not registered"); + + const assistantMessage = makeAssistantMessage("Composed offline.", "quiet planning"); + session.prompt = async (text: string): Promise => { + session.promptCalls.push(text); + session.isStreaming = true; + for (const listener of session.listeners()) { + listener({ + type: "message_update", + message: assistantMessage, + assistantMessageEvent: { type: "thinking_delta", delta: "quiet planning" }, + } as AgentSessionEvent); + } + for (const listener of session.listeners()) { + listener({ type: "message_end", message: assistantMessage } as AgentSessionEvent); + } + session.sessionManager.appendMessage(assistantMessage); + for (const listener of session.listeners()) { + listener({ type: "agent_end", messages: [assistantMessage] } as AgentSessionEvent); + } + session.isStreaming = false; + return true; + }; + + const response = await harness.agent.prompt({ + sessionId: created.sessionId, + prompt: [{ type: "text", text: "Say hello" }], + }); + expectAcpStructure(zPromptResponse, response); + + const messageChunks = harness.updates.filter( + update => update.sessionId === created.sessionId && update.update.sessionUpdate === "agent_message_chunk", + ); + expect(messageChunks).toHaveLength(1); + expect(messageChunks[0]?.update).toEqual( + expect.objectContaining({ + content: { type: "text", text: "Composed offline." }, + }), + ); + expectAcpNotifications(harness.updates); + + harness.abortController.abort(); + await Bun.sleep(0); + }); + it("replays assistant tool calls and matching results without duplicating the start", async () => { const harness = await createHarness(); const stored = new FakeAgentSession(harness.cwdA); @@ -1443,6 +1557,7 @@ describe("ACP agent", () => { expect(customMessage.content).toContain(`[Skill directory: ${skillDir}]`); expect(customMessage.content).toMatch(/[Rr]esolve any relative paths/); expect(customMessage.content).toContain("User: extra context"); + expect(session.customMessageOptions[0]).toEqual({ streamingBehavior: "steer" }); harness.abortController.abort(); await Bun.sleep(0); diff --git a/packages/coding-agent/test/agent-session-concurrent.test.ts b/packages/coding-agent/test/agent-session-concurrent.test.ts index 7b4b045d6..5d01d6b80 100644 --- a/packages/coding-agent/test/agent-session-concurrent.test.ts +++ b/packages/coding-agent/test/agent-session-concurrent.test.ts @@ -15,7 +15,7 @@ import { getBundledModel } from "@oh-my-pi/pi-catalog/models"; import { AsyncJobManager } from "@oh-my-pi/pi-coding-agent/async"; import type { Rule } from "@oh-my-pi/pi-coding-agent/capability/rule"; import { ModelRegistry } from "@oh-my-pi/pi-coding-agent/config/model-registry"; -import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings"; +import { type SettingPath, Settings } from "@oh-my-pi/pi-coding-agent/config/settings"; import { TtsrManager } from "@oh-my-pi/pi-coding-agent/export/ttsr"; import type { ExtensionRunner } from "@oh-my-pi/pi-coding-agent/extensibility/extensions"; import { AgentSession } from "@oh-my-pi/pi-coding-agent/session/agent-session"; @@ -68,7 +68,7 @@ describe("AgentSession concurrent prompt guard", () => { AsyncJobManager.resetForTests(); }); - async function createSession() { + async function createSession(settingsOverrides?: Partial>) { const model = getBundledModel("anthropic", "claude-sonnet-4-5")!; let abortSignal: AbortSignal | undefined; @@ -100,7 +100,7 @@ describe("AgentSession concurrent prompt guard", () => { }); const sessionManager = SessionManager.inMemory(); - const settings = Settings.isolated(); + const settings = Settings.isolated(settingsOverrides); const authStorage = await AuthStorage.create(path.join(tempDir, "testauth.db")); authStorages.push(authStorage); const modelRegistry = new ModelRegistry(authStorage, path.join(tempDir, "models.yml")); @@ -176,6 +176,86 @@ describe("AgentSession concurrent prompt guard", () => { await firstPrompt.catch(() => {}); }); + it("queues sendUserMessage as steer while streaming without AgentBusyError", async () => { + await createSession(); + + const firstPrompt = session.prompt("First message"); + await waitFor(() => session.isStreaming); + + // The first agent loop may dequeue a steer before the assertion runs, so + // observe agent.steer itself rather than the residual queue length. + const steered: AgentMessage[] = []; + const originalSteer = session.agent.steer.bind(session.agent); + session.agent.steer = (message: AgentMessage) => { + steered.push(message); + originalSteer(message); + }; + + // Extension path: no deliverAs while busy must queue, not throw. + await expect(session.sendUserMessage("hello from extension")).resolves.toBeUndefined(); + expect(steered).toHaveLength(1); + const queued = steered[0]; + expect(queued?.role).toBe("user"); + if (queued?.role === "user") { + expect(queued.content).toEqual([{ type: "text", text: "hello from extension" }]); + expect(queued.steering).toBe(true); + } + + session.agent.clearAllQueues(); + await session.abort(); + await firstPrompt.catch(() => {}); + }); + + it("sendUserMessage without deliverAs preserves prompt-flow keyword notices while streaming", async () => { + await createSession({ "magicKeywords.enabled": true, "magicKeywords.ultrathink": true }); + + const firstPrompt = session.prompt("First message"); + await waitFor(() => session.isStreaming); + + try { + await session.sendUserMessage("ultrathink fix via extension"); + const queuedShape = session.agent + .peekSteeringQueue() + .map(message => (message.role === "custom" ? message.customType : message.role)); + expect(queuedShape).toEqual(["ultrathink-notice", "user"]); + expect(session.getQueuedMessages()).toEqual({ + steering: ["ultrathink fix via extension"], + followUp: [], + }); + } finally { + session.agent.clearAllQueues(); + await session.abort(); + await firstPrompt.catch(() => {}); + } + }); + + it("sendUserMessage without deliverAs starts a normal prompt when idle", async () => { + await createSession(); + + let rejected: unknown; + let settled = false; + const turn = session + .sendUserMessage("Idle extension message") + .catch(error => { + rejected = error; + }) + .finally(() => { + settled = true; + }); + + try { + await waitFor(() => session.isStreaming || settled); + if (rejected) throw rejected; + + expect(session.isStreaming).toBe(true); + expect(settled).toBe(false); + expect(session.getQueuedMessages()).toEqual({ steering: [], followUp: [] }); + } finally { + await session.abort(); + await turn; + } + }); + it("delivers hidden nextTurn stop reactions through the next LLM call without exposing them in the visible queue", async () => { const model = getBundledModel("anthropic", "claude-sonnet-4-5")!; let firstStream: AssistantMessageEventStream | undefined; diff --git a/packages/coding-agent/test/bundled-agent-parsing.test.ts b/packages/coding-agent/test/bundled-agent-parsing.test.ts new file mode 100644 index 000000000..a4da8e57d --- /dev/null +++ b/packages/coding-agent/test/bundled-agent-parsing.test.ts @@ -0,0 +1,63 @@ +import { describe, expect, it } from "bun:test"; +import { Effort } from "@oh-my-pi/pi-ai"; +import { buildModel } from "@oh-my-pi/pi-catalog/build"; +import { resolveAgentModelPatterns, resolveModelOverride } from "@oh-my-pi/pi-coding-agent/config/model-resolver"; +import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings"; +import { getBundledAgent } from "@oh-my-pi/pi-coding-agent/task/agents"; + +describe("bundled agent parsing", () => { + it("lets reviewer inherit thinking effort from its model role", () => { + const reviewer = getBundledAgent("reviewer"); + + expect(reviewer).toBeDefined(); + expect(reviewer?.source).toBe("bundled"); + expect(reviewer?.model).toEqual(["pi/slow"]); + expect(reviewer?.thinkingLevel).toBeUndefined(); + }); + + it("lets plan inherit thinking effort from its model role", () => { + const plan = getBundledAgent("plan"); + + expect(plan).toBeDefined(); + expect(plan?.source).toBe("bundled"); + expect(plan?.model).toEqual(["pi/plan", "pi/slow"]); + expect(plan?.thinkingLevel).toBeUndefined(); + }); + + // Issue #4761: with `modelRoles.slow: ...:xhigh`, the role's explicit effort + // suffix must survive agent-pattern expansion and model resolution for the + // bundled agents routed at that role. The executor picks + // `agent.thinkingLevel ?? resolvedThinkingLevel` (task/executor.ts), so a + // bundled frontmatter pin would mask the suffix — reviewer/plan declare none + // (asserted above) and the resolved level below is what the subagent runs at. + it("resolves the configured slow-role effort suffix for reviewer and plan", () => { + const gpt55 = buildModel({ + id: "gpt-5.5", + name: "GPT-5.5 Codex", + api: "openai-codex-responses", + provider: "openai-codex", + baseUrl: "https://chatgpt.com/backend-api/codex", + reasoning: true, + thinking: { mode: "effort", efforts: [Effort.Low, Effort.Medium, Effort.High, Effort.XHigh] }, + input: ["text"], + cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 }, + contextWindow: 272000, + maxTokens: 128000, + }); + const settings = Settings.isolated({ + modelRoles: { slow: "openai-codex/gpt-5.5:xhigh", plan: "openai-codex/gpt-5.5:xhigh" }, + }); + const registry = { getAvailable: () => [gpt55] } as Parameters[1]; + + for (const name of ["reviewer", "plan"]) { + const agent = getBundledAgent(name); + expect(agent?.thinkingLevel).toBeUndefined(); + const patterns = resolveAgentModelPatterns({ agentModel: agent?.model, settings }); + const resolved = resolveModelOverride(patterns, registry, settings); + expect(resolved.model?.provider).toBe("openai-codex"); + expect(resolved.model?.id).toBe("gpt-5.5"); + expect(resolved.thinkingLevel).toBe(Effort.XHigh); + expect(resolved.explicitThinkingLevel).toBe(true); + } + }); +}); diff --git a/packages/coding-agent/test/extensions-runner.test.ts b/packages/coding-agent/test/extensions-runner.test.ts index d24518737..0a2680515 100644 --- a/packages/coding-agent/test/extensions-runner.test.ts +++ b/packages/coding-agent/test/extensions-runner.test.ts @@ -1169,6 +1169,7 @@ describe("ExtensionRunner", () => { setEditorText: () => {}, getEditorText: () => "", editor: async () => undefined, + addAutocompleteProvider: () => {}, setEditorComponent: () => {}, get theme() { return {} as never; diff --git a/packages/coding-agent/test/input-controller-escape.test.ts b/packages/coding-agent/test/input-controller-escape.test.ts index e7ae3af47..d2828106d 100644 --- a/packages/coding-agent/test/input-controller-escape.test.ts +++ b/packages/coding-agent/test/input-controller-escape.test.ts @@ -269,6 +269,16 @@ function abortViewSession(ctx: InteractiveModeContext): AbortViewSession { // so property access is explicit. return ctx.viewSession as unknown as AbortViewSession; } + +type MutableSessionState = InteractiveModeContext["session"] & { + isStreaming: boolean; +}; + +function mutableSessionState(ctx: InteractiveModeContext): MutableSessionState { + // Test harness installs a mutable fake AgentSession; keep the unchecked cast named + // so state mutations are explicit. + return ctx.session as MutableSessionState; +} beforeEach(async () => { await Settings.init({ inMemory: true }); }); @@ -438,111 +448,37 @@ describe("InputController escape behavior", () => { expect(spies.abort).not.toHaveBeenCalled(); }); - it("requires a second Esc within two seconds to abort streaming", () => { - const now = vi.spyOn(Date, "now"); - now.mockReturnValue(1_000); + it("aborts an active streaming turn on the first Esc without asking for confirmation", () => { const { ctx, editor, spies } = createContext(); - (ctx.session as { isStreaming: boolean }).isStreaming = true; + mutableSessionState(ctx).isStreaming = true; const controller = new InputController(ctx); controller.setupKeyHandlers(); editor.onEscape?.(); + expect(spies.abort).toHaveBeenCalledTimes(1); + expect(spies.abort).toHaveBeenCalledWith({ reason: USER_INTERRUPT_LABEL }); + expect(spies.showStatus).not.toHaveBeenCalledWith("Press Esc again within 2s to cancel streaming."); + }); + + it("aborts the submitted turn on the first Esc once the main session starts streaming", async () => { + const { ctx, editor, spies } = createContext(); + const submission = createSubmission({ text: "fix issue #4921" }); + spies.startPendingSubmission.mockReturnValue(submission); + const controller = new InputController(ctx); + + controller.setupKeyHandlers(); + controller.setupEditorSubmitHandler(); + await editor.onSubmit?.("fix issue #4921"); + mutableSessionState(ctx).isStreaming = true; + ctx.loadingAnimation = undefined; + + editor.onEscape?.(); + expect(spies.cancelPendingSubmission).not.toHaveBeenCalled(); - expect(spies.clearQueue).not.toHaveBeenCalled(); - expect(spies.abort).not.toHaveBeenCalled(); - expect(spies.showStatus).toHaveBeenCalledWith("Press Esc again within 2s to cancel streaming."); - - now.mockReturnValue(2_500); - editor.onEscape?.(); - expect(spies.abort).toHaveBeenCalledTimes(1); expect(spies.abort).toHaveBeenCalledWith({ reason: USER_INTERRUPT_LABEL }); - }); - - it("expires the streaming Esc arm instead of aborting on a late second press", () => { - const now = vi.spyOn(Date, "now"); - now.mockReturnValue(1_000); - const { ctx, editor, spies } = createContext(); - (ctx.session as { isStreaming: boolean }).isStreaming = true; - const controller = new InputController(ctx); - - controller.setupKeyHandlers(); - editor.onEscape?.(); - now.mockReturnValue(3_001); - editor.onEscape?.(); - - expect(spies.abort).not.toHaveBeenCalled(); - expect(spies.showStatus).toHaveBeenCalledTimes(2); - }); - - it("preserves the streaming Esc arm when streamingComponent appears between presses", () => { - // Pre-`message_start`: first Esc arms on the per-turn sentinel. `message_start` - // then publishes `ctx.streamingComponent`; the second Esc must still abort the - // same live turn instead of re-arming on the new component reference. - const now = vi.spyOn(Date, "now"); - now.mockReturnValue(1_000); - const { ctx, editor, spies } = createContext(); - (ctx.session as { isStreaming: boolean }).isStreaming = true; - const controller = new InputController(ctx); - - controller.setupKeyHandlers(); - editor.onEscape?.(); - (ctx as unknown as { streamingComponent: object }).streamingComponent = {}; - now.mockReturnValue(1_500); - editor.onEscape?.(); - - expect(spies.abort).toHaveBeenCalledTimes(1); - expect(spies.abort).toHaveBeenCalledWith({ reason: USER_INTERRUPT_LABEL }); - }); - - it("aborts on the second Esc even when ctx.streamingMessage was replaced by a delta in between", () => { - // `EventController` replaces `ctx.streamingMessage` with a fresh immutable - // snapshot on every `message_update`; the per-turn sentinel is unaffected so - // swapping the message must not invalidate the armed token. - const now = vi.spyOn(Date, "now"); - now.mockReturnValue(1_000); - const { ctx, editor, spies } = createContext(); - (ctx.session as { isStreaming: boolean }).isStreaming = true; - (ctx as unknown as { streamingComponent: object }).streamingComponent = {}; - (ctx as unknown as { streamingMessage: object }).streamingMessage = { content: [] }; - const controller = new InputController(ctx); - - controller.setupKeyHandlers(); - editor.onEscape?.(); - (ctx as unknown as { streamingMessage: object }).streamingMessage = { content: ["delta"] }; - now.mockReturnValue(1_500); - editor.onEscape?.(); - - expect(spies.abort).toHaveBeenCalledTimes(1); - expect(spies.abort).toHaveBeenCalledWith({ reason: USER_INTERRUPT_LABEL }); - }); - - it("clears the streaming Esc arm when the current turn ends", () => { - const now = vi.spyOn(Date, "now"); - now.mockReturnValue(1_000); - const { ctx, editor, spies, sessionListeners } = createContext(); - (ctx.session as { isStreaming: boolean }).isStreaming = true; - const controller = new InputController(ctx); - - controller.setupKeyHandlers(); - // Fallback arm (no streamingMessage/streamingComponent yet — pre-message_start). - editor.onEscape?.(); - expect(sessionListeners).toHaveLength(1); - - // Turn 1 ends; a new turn starts. session.subscribe receives both transitions, - // either of which must invalidate the still-armed fallback token so it cannot - // fast-abort the new turn's first Esc. - for (const listener of sessionListeners) { - listener({ type: "agent_end" }); - listener({ type: "agent_start" }); - } - - now.mockReturnValue(1_500); - editor.onEscape?.(); - - expect(spies.abort).not.toHaveBeenCalled(); - expect(spies.showStatus).toHaveBeenCalledTimes(2); + expect(spies.showStatus).not.toHaveBeenCalledWith("Press Esc again within 2s to cancel streaming."); }); it("returns focused subagent view to main on Esc instead of aborting", () => { diff --git a/packages/coding-agent/test/internal-urls/memory-protocol.test.ts b/packages/coding-agent/test/internal-urls/memory-protocol.test.ts index b88eee52d..a0efc7bee 100644 --- a/packages/coding-agent/test/internal-urls/memory-protocol.test.ts +++ b/packages/coding-agent/test/internal-urls/memory-protocol.test.ts @@ -238,6 +238,59 @@ describe("MemoryProtocolHandler — mnemopi bridge (issue #4443)", () => { }); }); + it("resolves memory:// to a read-only fact row (issue #4725)", async () => { + await withMnemopiSession(async ({ state }) => { + const beam = state.memory.beam; + beam.db + .prepare( + "INSERT INTO facts (fact_id, session_id, subject, predicate, object, timestamp, confidence) VALUES (?, ?, ?, ?, ?, ?, ?)", + ) + .run( + "0473bbdb8da6df92", + beam.sessionId, + "Glab", + "works-without", + "mise prefix", + "2026-07-01T00:00:00.000Z", + 0.9, + ); + + const router = InternalUrlRouter.instance(); + const resource = await router.resolve("memory://0473bbdb8da6df92"); + + expect(resource.content).toContain("id: 0473bbdb8da6df92"); + expect(resource.content).toContain("store: fact"); + expect(resource.content).toContain("Glab works-without mise prefix"); + }); + }); + + it("reports not_editable (not not_found) for memory_edit ops on a fact id (issue #4725)", async () => { + await withMnemopiSession(async ({ state }) => { + const beam = state.memory.beam; + beam.db + .prepare( + "INSERT INTO facts (fact_id, session_id, subject, predicate, object, timestamp, confidence) VALUES (?, ?, ?, ?, ?, ?, ?)", + ) + .run("fact-readonly", beam.sessionId, "service", "uses", "postgres", "2026-07-01T00:00:00.000Z", 0.9); + + expect(state.editScopedMemory("update", "fact-readonly", { content: "x" })).toMatchObject({ + status: "not_editable", + store: "fact", + }); + expect(state.editScopedMemory("forget", "fact-readonly")).toMatchObject({ + status: "not_editable", + store: "fact", + }); + expect(state.editScopedMemory("invalidate", "fact-readonly")).toMatchObject({ + status: "not_editable", + store: "fact", + }); + + // The fact row itself is untouched by the rejected edits. + expect(beam.db.prepare("SELECT fact_id FROM facts WHERE fact_id = ?").get("fact-readonly")).not.toBeNull(); + }); + }); + it("routes memory://root to the file-backed summary even when mnemopi is active", async () => { await withMnemopiSession(async () => { const router = InternalUrlRouter.instance(); diff --git a/packages/coding-agent/test/issue-4919-extension-autocomplete-provider.test.ts b/packages/coding-agent/test/issue-4919-extension-autocomplete-provider.test.ts new file mode 100644 index 000000000..31fb9c42d --- /dev/null +++ b/packages/coding-agent/test/issue-4919-extension-autocomplete-provider.test.ts @@ -0,0 +1,256 @@ +/** + * Issue #4919: a pi extension calling `ctx.ui.addAutocompleteProvider(...)` in its + * `session_start` handler crashed at load under omp — the method was absent from + * `ExtensionUIContext`, so the call threw `TypeError: ... is not a function` and + * (for extensions that wrap init in try/catch, e.g. @ff-labs/pi-fff) aborted the + * extension's entire initialization. + * + * These tests pin the pi-compatible contract: + * - headless contexts accept the factory as a no-op instead of throwing, and + * - interactive mode stacks each factory on top of the built-in editor provider. + * + * NOTE: imports are relative (`../src/...`) so the tests exercise this checkout + * even when `node_modules/@oh-my-pi/pi-coding-agent` resolves elsewhere. + */ + +import { afterAll, afterEach, beforeAll, beforeEach, describe, expect, it, vi } from "bun:test"; +import * as fs from "node:fs"; +import * as os from "node:os"; +import * as path from "node:path"; +import { Agent, type AgentTool } from "@oh-my-pi/pi-agent-core"; +import { type Api, Effort, type Model } from "@oh-my-pi/pi-ai"; +import type { AutocompleteProvider } from "@oh-my-pi/pi-tui"; +import { logger, TempDir } from "@oh-my-pi/pi-utils"; +import { type } from "arktype"; +import { ModelRegistry } from "../src/config/model-registry"; +import { resetSettingsForTest, Settings } from "../src/config/settings"; +import { loadExtensions } from "../src/extensibility/extensions/loader"; +import { ExtensionRunner } from "../src/extensibility/extensions/runner"; +import { InteractiveMode } from "../src/modes/interactive-mode"; +import { initTheme } from "../src/modes/theme/theme"; +import { AgentSession } from "../src/session/agent-session"; +import { AuthStorage } from "../src/session/auth-storage"; +import { SessionManager } from "../src/session/session-manager"; + +function makeTool(name: string): AgentTool { + return { + name, + label: name, + description: `Fake ${name}`, + parameters: type({}), + async execute() { + return { content: [{ type: "text" as const, text: "ok" }] }; + }, + }; +} + +/** + * Wrap `current` the way a well-behaved pi extension does: contribute items for + * its own trigger prefix, delegate everything else to the wrapped provider. + */ +function makeWrappingFactory(tag: string): (current: AutocompleteProvider) => AutocompleteProvider { + return current => ({ + async getSuggestions(lines, cursorLine, cursorCol) { + const line = lines[cursorLine] ?? ""; + if (line.startsWith("##")) { + const base = await current.getSuggestions(lines, cursorLine, cursorCol); + return { + items: [...(base?.items ?? []), { value: tag, label: tag }], + prefix: base?.prefix ?? line.slice(0, cursorCol), + }; + } + return current.getSuggestions(lines, cursorLine, cursorCol); + }, + applyCompletion(lines, cursorLine, cursorCol, item, prefix) { + return current.applyCompletion(lines, cursorLine, cursorCol, item, prefix); + }, + }); +} + +describe("extension autocomplete provider API (#4919)", () => { + let tempDir: TempDir; + let authStorage: AuthStorage; + let registry: ModelRegistry; + let model: Model; + let tools: AgentTool[]; + let originalHome: string | undefined; + let mode: InteractiveMode | undefined; + let session: AgentSession | undefined; + + beforeAll(async () => { + initTheme(); + resetSettingsForTest(); + // One empty temp dir doubles as the project cwd and the (isolated) home + // directory, keeping `refreshSlashCommandState`'s capability scan off the + // real home dir (mirrors the prompt-template autocomplete harness). + tempDir = TempDir.createSync("@pi-ext-autocomplete-"); + originalHome = process.env.HOME; + process.env.HOME = tempDir.path(); + await Settings.init({ inMemory: true, cwd: tempDir.path() }); + Settings.instance.set("startup.quiet", true); + authStorage = await AuthStorage.create(path.join(tempDir.path(), "testauth.db")); + authStorage.setRuntimeApiKey("anthropic", "test-key"); + registry = new ModelRegistry(authStorage, path.join(tempDir.path(), "models.yml")); + const resolved = registry.find("anthropic", "claude-sonnet-4-5"); + if (!resolved) throw new Error("Expected anthropic model claude-sonnet-4-5 to exist"); + model = resolved; + tools = [makeTool("read")]; + }); + + beforeEach(() => { + vi.spyOn(os, "homedir").mockReturnValue(tempDir.path()); + }); + + afterEach(async () => { + vi.restoreAllMocks(); + mode?.stop(); + await session?.dispose(); + mode = undefined; + session = undefined; + }); + + afterAll(() => { + authStorage?.close(); + if (originalHome === undefined) delete process.env.HOME; + else process.env.HOME = originalHome; + tempDir?.removeSync(); + resetSettingsForTest(); + }); + + function createHarness(): { mode: InteractiveMode; session: AgentSession } { + const manager = SessionManager.create(tempDir.path(), path.join(tempDir.path(), `active-${Bun.nanoseconds()}`)); + const created = new AgentSession({ + agent: new Agent({ + initialState: { + model, + systemPrompt: ["Test"], + tools, + messages: [], + thinkingLevel: Effort.Medium, + }, + }), + sessionManager: manager, + settings: Settings.isolated({ "compaction.enabled": false }), + modelRegistry: registry, + toolRegistry: new Map(tools.map(tool => [tool.name, tool])), + promptTemplates: [], + }); + const createdMode = new InteractiveMode(created, "test"); + session = created; + mode = createdMode; + return { mode: createdMode, session: created }; + } + + function captureAutocompleteProvider(target: InteractiveMode): { current: AutocompleteProvider | undefined } { + const slot: { current: AutocompleteProvider | undefined } = { current: undefined }; + vi.spyOn(target.editor, "setAutocompleteProvider").mockImplementation(provider => { + slot.current = provider; + }); + return slot; + } + + it("does not abort a session_start handler that registers a provider without UI", async () => { + // Mimics @ff-labs/pi-fff: registerAutocompleteProvider(ctx) runs first and + // unconditionally inside the try/catch that guards the whole init routine. + const extensionsDir = path.join(tempDir.path(), "runner-extensions"); + fs.mkdirSync(extensionsDir, { recursive: true }); + const markerPath = path.join(extensionsDir, "init-marker.txt"); + const extPath = path.join(extensionsDir, "fff-like.ts"); + fs.writeFileSync( + extPath, + `import * as fs from "node:fs"; +export default function (pi) { + pi.on("session_start", async (_event, ctx) => { + try { + ctx.ui.addAutocompleteProvider((current) => current); + // "Rest of init" — on baseline the call above throws and this never runs. + fs.writeFileSync(${JSON.stringify(markerPath)}, "initialized"); + } catch (error) { + fs.writeFileSync( + ${JSON.stringify(markerPath)}, + "failed: " + (error instanceof Error ? error.message : String(error)), + ); + } + }); +} +`, + ); + + const result = await loadExtensions([extPath], tempDir.path()); + expect(result.errors).toEqual([]); + const runner = new ExtensionRunner( + result.extensions, + result.runtime, + tempDir.path(), + SessionManager.inMemory(), + registry, + ); + const surfaced: string[] = []; + runner.onError(error => { + surfaced.push(error.error); + }); + + await runner.emit({ type: "session_start" }); + + expect(surfaced).toEqual([]); + expect(fs.readFileSync(markerPath, "utf8")).toBe("initialized"); + }); + + it("stacks extension factories on top of the built-in editor provider", async () => { + const created = createHarness(); + const slot = captureAutocompleteProvider(created.mode); + + // Registration before the first refresh (session_start fires before init's + // refreshSlashCommandState) must land once the base provider exists. + created.mode.addAutocompleteProvider(makeWrappingFactory("##fff-first")); + await created.mode.refreshSlashCommandState(tempDir.path()); + + const provider = slot.current; + expect(provider).toBeDefined(); + + // The extension's trigger prefix surfaces its items... + const extension = await provider!.getSuggestions(["##"], 0, 2); + expect(extension?.items.map(item => item.value)).toContain("##fff-first"); + + // ...while built-in slash completion still flows through the wrapper. + const slash = await provider!.getSuggestions(["/"], 0, 1); + expect(slash?.items.map(item => item.value)).toContain("model"); + + // Registration after the refresh re-applies immediately, preserving the chain. + created.mode.addAutocompleteProvider(makeWrappingFactory("##fff-second")); + const restacked = slot.current; + expect(restacked).toBeDefined(); + expect(restacked).not.toBe(provider); + + const chained = await restacked!.getSuggestions(["##"], 0, 2); + const values = chained?.items.map(item => item.value) ?? []; + expect(values).toContain("##fff-first"); + expect(values).toContain("##fff-second"); + }); + + it("skips broken factories without losing core autocomplete or healthy wrappers", async () => { + const created = createHarness(); + const slot = captureAutocompleteProvider(created.mode); + const warnSpy = vi.spyOn(logger, "warn").mockImplementation(() => {}); + + created.mode.addAutocompleteProvider(() => { + throw new Error("boom"); + }); + created.mode.addAutocompleteProvider(() => ({}) as AutocompleteProvider); + created.mode.addAutocompleteProvider(makeWrappingFactory("##healthy")); + await created.mode.refreshSlashCommandState(tempDir.path()); + + const provider = slot.current; + expect(provider).toBeDefined(); + + const slash = await provider!.getSuggestions(["/"], 0, 1); + expect(slash?.items.map(item => item.value)).toContain("model"); + + const extension = await provider!.getSuggestions(["##"], 0, 2); + expect(extension?.items.map(item => item.value)).toContain("##healthy"); + + expect(warnSpy.mock.calls.some(([message]) => String(message).includes("autocomplete provider factory"))).toBe( + true, + ); + }); +}); diff --git a/packages/coding-agent/test/keybindings-migration.test.ts b/packages/coding-agent/test/keybindings-migration.test.ts index ec113449f..bd54ff5a2 100644 --- a/packages/coding-agent/test/keybindings-migration.test.ts +++ b/packages/coding-agent/test/keybindings-migration.test.ts @@ -4,13 +4,32 @@ import * as os from "node:os"; import * as path from "node:path"; import { KeybindingsManager } from "@oh-my-pi/pi-coding-agent/config/keybindings"; import { matchesAppFollowUp } from "@oh-my-pi/pi-coding-agent/modes/utils/keybinding-matchers"; -import { setKeybindings } from "@oh-my-pi/pi-tui"; -import { removeWithRetries } from "@oh-my-pi/pi-utils"; +import { type KeybindingsConfig, setKeybindings } from "@oh-my-pi/pi-tui"; +import { + __resetDirsFromEnvForTests, + getAgentDir, + getProfileRootDir, + removeWithRetries, + setProfile, +} from "@oh-my-pi/pi-utils"; import { YAML } from "bun"; function ctrl(key: string): string { return String.fromCharCode(key.toLowerCase().charCodeAt(0) & 31); } + +async function writeKeybindingsYaml(agentDir: string, config: KeybindingsConfig): Promise { + await fs.mkdir(agentDir, { recursive: true }); + await Bun.write(path.join(agentDir, "keybindings.yml"), YAML.stringify(config, null, 2)); +} + +function restoreEnvValue(key: string, value: string | undefined): void { + if (value === undefined) { + delete process.env[key]; + } else { + process.env[key] = value; + } +} describe("KeybindingsManager.create", () => { beforeEach(() => { setKeybindings(KeybindingsManager.inMemory()); @@ -149,6 +168,117 @@ describe("KeybindingsManager.create", () => { } }); + it("inherits default user keybindings for a named profile without a profile keybindings file (#4867)", async () => { + const rootDir = await fs.mkdtemp(path.join(os.tmpdir(), "pi-keybindings-profile-")); + const defaultAgentDir = path.join(rootDir, "default", "agent"); + const profileAgentDir = path.join(rootDir, "profiles", "work", "agent"); + + await writeKeybindingsYaml(defaultAgentDir, { + "app.session.fork": "ctrl+f", + "tui.editor.deleteCharBackward": ["backspace", "ctrl+h"], + }); + + try { + const manager = KeybindingsManager.create(profileAgentDir, { inheritedAgentDir: defaultAgentDir }); + + expect(manager.getKeys("app.session.fork")).toEqual(["ctrl+f"]); + expect(manager.getKeys("tui.editor.deleteCharBackward")).toEqual(["backspace", "ctrl+h"]); + } finally { + await removeWithRetries(rootDir); + } + }); + + it("merges default user keybindings with profile overrides for a named profile (#4867)", async () => { + const rootDir = await fs.mkdtemp(path.join(os.tmpdir(), "pi-keybindings-profile-")); + const defaultAgentDir = path.join(rootDir, "default", "agent"); + const profileAgentDir = path.join(rootDir, "profiles", "work", "agent"); + + await writeKeybindingsYaml(defaultAgentDir, { + "app.session.fork": "ctrl+f", + "app.session.new": "ctrl+n", + }); + await writeKeybindingsYaml(profileAgentDir, { + "app.session.fork": "alt+f", + "app.clipboard.copyLine": "alt+l", + }); + + try { + const manager = KeybindingsManager.create(profileAgentDir, { inheritedAgentDir: defaultAgentDir }); + + expect(manager.getKeys("app.session.new")).toEqual(["ctrl+n"]); + expect(manager.getKeys("app.session.fork")).toEqual(["alt+f"]); + expect(manager.getKeys("app.clipboard.copyLine")).toEqual(["alt+l"]); + } finally { + await removeWithRetries(rootDir); + } + }); + + it("never writes migration output into the inherited default agent dir (#4867)", async () => { + const rootDir = await fs.mkdtemp(path.join(os.tmpdir(), "pi-keybindings-profile-")); + const defaultAgentDir = path.join(rootDir, "default", "agent"); + const profileAgentDir = path.join(rootDir, "profiles", "work", "agent"); + + // Legacy JSON in the default dir: loading it with a write-back path would + // materialize keybindings.yml there. The inherited load must stay read-only. + await fs.mkdir(defaultAgentDir, { recursive: true }); + await Bun.write( + path.join(defaultAgentDir, "keybindings.json"), + JSON.stringify({ "app.session.fork": "ctrl+f" }, null, 2), + ); + + try { + const manager = KeybindingsManager.create(profileAgentDir, { inheritedAgentDir: defaultAgentDir }); + + expect(manager.getKeys("app.session.fork")).toEqual(["ctrl+f"]); + expect(await Bun.file(path.join(defaultAgentDir, "keybindings.yml")).exists()).toBe(false); + } finally { + await removeWithRetries(rootDir); + } + }); + + it("merges default user keybindings when create uses the active profile with no arguments (#4867)", async () => { + const originalConfigDir = process.env.PI_CONFIG_DIR; + const originalAgentDirEnv = process.env.PI_CODING_AGENT_DIR; + const originalOmpProfile = process.env.OMP_PROFILE; + const originalPiProfile = process.env.PI_PROFILE; + const configRootDir = await fs.mkdtemp(path.join(os.tmpdir(), "pi-keybindings-active-profile-")); + + try { + process.env.PI_CONFIG_DIR = path.relative(os.homedir(), configRootDir); + restoreEnvValue("PI_CODING_AGENT_DIR", originalAgentDirEnv); + restoreEnvValue("OMP_PROFILE", originalOmpProfile); + restoreEnvValue("PI_PROFILE", originalPiProfile); + __resetDirsFromEnvForTests(); + + const defaultAgentDir = path.join(getProfileRootDir(undefined), "agent"); + const profileAgentDir = path.join(getProfileRootDir("work"), "agent"); + await writeKeybindingsYaml(defaultAgentDir, { + "app.session.fork": "ctrl+f", + "app.session.new": "ctrl+n", + }); + await writeKeybindingsYaml(profileAgentDir, { + "app.session.fork": "alt+f", + "app.clipboard.copyLine": "alt+l", + }); + + setProfile("work"); + + expect(getAgentDir()).toBe(profileAgentDir); + const manager = KeybindingsManager.create(); + + expect(manager.getKeys("app.session.new")).toEqual(["ctrl+n"]); + expect(manager.getKeys("app.session.fork")).toEqual(["alt+f"]); + expect(manager.getKeys("app.clipboard.copyLine")).toEqual(["alt+l"]); + } finally { + restoreEnvValue("PI_CONFIG_DIR", originalConfigDir); + restoreEnvValue("PI_CODING_AGENT_DIR", originalAgentDirEnv); + restoreEnvValue("OMP_PROFILE", originalOmpProfile); + restoreEnvValue("PI_PROFILE", originalPiProfile); + __resetDirsFromEnvForTests(); + await removeWithRetries(configRootDir); + } + }); + it("defaults model selection to Alt+M and display reset to Ctrl+L", () => { const manager = KeybindingsManager.inMemory(); diff --git a/packages/coding-agent/test/model-discovery.test.ts b/packages/coding-agent/test/model-discovery.test.ts index dd7d89913..393e60645 100644 --- a/packages/coding-agent/test/model-discovery.test.ts +++ b/packages/coding-agent/test/model-discovery.test.ts @@ -3,6 +3,7 @@ import * as fs from "node:fs"; import * as os from "node:os"; import * as path from "node:path"; import { Effort, type FetchImpl, type Model } from "@oh-my-pi/pi-ai"; +import type { OAuthCredentials } from "@oh-my-pi/pi-ai/oauth/types"; import { buildModel } from "@oh-my-pi/pi-catalog/build"; import { writeModelCache } from "@oh-my-pi/pi-catalog/model-cache"; import type { OpenAICompat } from "@oh-my-pi/pi-catalog/types"; @@ -19,15 +20,18 @@ describe("ModelRegistry runtime discovery", () => { let originalOllamaBaseUrl: string | undefined; let originalOllamaHost: string | undefined; let originalOllamaContextLength: string | undefined; + let originalAnthropicApiKey: string | undefined; beforeEach(async () => { resetSettingsForTest(); originalOllamaBaseUrl = Bun.env.OLLAMA_BASE_URL; originalOllamaHost = Bun.env.OLLAMA_HOST; originalOllamaContextLength = Bun.env.OLLAMA_CONTEXT_LENGTH; + originalAnthropicApiKey = Bun.env.ANTHROPIC_API_KEY; delete Bun.env.OLLAMA_BASE_URL; delete Bun.env.OLLAMA_HOST; delete Bun.env.OLLAMA_CONTEXT_LENGTH; + delete Bun.env.ANTHROPIC_API_KEY; tempDir = path.join(os.tmpdir(), `pi-test-model-registry-${Snowflake.next()}`); fs.mkdirSync(tempDir, { recursive: true }); modelsJsonPath = path.join(tempDir, "models.json"); @@ -55,6 +59,11 @@ describe("ModelRegistry runtime discovery", () => { } else { Bun.env.OLLAMA_CONTEXT_LENGTH = originalOllamaContextLength; } + if (originalAnthropicApiKey === undefined) { + delete Bun.env.ANTHROPIC_API_KEY; + } else { + Bun.env.ANTHROPIC_API_KEY = originalAnthropicApiKey; + } authStorage.close(); if (tempDir && fs.existsSync(tempDir)) { removeSyncWithRetries(tempDir); @@ -115,6 +124,197 @@ describe("ModelRegistry runtime discovery", () => { }; } + async function useAuthStorageWithRefreshTracker() { + authStorage.close(); + const refreshCalls: string[] = []; + authStorage = await AuthStorage.create(":memory:", { + refreshOAuthCredential: async (provider, _credentialId, credential): Promise => { + refreshCalls.push(provider); + return { + ...credential, + access: provider === "anthropic" ? "sk-ant-oat-fresh-anthropic" : `fresh-${provider}`, + expires: Date.now() + 3_600_000, + }; + }, + }); + return { refreshCalls }; + } + + type AnthropicDiscoveryCapture = { + modelListAuthorization?: string | null; + modelListXApiKey?: string | null; + modelListCalls: number; + }; + + function mockAnthropicModelsDiscovery(capture: AnthropicDiscoveryCapture): FetchImpl { + const endpointPrefix = "https://api.anthropic.com/"; + return async (input, init) => { + const url = String(input); + if (url === "https://models.dev/api.json") { + return Response.json({}); + } + if (url.startsWith(endpointPrefix) && url.endsWith("/models")) { + const headers = new Headers(init?.headers); + capture.modelListAuthorization = headers.get("authorization"); + capture.modelListXApiKey = headers.get("x-api-key"); + capture.modelListCalls++; + return Response.json({ + data: [{ id: "claude-regression-4893", display_name: "Claude Regression 4893" }], + }); + } + throw new Error(`Unexpected URL: ${url}`); + }; + } + + test("refreshProvider online refreshes expired anthropic OAuth before model discovery", async () => { + const { refreshCalls } = await useAuthStorageWithRefreshTracker(); + await authStorage.set("anthropic", { + type: "oauth", + access: "sk-ant-oat-expired-anthropic", + refresh: "refresh-anthropic", + expires: Date.now() - 60_000, + }); + const capture: AnthropicDiscoveryCapture = { modelListCalls: 0 }; + const registry = new ModelRegistry(authStorage, modelsJsonPath, { + fetch: mockAnthropicModelsDiscovery(capture), + }); + + await registry.refreshProvider("anthropic", "online"); + + expect(refreshCalls).toEqual(["anthropic"]); + expect(capture.modelListCalls).toBe(1); + expect(capture.modelListAuthorization).toBe("Bearer sk-ant-oat-fresh-anthropic"); + expect(capture.modelListXApiKey).toBeNull(); + expect(registry.find("anthropic", "claude-regression-4893")).toBeDefined(); + }); + + test("refreshProvider online does not refresh unrelated expired OAuth credentials", async () => { + const { refreshCalls } = await useAuthStorageWithRefreshTracker(); + await authStorage.set("anthropic", { + type: "oauth", + access: "sk-ant-oat-expired-anthropic", + refresh: "refresh-anthropic", + expires: Date.now() - 60_000, + }); + await authStorage.set("openai", { + type: "oauth", + access: "expired-openai", + refresh: "refresh-openai", + expires: Date.now() - 60_000, + }); + const capture: AnthropicDiscoveryCapture = { modelListCalls: 0 }; + const registry = new ModelRegistry(authStorage, modelsJsonPath, { + fetch: mockAnthropicModelsDiscovery(capture), + }); + + await registry.refreshProvider("anthropic", "online"); + + expect(refreshCalls).toEqual(["anthropic"]); + expect(authStorage.getOAuthCredential("openai")?.access).toBe("expired-openai"); + expect(capture.modelListCalls).toBe(1); + }); + + test("refreshProvider offline does not touch expired OAuth credentials", async () => { + const { refreshCalls } = await useAuthStorageWithRefreshTracker(); + await authStorage.set("anthropic", { + type: "oauth", + access: "sk-ant-oat-expired-anthropic", + refresh: "refresh-anthropic", + expires: Date.now() - 60_000, + }); + const registry = new ModelRegistry(authStorage, modelsJsonPath, { + fetch: async input => { + throw new Error(`Offline discovery should not fetch ${String(input)}`); + }, + }); + + await registry.refreshProvider("anthropic", "offline"); + + expect(refreshCalls).toEqual([]); + expect(authStorage.getOAuthCredential("anthropic")?.access).toBe("sk-ant-oat-expired-anthropic"); + }); + test("online-if-uncached refreshes expired OAuth when the discovery cache is stale for the model manager", async () => { + const { refreshCalls } = await useAuthStorageWithRefreshTracker(); + await authStorage.set("anthropic", { + type: "oauth", + access: "sk-ant-oat-expired-anthropic", + refresh: "refresh-anthropic", + expires: Date.now() - 60_000, + }); + // Older than the model manager's 2h default TTL: the manager WILL fetch, + // so the preflight must mint a fresh bearer first. + writeModelCache("anthropic", Date.now() - 3 * 60 * 60 * 1000, [], true, "", cacheDbPath); + const capture: AnthropicDiscoveryCapture = { modelListCalls: 0 }; + const registry = new ModelRegistry(authStorage, modelsJsonPath, { + fetch: mockAnthropicModelsDiscovery(capture), + }); + + await registry.refreshProvider("anthropic", "online-if-uncached"); + + expect(refreshCalls).toEqual(["anthropic"]); + expect(capture.modelListCalls).toBe(1); + expect(capture.modelListAuthorization).toBe("Bearer sk-ant-oat-fresh-anthropic"); + }); + + test("online-if-uncached leaves expired OAuth untouched when the discovery cache is fresh", async () => { + const { refreshCalls } = await useAuthStorageWithRefreshTracker(); + await authStorage.set("anthropic", { + type: "oauth", + access: "sk-ant-oat-expired-anthropic", + refresh: "refresh-anthropic", + expires: Date.now() - 60_000, + }); + // Fresh authoritative cache: the manager will not fetch, so opening a + // cached model selector must not rotate (or risk disabling) credentials. + writeModelCache("anthropic", Date.now() - 60_000, [], true, "", cacheDbPath); + const capture: AnthropicDiscoveryCapture = { modelListCalls: 0 }; + const registry = new ModelRegistry(authStorage, modelsJsonPath, { + fetch: mockAnthropicModelsDiscovery(capture), + }); + + await registry.refreshProvider("anthropic", "online-if-uncached"); + + expect(refreshCalls).toEqual([]); + expect(capture.modelListCalls).toBe(0); + expect(authStorage.getOAuthCredential("anthropic")?.access).toBe("sk-ant-oat-expired-anthropic"); + }); + + test("configured discovery suppresses built-in special OAuth discovery", async () => { + await authStorage.set("google-gemini-cli", { + type: "oauth", + access: "fresh-google-gemini-cli", + refresh: "refresh-google-gemini-cli", + expires: Date.now() + 3_600_000, + }); + writeRawModelsJson({ + "google-gemini-cli": { + baseUrl: "http://127.0.0.1:4893", + api: "openai-completions", + auth: "none", + discovery: { type: "openai-models-list" }, + }, + }); + const unexpectedUrls: string[] = []; + const fetchMock: FetchImpl = async input => { + const url = String(input); + if (url === "http://127.0.0.1:4893/v1/models") { + return Response.json({ + data: [{ id: "configured-gemini-cli-model", context_length: 65_536 }], + }); + } + unexpectedUrls.push(url); + throw new Error(`Unexpected URL: ${url}`); + }; + const registry = new ModelRegistry(authStorage, modelsJsonPath, { fetch: fetchMock }); + + await registry.refreshProvider("google-gemini-cli", "online"); + + expect(unexpectedUrls).toEqual([]); + const configuredModel = registry.find("google-gemini-cli", "configured-gemini-cli-model"); + expect(configuredModel?.baseUrl).toBe("http://127.0.0.1:4893"); + expect(configuredModel?.contextWindow).toBe(65_536); + }); + test("auto-discovers ollama models without provider config", async () => { const fetchMock = mockOllamaDiscovery(["phi4-mini"]); const registry = new ModelRegistry(authStorage, modelsJsonPath, { fetch: fetchMock }); diff --git a/packages/coding-agent/test/model-registry-runtime-provider.test.ts b/packages/coding-agent/test/model-registry-runtime-provider.test.ts index 467cc0ec8..977e8a665 100644 --- a/packages/coding-agent/test/model-registry-runtime-provider.test.ts +++ b/packages/coding-agent/test/model-registry-runtime-provider.test.ts @@ -261,6 +261,52 @@ describe("ModelRegistry runtime provider registration", () => { }); }); + test("configured discovery suppresses extension fetchDynamicModels for the same provider", async () => { + const providerName = "runtime-configured-provider"; + fs.writeFileSync( + modelsJsonPath, + JSON.stringify({ + providers: { + [providerName]: { + baseUrl: "http://127.0.0.1:4893", + api: "openai-completions", + auth: "none", + discovery: { type: "openai-models-list" }, + }, + }, + }), + ); + const configuredFetch: FetchImpl = async input => { + const url = String(input); + if (url === "http://127.0.0.1:4893/v1/models") { + return Response.json({ + data: [{ id: "shared-runtime-model", context_length: 32_768 }], + }); + } + throw new Error(`Unexpected URL: ${url}`); + }; + const configuredRegistry = new ModelRegistry(authStorage, modelsJsonPath, { fetch: configuredFetch }); + let runtimeFetchCalls = 0; + configuredRegistry.registerProvider( + providerName, + { + baseUrl: "https://runtime.example.com/v1", + apiKey: "RUNTIME_KEY", + api: "openai-completions", + fetchDynamicModels: async () => { + runtimeFetchCalls++; + return [{ ...baseModel, id: "shared-runtime-model", contextWindow: 999_999 }]; + }, + }, + "ext://runtime", + ); + + await configuredRegistry.refreshProvider(providerName, "online"); + + expect(runtimeFetchCalls).toBe(0); + expect(configuredRegistry.find(providerName, "shared-runtime-model")?.contextWindow).toBe(32_768); + }); + test("refreshRuntimeProviders times out extension fetchDynamicModels that never resolves", async () => { vi.useFakeTimers(); const hangingFetch = Promise.withResolvers[number][]>(); diff --git a/packages/coding-agent/test/modes/components/transcript-container.test.ts b/packages/coding-agent/test/modes/components/transcript-container.test.ts index 73f6bbaa8..ae08beade 100644 --- a/packages/coding-agent/test/modes/components/transcript-container.test.ts +++ b/packages/coding-agent/test/modes/components/transcript-container.test.ts @@ -727,3 +727,163 @@ describe("TranscriptContainer renderViewportTail", () => { expect([...container.renderViewportTail(W, 0)]).toEqual([]); }); }); + +// A displaceable snapshot (todo/poll card): kept unfinalized only so a matching +// follow-up call can retract it. Mirrors ToolExecutionComponent.seal — sealing +// finalizes the block in place and it stops reporting displaceable. A pending +// tool starts non-displaceable and becomes a displaceable snapshot only when +// its successful result arrives (`makeDisplaceable`). +class DisplaceableBlock implements Component { + sealCount = 0; + #sealed = false; + #displaceable: boolean; + #lines: string[]; + constructor(lines: string[], displaceable = true) { + this.#lines = lines; + this.#displaceable = displaceable; + } + makeDisplaceable(): void { + this.#displaceable = true; + } + isTranscriptBlockFinalized(): boolean { + return this.#sealed; + } + isDisplaceableBlock(): boolean { + return this.#displaceable && !this.#sealed; + } + seal(): void { + this.sealCount++; + this.#sealed = true; + } + invalidate(): void {} + render(_width: number): string[] { + return [...this.#lines]; + } +} + +// Seal-on-commit: rows on the native-scrollback tape are immutable, so once the +// commit boundary covers any of a displaceable snapshot's rows the container +// must seal it in place — retracting it would strand an orphaned copy in +// terminal history, and left unfinalized it would pin the live-region seam +// open. setNativeScrollbackCommittedRows is a pure store; the seal walk runs at +// the start of the NEXT render, over the previous frame's segments (the +// geometry the committed count was computed against), before the seam scan so +// the seam unpins in that same frame. +describe("TranscriptContainer seal-on-commit", () => { + const W = 40; + + // history(0) | sep(1) | todo-header(2) | todo-body(3); the leading + // separator row belongs to the card's segment (segment.startRow = 1). + function cardAfterHistory(displaceable = true): { container: TranscriptContainer; card: DisplaceableBlock } { + const container = new TranscriptContainer(); + container.addChild(new MutableBlock(["history"])); + const card = new DisplaceableBlock(["todo-header", "todo-body"], displaceable); + container.addChild(card); + expect(container.render(W)).toEqual(["history", "", "todo-header", "todo-body"]); + return { container, card }; + } + + it("seals on the next render once the boundary covers the block's rows", () => { + const { container, card } = cardAfterHistory(); + // The unsealed card pins the live-region seam at its own rows. + expect(container.getNativeScrollbackLiveRegionStart()).toBe(2); + // Rows 0..2 (through the card's header) are immutable history now. + container.setNativeScrollbackCommittedRows(3); + // The setter is a pure store: sealing waits for the next compose. + expect(card.sealCount).toBe(0); + container.render(W); + expect(card.sealCount).toBe(1); + expect(container.isBlockUncommitted(card)).toBe(false); + // The seal pre-pass ran before the seam scan: the SAME render already + // reports the seam unpinned (no still-mutating block left). + expect(container.getNativeScrollbackLiveRegionStart()).toBeUndefined(); + }); + + it("does not seal while the boundary stays above the block", () => { + const { container, card } = cardAfterHistory(); + // Only "history" committed; the card's rows are all still retractable. + container.setNativeScrollbackCommittedRows(1); + container.render(W); + expect(card.sealCount).toBe(0); + expect(container.isBlockUncommitted(card)).toBe(true); + }); + + it("never seals across same-value or decreasing republishes above the block", () => { + const { container, card } = cardAfterHistory(); + // The engine republishes the committed count every frame (compose and + // post-emit): repeated same-value and decreasing stores above the + // card's rows never accumulate into a seal. + container.setNativeScrollbackCommittedRows(1); + container.render(W); + container.setNativeScrollbackCommittedRows(1); + container.render(W); + container.setNativeScrollbackCommittedRows(0); + container.render(W); + expect(card.sealCount).toBe(0); + expect(container.isBlockUncommitted(card)).toBe(true); + }); + + it("seals exactly once as the boundary sweeps past the block in stages", () => { + const container = new TranscriptContainer(); + container.addChild(new MutableBlock(["history"])); + const card = new DisplaceableBlock(["todo-header", "todo-body"]); + container.addChild(card); + container.addChild(new MutableBlock(["tail"])); + expect(container.render(W)).toEqual(["history", "", "todo-header", "todo-body", "", "tail"]); + // First crossing (through the header) seals; the sealed block stops + // reporting displaceable. + container.setNativeScrollbackCommittedRows(3); + container.render(W); + expect(card.sealCount).toBe(1); + // A later sweep past the whole block, and every subsequent render at + // that boundary, must not seal again. + container.setNativeScrollbackCommittedRows(6); + container.render(W); + container.render(W); + expect(card.sealCount).toBe(1); + }); + + it("never seals a displaceable block with an empty contribution", () => { + const container = new TranscriptContainer(); + container.addChild(new MutableBlock(["history"])); + const empty = new DisplaceableBlock([]); + container.addChild(empty); + container.addChild(new MutableBlock(["tail"])); + expect(container.render(W)).toEqual(["history", "", "tail"]); + // None of the block's rows are on the tape: nothing to seal, ever. + container.setNativeScrollbackCommittedRows(3); + container.render(W); + expect(empty.sealCount).toBe(0); + expect(container.isBlockUncommitted(empty)).toBe(true); + }); + + it("walks past blocks without the displaceable protocol", () => { + const container = new TranscriptContainer(); + const plain = new MutableBlock(["plain-block"]); + container.addChild(plain); + const card = new DisplaceableBlock(["todo-header"]); + container.addChild(card); + expect(container.render(W)).toEqual(["plain-block", "", "todo-header"]); + container.setNativeScrollbackCommittedRows(3); + // The pre-pass visits the plain block first (its rows also committed); + // absent duck-typed methods are a no-op and the card below still seals. + container.render(W); + expect(card.sealCount).toBe(1); + expect(container.isBlockUncommitted(plain)).toBe(false); + }); + + it("seals a block that became displaceable after its rows committed", () => { + // A pending tool's preview rows scroll into native scrollback before + // its successful result arrives; only then does the block become a + // displaceable snapshot. The walk runs every render, so the flip is + // caught on the next compose — not only when the boundary moves. + const { container, card } = cardAfterHistory(false); + container.setNativeScrollbackCommittedRows(3); + container.render(W); + // Rows committed while not displaceable: nothing to seal yet. + expect(card.sealCount).toBe(0); + card.makeDisplaceable(); + container.render(W); + expect(card.sealCount).toBe(1); + }); +}); diff --git a/packages/coding-agent/test/modes/controllers/event-controller-toolcall-finalize.test.ts b/packages/coding-agent/test/modes/controllers/event-controller-toolcall-finalize.test.ts index cd89df02b..01688768a 100644 --- a/packages/coding-agent/test/modes/controllers/event-controller-toolcall-finalize.test.ts +++ b/packages/coding-agent/test/modes/controllers/event-controller-toolcall-finalize.test.ts @@ -40,6 +40,10 @@ function makeStreamingMessage(content: AssistantMessage["content"]): AssistantMe }; } +// Components the controller mounts during a dispatch (pending tool previews). +// Sealed in afterEach so their spinner intervals never outlive the test file. +const mountedComponents: { seal?(): void }[] = []; + function createFixture(streamingMessage: AssistantMessage) { const markTranscriptBlockFinalized = vi.fn(); const streamingComponent = { @@ -49,14 +53,14 @@ function createFixture(streamingMessage: AssistantMessage) { const ctx = { isInitialized: true, init: vi.fn(async () => {}), - ui: { requestRender: vi.fn() }, + ui: { requestRender: vi.fn(), requestComponentRender: vi.fn() }, statusLine: { invalidate: vi.fn() }, updateEditorTopBorder: vi.fn(), streamingComponent, streamingMessage, pendingTools: new Map(), noteDisplayableThinkingContent: vi.fn(() => false), - chatContainer: { addChild: vi.fn() }, + chatContainer: { addChild: vi.fn((child: { seal?(): void }) => mountedComponents.push(child)) }, toolOutputExpanded: false, settings, session: { getToolByName: () => undefined }, @@ -84,6 +88,7 @@ async function dispatchUpdate(message: AssistantMessage) { describe("EventController finalizes assistant block when tool-call args stream", () => { afterEach(() => { + for (const component of mountedComponents.splice(0)) component.seal?.(); resetSettingsForTest(); vi.restoreAllMocks(); }); diff --git a/packages/coding-agent/test/read-tool-group.test.ts b/packages/coding-agent/test/read-tool-group.test.ts index e5b6c43da..f1d6da276 100644 --- a/packages/coding-agent/test/read-tool-group.test.ts +++ b/packages/coding-agent/test/read-tool-group.test.ts @@ -250,6 +250,45 @@ describe("ReadToolGroupComponent", () => { expect(extractLinkTexts(rendered)).not.toContain("src/example.ts:7-9"); }); + it("renders separate selector grouped summary paths while linking only the base path", () => { + settings.override("tui.hyperlinks", "always"); + const component = new ReadToolGroupComponent(); + const resolvedPath = path.resolve("/workspace/src/grouped.ts"); + component.updateArgs({ path: "src/grouped.ts", selector: "2-3" }, "read-split-selector"); + component.updateResult( + { + content: [{ type: "text", text: "line 2" }], + details: { meta: { source: { type: "path", value: resolvedPath } } }, + }, + false, + "read-split-selector", + ); + + const rendered = component.render(120).join("\n"); + + const groupedUri = new URL(url.pathToFileURL(path.resolve(resolvedPath)).href); + groupedUri.searchParams.set("line", "2"); + expect(Bun.stripANSI(rendered)).toContain("Read src/grouped.ts:2-3"); + expect(extractLinkUris(rendered)).toContain(groupedUri.href); + expect(extractLinkTexts(rendered)).toContain("src/grouped.ts"); + expect(extractLinkTexts(rendered)).not.toContain("src/grouped.ts:2-3"); + }); + + it("ignores non-string selectors from malformed runtime args", () => { + const component = new ReadToolGroupComponent(); + const malformedArgs = { path: "src/example.ts", selector: 10 } as unknown as { + path: string; + selector: string; + }; + + expect(() => component.updateArgs(malformedArgs, "read-malformed-selector")).not.toThrow(); + + const plain = Bun.stripANSI(component.render(120).join("\n")); + + expect(plain).toContain("Read src/example.ts"); + expect(plain).not.toContain("src/example.ts:10"); + }); + it("links inline preview titles when the summary row is suppressed", () => { settings.override("tui.hyperlinks", "always"); const component = new ReadToolGroupComponent({ showContentPreview: true }); diff --git a/packages/coding-agent/test/rpc-skill-command.test.ts b/packages/coding-agent/test/rpc-skill-command.test.ts index 068b55bfb..1100ace66 100644 --- a/packages/coding-agent/test/rpc-skill-command.test.ts +++ b/packages/coding-agent/test/rpc-skill-command.test.ts @@ -16,6 +16,7 @@ describe("tryRunRpcSkillCommand", () => { ); let message: Pick | undefined; + let options: { streamingBehavior?: "steer" | "followUp" } | undefined; const handled = await tryRunRpcSkillCommand( { @@ -23,8 +24,9 @@ describe("tryRunRpcSkillCommand", () => { skills: [ { name: "reviewer", description: "Review code", filePath: skillPath, baseDir: dir, source: "project" }, ], - async promptCustomMessage(nextMessage: typeof message) { + async promptCustomMessage(nextMessage: typeof message, nextOptions?: typeof options) { message = nextMessage; + options = nextOptions; }, }, "/skill:reviewer focus on risks", @@ -39,10 +41,49 @@ describe("tryRunRpcSkillCommand", () => { expect(message?.content).toContain("User: focus on risks"); expect(message?.display).toBe(true); expect(message?.attribution).toBe("user"); + expect(options).toEqual({ streamingBehavior: "steer" }); await removeWithRetries(dir); }); + test("honors the RPC prompt streaming behavior for registered /skill commands", async () => { + const dir = await fs.mkdtemp(path.join(os.tmpdir(), `omp-rpc-skill-${Snowflake.next()}-`)); + const skillPath = path.join(dir, "SKILL.md"); + await Bun.write( + skillPath, + "---\nname: reviewer\ndescription: Review code\n---\n\nReview the supplied code carefully.\n", + ); + + let options: { streamingBehavior?: "steer" | "followUp" } | undefined; + try { + const handled = await tryRunRpcSkillCommand( + { + skillsSettings: { enableSkillCommands: true }, + skills: [ + { + name: "reviewer", + description: "Review code", + filePath: skillPath, + baseDir: dir, + source: "project", + }, + ], + async promptCustomMessage(nextMessage, nextOptions) { + expect(nextMessage.customType).toBe(SKILL_PROMPT_MESSAGE_TYPE); + options = nextOptions; + }, + }, + "/skill:reviewer wait for the current turn", + "followUp", + ); + + expect(handled).toEqual({ agentInvoked: true }); + expect(options?.streamingBehavior).toBe("followUp"); + } finally { + await removeWithRetries(dir); + } + }); + test("ignores unknown skill commands so normal prompt handling can continue", async () => { const handled = await tryRunRpcSkillCommand( { diff --git a/packages/coding-agent/test/settings-manager.test.ts b/packages/coding-agent/test/settings-manager.test.ts index a8bf84f85..daf6e15a0 100644 --- a/packages/coding-agent/test/settings-manager.test.ts +++ b/packages/coding-agent/test/settings-manager.test.ts @@ -70,6 +70,51 @@ describe("Settings", () => { await Bun.sleep(0); await tempDir?.remove(); }); + + describe("main config file selection", () => { + it("loads and updates an existing config.yaml without creating config.yml", async () => { + const yamlConfigPath = path.join(agentDir, "config.yaml"); + await Bun.write(yamlConfigPath, YAML.stringify({ setupVersion: 1 }, null, 2)); + + const settings = await Settings.init({ cwd: projectDir, agentDir }); + expect(settings.get("setupVersion")).toBe(1); + + settings.set("setupVersion", 2); + await settings.flush(); + + const savedSettings = YAML.parse(await Bun.file(yamlConfigPath).text()) as Record; + expect(savedSettings.setupVersion).toBe(2); + expect(await Bun.file(getConfigPath()).exists()).toBe(false); + }); + + it("clones the selected config.yaml path for persisted settings", async () => { + const yamlConfigPath = path.join(agentDir, "config.yaml"); + await Bun.write(yamlConfigPath, YAML.stringify({ setupVersion: 1 }, null, 2)); + + const settings = await Settings.init({ cwd: projectDir, agentDir }); + const cloned = await settings.cloneForCwd(tempDir.join("other-project")); + + cloned.set("setupVersion", 2); + await cloned.flush(); + + const savedSettings = YAML.parse(await Bun.file(yamlConfigPath).text()) as Record; + expect(savedSettings.setupVersion).toBe(2); + expect(await Bun.file(getConfigPath()).exists()).toBe(false); + }); + + it("creates config.yml for new persisted settings when no main config exists", async () => { + const yamlConfigPath = path.join(agentDir, "config.yaml"); + + const settings = await Settings.init({ cwd: projectDir, agentDir }); + settings.set("setupVersion", 1); + await settings.flush(); + + expect(await Bun.file(getConfigPath()).exists()).toBe(true); + expect(await Bun.file(yamlConfigPath).exists()).toBe(false); + expect((await readSettings()).setupVersion).toBe(1); + }); + }); + describe("defaults", () => { it("keeps eight inline images live by default", async () => { const settings = await Settings.init({ cwd: projectDir, agentDir }); diff --git a/packages/coding-agent/test/tool-execution-write-repaint.test.ts b/packages/coding-agent/test/tool-execution-write-repaint.test.ts new file mode 100644 index 000000000..d011987e5 --- /dev/null +++ b/packages/coding-agent/test/tool-execution-write-repaint.test.ts @@ -0,0 +1,140 @@ +import { afterEach, beforeAll, describe, expect, it, vi } from "bun:test"; +import { ToolExecutionComponent } from "@oh-my-pi/pi-coding-agent/modes/components/tool-execution"; +import { initTheme } from "@oh-my-pi/pi-coding-agent/modes/theme/theme"; +import { type Component, TUI } from "@oh-my-pi/pi-tui"; +import { StressRenderScheduler } from "../../tui/test/render-stress-scheduler"; +import { VirtualTerminal } from "../../tui/test/virtual-terminal"; + +function writeArgs(lineCount: number) { + return { + path: "notes.txt", + content: Array.from({ length: lineCount }, (_, i) => `line ${i + 1}`).join("\n"), + }; +} + +function partialWriteResult(text = "Writing notes.txt...") { + return { content: [{ type: "text", text }] }; +} + +class Footer implements Component { + constructor(readonly rows: number) {} + invalidate(): void {} + render(_width: number): string[] { + return Array.from({ length: this.rows }, (_, i) => `editor-${i}`); + } +} + +function plainBuffer(term: VirtualTerminal): string[] { + return term + .getScrollBuffer() + .map(row => Bun.stripANSI(row).trimEnd()) + .filter(Boolean); +} + +describe("ToolExecutionComponent write repaint seam", () => { + const components: ToolExecutionComponent[] = []; + + beforeAll(async () => { + await initTheme(); + }); + + afterEach(() => { + for (const component of components) component.stopAnimation(); + components.length = 0; + vi.restoreAllMocks(); + }); + + function makeComponent(args: unknown) { + const resetDisplay = vi.fn(); + const ui = { requestRender() {}, requestComponentRender() {}, resetDisplay } as unknown as TUI; + const component = new ToolExecutionComponent("write", args, {}, undefined, ui); + components.push(component); + resetDisplay.mockClear(); + return { component, resetDisplay }; + } + + it("forces a viewport repaint when a painted collapsed tail window receives its first result", () => { + // 20 lines > WRITE_STREAMING_PREVIEW_LINES (12): the pending preview is a + // tail window the first-result render re-anchors to the top of the file. + const { component, resetDisplay } = makeComponent(writeArgs(20)); + component.render(80); + + component.updateResult(partialWriteResult(), true); + + expect(resetDisplay).toHaveBeenCalledTimes(1); + }); + + it("does not repaint when the pending tail window never reaches the terminal", () => { + const { component, resetDisplay } = makeComponent(writeArgs(20)); + // No render() before the result: a resetDisplay here would wipe native + // scrollback for a shape the user never saw. + + component.updateResult(partialWriteResult(), true); + + expect(resetDisplay).not.toHaveBeenCalled(); + }); + + it("does not repaint a collapsed preview that fits the streaming window", () => { + // 12 lines render top-anchored without a tail window, so the first result + // does not re-anchor the frame; wiping scrollback would be gratuitous. + const { component, resetDisplay } = makeComponent(writeArgs(12)); + component.render(80); + + component.updateResult(partialWriteResult(), true); + + expect(resetDisplay).not.toHaveBeenCalled(); + }); + + it("does not repaint an expanded pending preview", () => { + // Expanded previews show the whole file top-anchored — no tail window to + // re-anchor. + const { component, resetDisplay } = makeComponent(writeArgs(20)); + component.setExpanded(true); + component.render(80); + + component.updateResult(partialWriteResult(), true); + + expect(resetDisplay).not.toHaveBeenCalled(); + }); + + it("removes stale pending tail rows from the terminal buffer when the first partial result arrives", async () => { + const term = new VirtualTerminal(80, 8, 1_000); + const scheduler = new StressRenderScheduler(); + const tui = new TUI(term, undefined, { renderScheduler: scheduler }); + const component = new ToolExecutionComponent("write", writeArgs(20), {}, undefined, tui); + components.push(component); + tui.addChild(component); + tui.addChild(new Footer(5)); + + try { + tui.start(); + await scheduler.drain(term); + const pendingRows = plainBuffer(term); + expect(pendingRows.some(row => row.includes("… (8 earlier lines)"))).toBe(true); + expect(pendingRows.some(row => row.includes("… (streaming)"))).toBe(true); + expect(pendingRows.some(row => row.includes("20 line 20"))).toBe(true); + + component.setArgsComplete(); + tui.requestRender(); + await scheduler.drain(term); + + component.updateResult(partialWriteResult(), true); + tui.requestRender(); + await scheduler.drain(term); + + const rows = plainBuffer(term); + // The stale pending tail window must not survive above the new frame. + expect(rows.some(row => row.includes("… (streaming)"))).toBe(false); + expect(rows.some(row => row.includes("earlier lines"))).toBe(false); + expect(rows.some(row => row.includes("20 line 20"))).toBe(false); + // The first partial-result frame is what remains: progress line plus the + // top-anchored preview. + expect(rows.some(row => row.includes("Writing notes.txt..."))).toBe(true); + expect(rows.some(row => row.includes(" 1 line 1"))).toBe(true); + expect(rows.some(row => row.includes("… 14 more lines"))).toBe(true); + } finally { + tui.stop(); + await term.flush(); + } + }); +}); diff --git a/packages/coding-agent/test/tools.test.ts b/packages/coding-agent/test/tools.test.ts index a3baf9bbd..90c001c02 100644 --- a/packages/coding-agent/test/tools.test.ts +++ b/packages/coding-agent/test/tools.test.ts @@ -330,6 +330,34 @@ describe("Coding Agent Tools", () => { expect(result.details?.truncation).toBeUndefined(); }); + it("treats empty optional selector as omitted for read", async () => { + const testFile = path.join(testDir, "read-empty-selector.txt"); + const content = "alpha\nselector target\nomega"; + fs.writeFileSync(testFile, content); + + const omitted = getTextOutput(await readTool.execute("test-read-empty-selector-omitted", { path: testFile })); + expect(omitted).toContain("alpha"); + expect(omitted).toContain("selector target"); + expect(omitted).toContain("omega"); + + for (const { name, selector } of [ + { name: "empty", selector: "" }, + { name: "whitespace", selector: " \t\n " }, + ]) { + const withOptionalSelector = getTextOutput( + await readTool.execute(`test-read-empty-selector-${name}`, { + path: testFile, + selector, + }), + ); + expect(withOptionalSelector).toBe(omitted); + } + + await expect( + readTool.execute("test-read-empty-selector-malformed", { path: testFile, selector: "-100" }), + ).rejects.toThrow(/Invalid selector/); + }); + it("truncates lines wider than the read column cap, leaving narrow lines untouched", async () => { const wideLine = "x".repeat(1500); const testFile = path.join(testDir, "wide.txt"); @@ -1649,6 +1677,42 @@ function b() { expect(output).toMatch(/\*2\|match line/); }); + it("treats empty optional selector as omitted for search", async () => { + const testFile = path.join(testDir, "grep-empty-selector.txt"); + fs.writeFileSync(testFile, "before\nneedle empty selector\nbetween\nneedle whitespace selector\nafter"); + + const omitted = getTextOutput( + await searchTool.execute("test-search-empty-selector-omitted", { + pattern: "needle", + path: testFile, + }), + ); + expect(omitted).toMatch(/\*2\|needle empty selector/); + expect(omitted).toMatch(/\*4\|needle whitespace selector/); + + for (const { name, selector } of [ + { name: "empty", selector: "" }, + { name: "whitespace", selector: " \t\n " }, + ]) { + const withOptionalSelector = getTextOutput( + await searchTool.execute(`test-search-empty-selector-${name}`, { + pattern: "needle", + path: testFile, + selector, + }), + ); + expect(withOptionalSelector).toBe(omitted); + } + + await expect( + searchTool.execute("test-search-empty-selector-malformed", { + pattern: "needle", + path: testFile, + selector: "not-a-range", + }), + ).rejects.toThrow(/selector "not-a-range" is invalid/); + }); + it("flags a zero-match search as contextually useless", async () => { fs.writeFileSync(path.join(testDir, "plain.txt"), "nothing interesting here\n"); diff --git a/packages/coding-agent/test/tools/grep-directory-selector.test.ts b/packages/coding-agent/test/tools/grep-directory-selector.test.ts new file mode 100644 index 000000000..f164fc20b --- /dev/null +++ b/packages/coding-agent/test/tools/grep-directory-selector.test.ts @@ -0,0 +1,99 @@ +import { afterEach, beforeEach, describe, expect, it } from "bun:test"; +import * as fs from "node:fs/promises"; +import * as os from "node:os"; +import * as path from "node:path"; +import { Settings } from "@oh-my-pi/pi-coding-agent/config/settings"; +import type { ToolSession } from "@oh-my-pi/pi-coding-agent/tools"; +import { GrepTool } from "@oh-my-pi/pi-coding-agent/tools/grep"; +import { removeWithRetries } from "@oh-my-pi/pi-utils"; + +function resultText(result: { content: Array<{ type: string; text?: string }> }): string { + return result.content + .filter(entry => entry.type === "text") + .map(entry => entry.text ?? "") + .join("\n"); +} + +describe("grep explicit line selector on directory searches", () => { + let testDir: string; + + beforeEach(async () => { + testDir = await fs.mkdtemp(path.join(os.tmpdir(), "grep-directory-selector-")); + }); + + afterEach(async () => { + await removeWithRetries(testDir); + }); + + function createSession(): ToolSession { + return { + cwd: testDir, + hasUI: false, + getSessionFile: () => null, + getSessionSpawns: () => "*", + settings: Settings.isolated({ "grep.contextBefore": 0, "grep.contextAfter": 0 }), + }; + } + + it("filters matches by per-file line number instead of rejecting the directory", async () => { + const appDir = path.join(testDir, "scripts", "app"); + await fs.mkdir(appDir, { recursive: true }); + await Bun.write(path.join(appDir, "one.ts"), "outside one\ninside one\noutside one again\n"); + await Bun.write(path.join(appDir, "two.ts"), "outside two\ninside two\noutside two again\n"); + + const result = await new GrepTool(createSession()).execute("grep-directory-selector", { + pattern: "inside|outside", + path: "scripts/app", + selector: "2-2", + }); + + const text = resultText(result); + expect(text).toContain("inside one"); + expect(text).toContain("inside two"); + expect(text).not.toContain("outside one"); + expect(text).not.toContain("outside two"); + expect(text).not.toContain("Line-range selector requires a single file"); + }); + + it("fetches enough directory matches before applying an explicit later line selector", async () => { + const appDir = path.join(testDir, "scripts", "hot"); + await fs.mkdir(appDir, { recursive: true }); + const content = `${Array.from( + { length: 30 }, + (_, index) => `cap-needle line ${String(index + 1).padStart(2, "0")}`, + ).join("\n")}\n`; + await Bun.write(path.join(appDir, "many.ts"), content); + + const result = await new GrepTool(createSession()).execute("grep-directory-selector-cap", { + pattern: "cap-needle", + path: "scripts/hot", + selector: "25-25", + }); + + const text = resultText(result); + expect(text).toContain("cap-needle line 25"); + expect(text).not.toContain("cap-needle line 24"); + expect(text).not.toContain("cap-needle line 26"); + }); + + it("supports open-ended selectors on directories with a finite fetch budget", async () => { + const appDir = path.join(testDir, "scripts", "tail"); + await fs.mkdir(appDir, { recursive: true }); + const content = `${Array.from( + { length: 60 }, + (_, index) => `open-needle line ${String(index + 1).padStart(2, "0")}`, + ).join("\n")}\n`; + await Bun.write(path.join(appDir, "long.ts"), content); + + const result = await new GrepTool(createSession()).execute("grep-directory-selector-open", { + pattern: "open-needle", + path: "scripts/tail", + selector: "35-", + }); + + const text = resultText(result); + expect(text).toContain("open-needle line 35"); + expect(text).not.toContain("open-needle line 34"); + expect(text).not.toContain("Line-range selector requires a single file"); + }); +}); diff --git a/packages/coding-agent/test/tools/read-renderer.test.ts b/packages/coding-agent/test/tools/read-renderer.test.ts index babab0ba0..2a128bd34 100644 --- a/packages/coding-agent/test/tools/read-renderer.test.ts +++ b/packages/coding-agent/test/tools/read-renderer.test.ts @@ -83,6 +83,46 @@ describe("readToolRenderer hyperlinks", () => { expect(extractLinkTexts(rendered)).not.toContain(`${examplePath}:10-12`); }); + it("renders separate selector read call paths while linking only the base path", async () => { + settings.override("tui.hyperlinks", "always"); + const theme = await getThemeByName("dark"); + expect(theme).toBeDefined(); + + const examplePath = path.resolve("/tmp/omp-read/separate-selector.ts"); + const component = readToolRenderer.renderCall( + { path: examplePath, selector: "10-12" }, + { expanded: false, isPartial: false }, + theme!, + ); + + const rendered = component.render(200).join("\n"); + expect(Bun.stripANSI(rendered)).toContain(`${examplePath}:10-12`); + const exampleUri = new URL(url.pathToFileURL(path.resolve(examplePath)).href); + exampleUri.searchParams.set("line", "10"); + expect(extractLinkUris(rendered)).toContain(exampleUri.href); + expect(extractLinkTexts(rendered)).toContain(examplePath); + expect(extractLinkTexts(rendered)).not.toContain(`${examplePath}:10-12`); + }); + + it("renders separate raw read selectors while linking only the base path", async () => { + settings.override("tui.hyperlinks", "always"); + const theme = await getThemeByName("dark"); + expect(theme).toBeDefined(); + + const examplePath = path.resolve("/tmp/omp-read/raw-selector.ts"); + const component = readToolRenderer.renderCall( + { path: examplePath, selector: "raw" }, + { expanded: false, isPartial: false }, + theme!, + ); + + const rendered = component.render(200).join("\n"); + expect(Bun.stripANSI(rendered)).toContain(`${examplePath}:raw`); + expect(extractLinkUris(rendered)).toContain(url.pathToFileURL(path.resolve(examplePath)).href); + expect(extractLinkTexts(rendered)).toContain(examplePath); + expect(extractLinkTexts(rendered)).not.toContain(`${examplePath}:raw`); + }); + it("links HTTP read result headers to the final URL", async () => { settings.override("tui.hyperlinks", "always"); const theme = await getThemeByName("dark"); diff --git a/packages/coding-agent/test/utils/clipboard.test.ts b/packages/coding-agent/test/utils/clipboard.test.ts index 5e7b5f53d..b9598bce0 100644 --- a/packages/coding-agent/test/utils/clipboard.test.ts +++ b/packages/coding-agent/test/utils/clipboard.test.ts @@ -166,6 +166,23 @@ describe("readImageFromClipboard dispatch", () => { expect(calls[0]?.cmd).toContain("-Sta"); }); + it("falls back to PowerShell when native Windows image conversion fails", async () => { + setPlatform("win32"); + const calls: SpawnCall[] = []; + spyPowershell(calls, RED_1X1_PNG_BASE64); + vi.spyOn(native, "readImageFromClipboard").mockRejectedValue( + new Error("The clipboard image could not be converted to the appropriate format."), + ); + + const image = await readImageFromClipboard(); + + expect(calls).toHaveLength(1); + expect(calls[0]?.cmd[0]).toBe("powershell.exe"); + expect(calls[0]?.cmd).toContain("-Sta"); + expect(image?.mimeType).toBe("image/png"); + expect(Array.from(image!.data.subarray(0, 8))).toEqual([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a]); + }); + it("delegates straight to the native bridge on non-WSL linux with a display", async () => { setPlatform("linux"); process.env.DISPLAY = ":0"; diff --git a/packages/hashline/package.json b/packages/hashline/package.json index 742cc1670..a3c5f7cf6 100644 --- a/packages/hashline/package.json +++ b/packages/hashline/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/hashline", - "version": "16.3.12", + "version": "16.3.14", "description": "Hashline: a compact, line-anchored patch language and applier. Pluggable FS/IO so it works over disk, in-memory, or any custom backend.", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/mnemopi/package.json b/packages/mnemopi/package.json index 2f37b76b6..9a4845089 100644 --- a/packages/mnemopi/package.json +++ b/packages/mnemopi/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/pi-mnemopi", - "version": "16.3.12", + "version": "16.3.14", "description": "Local SQLite memory engine for Oh My Pi agents", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/mnemopi/src/core/beam/store.ts b/packages/mnemopi/src/core/beam/store.ts index 5d0d2f9dd..225452fc5 100644 --- a/packages/mnemopi/src/core/beam/store.ts +++ b/packages/mnemopi/src/core/beam/store.ts @@ -665,7 +665,46 @@ export function get(beam: BeamMemoryState, memoryId: string): Row | null { WHERE id = ? AND (session_id = ? OR scope = 'global') `) .get(memoryId, beam.sessionId) as Row | null | undefined; - return episodic == null ? null : { ...episodic, metadata: episodic.metadata_json, memory_store: "episodic" }; + if (episodic != null) return { ...episodic, metadata: episodic.metadata_json, memory_store: "episodic" }; + + return getFact(beam, memoryId); +} + +/** + * Read-only resolution for ids minted from the `facts` table. `recall` + * surfaces `facts.fact_id` as a result id (`factRecall`), so `get` must + * resolve those ids too — otherwise every surfaced fact id is a dead end + * for the read path (issue #4725). Visibility mirrors `factRecall`: + * same-session facts plus explicitly global ones (`scope` is an optional + * column on `facts`; `SELECT *` tolerates banks without it, in which case + * only same-session facts resolve). The row is shaped like the + * working/episodic hits with the full triple as content; + * `memory_store: "fact"` marks it read-only — no update/forget/invalidate + * path mutates `facts`. + */ +function getFact(beam: BeamMemoryState, memoryId: string): Row | null { + const fact = beam.db.prepare("SELECT * FROM facts WHERE fact_id = ?").get(memoryId) as Row | null | undefined; + if (fact == null) return null; + if (fact.session_id !== beam.sessionId && fact.scope !== "global") return null; + const subject = typeof fact.subject === "string" ? fact.subject : ""; + const predicate = typeof fact.predicate === "string" ? fact.predicate : ""; + const object = typeof fact.object === "string" ? fact.object : ""; + return { + id: fact.fact_id, + content: [subject, predicate, object].filter(part => part.length > 0).join(" "), + source: "facts", + timestamp: fact.timestamp ?? null, + session_id: fact.session_id ?? null, + importance: fact.confidence ?? null, + metadata: JSON.stringify({ + subject, + predicate, + object, + source_msg_id: fact.source_msg_id ?? null, + }), + created_at: fact.created_at ?? null, + memory_store: "fact", + }; } export function forgetWorking(beam: BeamMemoryState, memoryId: string): boolean { diff --git a/packages/mnemopi/test/beam-store.test.ts b/packages/mnemopi/test/beam-store.test.ts index f9af3a8b5..fb09c7a62 100644 --- a/packages/mnemopi/test/beam-store.test.ts +++ b/packages/mnemopi/test/beam-store.test.ts @@ -1,4 +1,5 @@ import { afterEach, describe, expect, it } from "bun:test"; +import { recallEnhanced } from "@oh-my-pi/pi-mnemopi/core/beam/recall"; import { initBeam } from "@oh-my-pi/pi-mnemopi/core/beam/schema"; import { exportToDict, @@ -217,3 +218,70 @@ describe("beam store free functions", () => { expect(scratchpadRead(dest).map(row => row.content)).toEqual([]); }); }); + +describe("fact-id read path (issue #4725)", () => { + function insertFact( + beam: BeamMemoryState, + factId: string, + sessionId: string, + subject: string, + predicate: string, + object: string, + confidence = 0.9, + ): void { + beam.db + .prepare( + "INSERT INTO facts (fact_id, session_id, subject, predicate, object, timestamp, confidence) VALUES (?, ?, ?, ?, ?, ?, ?)", + ) + .run(factId, sessionId, subject, predicate, object, "2026-05-30T00:00:00.000Z", confidence); + } + + it("resolves an id surfaced by fact recall to a read-only fact row", async () => { + const beam = makeState(); + insertFact(beam, "fact-postgres", beam.sessionId, "service", "uses", "postgres database", 0.91); + + const results = await recallEnhanced(beam, "postgres", 5, { includeFacts: true }); + const surfaced = results.find(result => result.source === "facts"); + expect(surfaced?.id).toBe("fact-postgres"); + + // memory:// reads and memory_edit both resolve ids via get(); a + // surfaced fact id must not be a dead end. + const row = get(beam, "fact-postgres"); + expect(row).toMatchObject({ + id: "fact-postgres", + content: "service uses postgres database", + source: "facts", + importance: 0.91, + session_id: beam.sessionId, + memory_store: "fact", + }); + expect(JSON.parse(String(row?.metadata))).toMatchObject({ + subject: "service", + predicate: "uses", + object: "postgres database", + }); + }); + + it("keeps fact reads session-scoped like fact recall, honoring explicit global scope", () => { + const beam = makeState(); + insertFact(beam, "fact-other", "session-other", "service", "uses", "postgres database"); + expect(get(beam, "fact-other")).toBeNull(); + + beam.db.run("ALTER TABLE facts ADD COLUMN scope TEXT DEFAULT 'session'"); + beam.db.run("UPDATE facts SET scope = 'global' WHERE fact_id = 'fact-other'"); + expect(get(beam, "fact-other")?.memory_store).toBe("fact"); + }); + + it("keeps working rows first on id collision and never deletes facts via forgetWorking", () => { + const beam = makeState(); + insertFact(beam, "shared-id", beam.sessionId, "service", "uses", "postgres database"); + const workingId = remember(beam, "working row shadowing a fact id"); + beam.db.prepare("UPDATE working_memory SET id = ? WHERE id = ?").run("shared-id", workingId); + + expect(get(beam, "shared-id")?.memory_store).toBe("working"); + + expect(forgetWorking(beam, "fact-missing")).toBe(false); + expect(forgetWorking(beam, "shared-id")).toBe(true); + expect(get(beam, "shared-id")?.memory_store).toBe("fact"); + }); +}); diff --git a/packages/natives/CHANGELOG.md b/packages/natives/CHANGELOG.md index f16aff66a..fb1e08bea 100644 --- a/packages/natives/CHANGELOG.md +++ b/packages/natives/CHANGELOG.md @@ -2,6 +2,14 @@ ## [Unreleased] +## [16.3.13] - 2026-07-09 + +### Fixed + +- Fixed unbounded memory growth in the native bash output bridge when a command produces output faster than the JS event loop consumes it: the shell streaming path now uses a bounded chunk queue with real backpressure (pipe readers park until the JS callback catches up, parking the child on its pipe) instead of buffering the entire surplus in memory. No output is dropped — the rolling tail view, `[raw output: artifact://…]` lossless capture, and byte accounting are unaffected ([#4078](https://github.com/can1357/oh-my-pi/issues/4078)). +- Fixed `readImageFromClipboard` on Windows failing with "could not be converted to the appropriate format" for screenshots taken by Qt-based tools such as PixPin and Snipaste. arboard hands their `CF_DIBV5` payload (`BI_RGB` plus an alpha mask, rewritten to `BI_BITFIELDS`) to a header-less BMP decode that mis-places the pixel offset for V4/V5 bitfield headers; the native reader now falls back to decoding the raw `CF_DIB` clipboard bytes directly, so image paste no longer depends on the PowerShell bridge. ([#3426](https://github.com/can1357/oh-my-pi/issues/3426)) +- Fixed OMP being killed outright (OOM on memory-capped hosts such as WSL) when an output-heavy bash command hit its timeout: the unbounded output-bridge backlog could grow by gigabytes before cancellation and starve the JS event loop far past the deadline; with the bounded backpressured bridge the run resolves at its deadline with flat memory ([#4866](https://github.com/can1357/oh-my-pi/issues/4866)). + ## [16.3.12] - 2026-07-08 ### Fixed diff --git a/packages/natives/native/index.d.ts b/packages/natives/native/index.d.ts index 6649dc486..3220cfb8c 100644 --- a/packages/natives/native/index.d.ts +++ b/packages/natives/native/index.d.ts @@ -170,7 +170,7 @@ export declare function __ompInstallTokioRuntime(): void * `packages/natives/native/index.js` (which derives the name from * `package.json#version`). */ -export declare function __piNativesV16_3_12(): void +export declare function __piNativesV16_3_14(): void /** * Apply conservative pre-execution rewrites to a bash command. diff --git a/packages/natives/native/index.js b/packages/natives/native/index.js index d7314c43d..09c13c442 100644 --- a/packages/natives/native/index.js +++ b/packages/natives/native/index.js @@ -24,7 +24,7 @@ export const Shell = nativeBindings.Shell; // functions export const __ompInstallTokioRuntime = nativeBindings.__ompInstallTokioRuntime; -export const __piNativesV16_3_12 = nativeBindings.__piNativesV16_3_12; +export const __piNativesV16_3_14 = nativeBindings.__piNativesV16_3_14; export const applyBashFixups = nativeBindings.applyBashFixups; export const astEdit = nativeBindings.astEdit; export const astGrep = nativeBindings.astGrep; diff --git a/packages/natives/package.json b/packages/natives/package.json index c0f0e2add..45c07ec2f 100644 --- a/packages/natives/package.json +++ b/packages/natives/package.json @@ -1,6 +1,6 @@ { "name": "@oh-my-pi/pi-natives", - "version": "16.3.12", + "version": "16.3.14", "description": "Native Rust bindings for grep, clipboard, image processing, syntax highlighting, PTY, and shell operations via N-API", "type": "module", "homepage": "https://omp.sh", diff --git a/packages/natives/test/issue-4866-repro.test.ts b/packages/natives/test/issue-4866-repro.test.ts new file mode 100644 index 000000000..b26e97fdb --- /dev/null +++ b/packages/natives/test/issue-4866-repro.test.ts @@ -0,0 +1,109 @@ +/** + * Regression for https://github.com/can1357/oh-my-pi/issues/4866. + * + * "When bash command times out, it exits/crashes OMP as a whole" (WSL). + * + * Root cause: the native shell output bridge (`bridge_chunks` in + * `crates/pi-natives/src/shell.rs` + `emit_chunk` in + * `crates/pi-shell/src/shell.rs`) queued decoded output chunks into an + * unbounded cross-thread channel and fired the JS threadsafe function + * non-blocking, with no backpressure. A producer outrunning the JS consumer + * (`yes | cat` runs as in-process uutils builtins at memory speed; any + * output-heavy long task qualifies) ballooned the native queue by gigabytes + * before the timeout fired, and the callback flood then kept the JS event + * loop saturated so the deadline machinery ran tens of seconds late. + * Measured on the pre-fix baseline (macOS arm64): a `timeoutMs: 1500` run + * through the bash executor resolved after ~30-36 s having forwarded ~6.9 GB, + * with process RSS pinned at ~7 GB. On WSL's memory-capped VM that backlog + * trips the Linux OOM killer, which SIGKILLs the whole OMP process — the + * reported "crashes OMP as a whole". + * + * This test models the real consumer (OutputSink sanitize/tail/render work) + * with a deliberately slow `onChunk` (~1 ms per callback) and pins the fixed + * contract for both the one-shot (`executeShell`) and persistent-session + * (`Shell.run`) paths: + * 1. The run resolves near its deadline (raced against a generous window) + * instead of being dragged out by an unbounded backlog drain. On the + * pre-fix bridge the drain alone needs minutes (tens of thousands of + * queued 64 KiB batches through a ~1 ms consumer). + * 2. Native memory stays bounded: RSS growth over the run stays far under + * the gigabytes the unbounded queue accumulated (bounded(64) queue × + * 64 KiB batches plus JS churn). + * 3. The run still reports `timedOut`, so timeout annotation and session + * quarantine behave as before. + * + * Bounds carry >4x headroom on both sides of every threshold (fixed path + * measured: resolve ≈1 s, RSS delta ≈60 MiB; baseline: unresolved at 6 s, + * RSS delta ≥2 GiB), so the test stays robust on slow CI hosts while the + * failure mode overshoots by orders of magnitude. + */ +import { describe, expect, it } from "bun:test"; +import { executeShell, Shell, type ShellRunResult } from "../native/index.js"; + +/** `yes` and `cat` are in-process uutils builtins: output is produced at + * memory speed, which is what made the unbounded bridge lethal. */ +const FAST_PRODUCER = "yes issue-4866-crash-line | cat"; +const TIMEOUT_MS = 800; +/** Window the timed-out run must resolve within (fixed path: ~1 s; pre-fix + * baseline is still draining its multi-GB backlog minutes later). */ +const RESOLVE_WINDOW_MS = 8_000; +/** RSS growth budget. Fixed path: tens of MiB. Pre-fix: multiple GiB. */ +const MAX_RSS_DELTA_BYTES = 512 * 1024 * 1024; +/** Per-callback consumer cost emulating OutputSink/TUI work. */ +const CONSUMER_STALL_MS = 1; +const TEST_BUDGET_MS = 60_000; + +const posixIt = process.platform === "win32" ? it.skip : it; + +// Real-clock integration test (ts-no-test-timers exception): the run under +// test is a native tokio shell execution behind the N-API boundary — fake JS +// timers cannot advance the native runtime's clock, and the defect being +// pinned is precisely a real-time liveness failure (the JS event loop and +// deadline machinery starved by the callback flood). The stall emulates +// synchronous per-callback consumer cost (CPU work, not scheduling), and the +// resolve window is a liveness bound, not a synchronization guess. +async function runTimedOutFastProducer( + run: (onChunk: (err: Error | null, chunk: string) => void) => Promise, +): Promise { + const rssBefore = process.memoryUsage.rss(); + const slowConsumer = (_err: Error | null, chunk: string) => { + if (chunk) Bun.sleepSync(CONSUMER_STALL_MS); + }; + + const settled = run(slowConsumer).then(result => ({ done: true as const, result })); + const raced = await Promise.race([settled, Bun.sleep(RESOLVE_WINDOW_MS).then(() => ({ done: false as const }))]); + const rssDelta = process.memoryUsage.rss() - rssBefore; + + // (2) Bounded native memory — the unbounded bridge queued gigabytes here. + expect(rssDelta).toBeLessThan(MAX_RSS_DELTA_BYTES); + // (1) Timely resolution — the unbounded bridge dragged the run out for + // minutes past its deadline. + expect(raced.done).toBe(true); + if (raced.done) { + // (3) Timeout is still reported as such. + expect(raced.result.timedOut).toBe(true); + } +} + +describe("issue 4866: bash timeout must not flood the output bridge", () => { + posixIt( + "one-shot executeShell: fast producer with slow consumer times out near its deadline with bounded memory", + async () => { + await runTimedOutFastProducer(onChunk => + executeShell({ command: FAST_PRODUCER, timeoutMs: TIMEOUT_MS }, onChunk), + ); + }, + TEST_BUDGET_MS, + ); + + posixIt( + "persistent Shell.run: fast producer with slow consumer times out near its deadline with bounded memory", + async () => { + const shell = new Shell(); + await runTimedOutFastProducer(onChunk => + shell.run({ command: FAST_PRODUCER, timeoutMs: TIMEOUT_MS }, onChunk), + ); + }, + TEST_BUDGET_MS, + ); +}); diff --git a/packages/snapcompact/package.json b/packages/snapcompact/package.json index 3e52a7a2c..11b7bc0d4 100644 --- a/packages/snapcompact/package.json +++ b/packages/snapcompact/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/snapcompact", - "version": "16.3.12", + "version": "16.3.14", "description": "Bitmap-frame context compression for vision-capable LLMs", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/stats/package.json b/packages/stats/package.json index 10c8e2e97..a20592455 100644 --- a/packages/stats/package.json +++ b/packages/stats/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/omp-stats", - "version": "16.3.12", + "version": "16.3.14", "description": "Local observability dashboard for pi AI usage statistics", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/swarm-extension/package.json b/packages/swarm-extension/package.json index ccc809a1b..f5fee963b 100644 --- a/packages/swarm-extension/package.json +++ b/packages/swarm-extension/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/swarm-extension", - "version": "16.3.12", + "version": "16.3.14", "description": "Swarm orchestration extension for omp", "homepage": "https://omp.sh", "author": "Derek Rynd", diff --git a/packages/tui/CHANGELOG.md b/packages/tui/CHANGELOG.md index 9f4b3518e..080025328 100644 --- a/packages/tui/CHANGELOG.md +++ b/packages/tui/CHANGELOG.md @@ -5,6 +5,19 @@ ### Added - Added optional right-border scrollbar to the `Editor` component (`setScrollbarVisible`): shows a thumb glyph on the right border when content overflows `maxHeight`, enabling scrollable multi-line editors (e.g. advisor instructions) without losing the submit hint off-screen. +## [16.3.14] - 2026-07-09 + +### Fixed + +- Fixed race condition where scrollback rows could be incorrectly retracted between render frames + +## [16.3.13] - 2026-07-09 + +### Fixed + +- Fixed late terminal appearance subscribers missing the already-detected OSC 11 light/dark result, so theme auto-detection picks up the terminal appearance even when the response arrives before the UI subscribes ([#4731](https://github.com/can1357/oh-my-pi/issues/4731)). +- Fixed slash command Tab completion reopening the file autocomplete drawer after accepting no-argument commands ([#4808](https://github.com/can1357/oh-my-pi/issues/4808)). + ## [16.3.12] - 2026-07-08 ### Fixed diff --git a/packages/tui/package.json b/packages/tui/package.json index 682e2b532..e672194e4 100644 --- a/packages/tui/package.json +++ b/packages/tui/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/pi-tui", - "version": "16.3.12", + "version": "16.3.14", "description": "Terminal User Interface library with differential rendering for efficient text-based applications", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/tui/src/autocomplete.ts b/packages/tui/src/autocomplete.ts index df6b34b1e..54298a60d 100644 --- a/packages/tui/src/autocomplete.ts +++ b/packages/tui/src/autocomplete.ts @@ -437,6 +437,9 @@ export class CombinedAutocompleteProvider implements AutocompleteProvider { const argumentText = commandText.slice(spaceIndex + 1); // Text after space const command = this.#commands.find(cmd => commandMatchesNameOrAlias(cmd, commandName)); + if (command && "allowArgs" in command && command.allowArgs === false && !/\S/.test(argumentText)) { + return null; + } if (command && (!("allowArgs" in command) || command.allowArgs !== false)) { if (!("getArgumentCompletions" in command) || !command.getArgumentCompletions) { return null; // No argument completion for this command diff --git a/packages/tui/src/terminal.ts b/packages/tui/src/terminal.ts index ed9210af9..b35dfa1f5 100644 --- a/packages/tui/src/terminal.ts +++ b/packages/tui/src/terminal.ts @@ -382,6 +382,8 @@ export interface Terminal { * Register a callback for terminal appearance (dark/light) changes. * Detection uses OSC 11 background color query with Mode 2031 as a change trigger. * Fires when the detected appearance changes, including the initial detection. + * Subscribers registered after detection are invoked immediately with the + * already-detected appearance so late subscribers never miss it. */ onAppearanceChange(callback: (appearance: TerminalAppearance) => void): void; /** The last detected terminal appearance, or undefined if not yet known. */ @@ -516,6 +518,17 @@ export class ProcessTerminal implements Terminal { onAppearanceChange(callback: (appearance: TerminalAppearance) => void): void { this.#appearanceCallbacks.push(callback); + // Replay an already-detected appearance: the startup OSC 11 response can + // arrive before consumers (e.g. the theme bridge) subscribe, and the + // dedup in #handleOsc11Response would otherwise suppress the value for + // them forever (#4731). + if (this.#appearance) { + try { + callback(this.#appearance); + } catch { + /* ignore callback errors */ + } + } } onPrivateModeReport(callback: (mode: number, supported: boolean) => void): void { @@ -704,7 +717,10 @@ export class ProcessTerminal implements Terminal { const decrpmResponsePattern = /^\x1b\[\?(\d+);(\d+)\$y$/; // In-band resize report (DEC mode 2048): \x1b[48;rows;cols;yPixels;xPixels t - const inBandResizePattern = /^\x1b\[48;(\d+);(\d+);(\d+);(\d+)t$/; + // Any field may carry `:`-separated subparameters, which clients MUST + // ignore per spec (#4748): capture the leading digits of each field and + // skip the subparameter tail instead of dropping the whole report. + const inBandResizePattern = /^\x1b\[48;(\d+)(?::[\d:]*)?;(\d+)(?::[\d:]*)?;(\d+)(?::[\d:]*)?;(\d+)(?::[\d:]*)?t$/; this.#stdinBuffer.on("data", (sequence: string) => { // Fast path for plain-text bytes: every escape-probe regex below @@ -776,7 +792,7 @@ export class ProcessTerminal implements Terminal { // reassembled sequence that turns out not to be a resize report (e.g. a // split kitty `\x1b[48;…u` for a digit key) is forwarded to the input // handler rather than dropped. - const inBandResizePartialPattern = /^\x1b\[4[\d;]*$/; + const inBandResizePartialPattern = /^\x1b\[4[\d;:]*$/; const isInBandResizePartial = this.#inBandResizeActive && inBandResizePartialPattern.test(sequence); if (this.#inBandResizeBuffer && sequence.startsWith("\x1b")) { // A new escape interrupted the partial; the stale partial is @@ -1185,9 +1201,9 @@ export class ProcessTerminal implements Terminal { * `rows` before the `resize` event fires, so they are authoritative for the * new cell geometry. A cached DEC 2048 report can be stale: the matching * post-resize report may be dropped (split across stdin reads past the flush - * window) or carry `:`-subparameters the parser skips, leaving the getters - * pinned to the old size — which freezes the rendered width because the - * renderer reflows against {@link columns}/{@link rows}, not the live OS + * window, or interrupted by another escape mid-reassembly), leaving the + * getters pinned to the old size — which freezes the rendered width because + * the renderer reflows against {@link columns}/{@link rows}, not the live OS * value. Drop a cached dimension that disagrees with the live OS value; the * terminal's next valid in-band report re-seeds pixel sizing. */ diff --git a/packages/tui/src/tui.ts b/packages/tui/src/tui.ts index 0d44beb4a..a916af1d5 100644 --- a/packages/tui/src/tui.ts +++ b/packages/tui/src/tui.ts @@ -1137,8 +1137,14 @@ export class TUI extends Container { // Feed the engine's committed-row claim (from the previous frame's // emit) before rendering so the child can skip re-deriving blocks // that already live in immutable native scrollback. Reused segments - // skip this: they never call render(), so the signal is moot. - setNativeScrollbackCommittedRows(child, Math.max(0, this.#committedRows - offset)); + // skip this: they never call render(), so the signal is moot. The + // claim is in the previous frame's coordinates and never exceeds + // the rows the child actually contributed there — history that + // advanced into LATER root children must not read as this child's + // own future rows being pre-committed. + const prevRows = previous !== undefined && previous.component === child ? previous.rowCount : 0; + const prevStart = previous !== undefined && previous.component === child ? previous.start : offset; + setNativeScrollbackCommittedRows(child, Math.min(prevRows, Math.max(0, this.#committedRows - prevStart))); childLines = child.render(width); const liveRegionStart = getNativeScrollbackLiveRegionStart(child); if (liveRegionStart !== undefined) { @@ -2802,6 +2808,7 @@ export class TUI extends Container { this.#committedPrefixAuditRows = Math.min(chunkTo, finalBoundary); this.#clearScrollbackOnNextRender = false; this.#hasEverRendered = true; + this.#publishCommittedRows(); if (!firstPaint && frameLength > height) this.#armPostFullPaintSettle(); return; } @@ -2829,6 +2836,7 @@ export class TUI extends Container { } else { this.#committedPrefixAuditRows = Math.min(preAuditRows, this.#committedRows); } + this.#publishCommittedRows(); } /** @@ -2853,6 +2861,25 @@ export class TUI extends Container { } } + /** + * Push the post-emit committed-row count to root children that implement + * {@link NativeScrollbackCommittedRows}. Compose feeds the same signal + * before each child render (see {@link render}), but guards that run + * BETWEEN frames — e.g. a controller consulting the transcript's + * committed boundary to decide whether a displaceable block may still be + * retracted — would otherwise observe a count one frame stale and retract + * rows that just entered immutable native scrollback, stranding an + * orphaned copy above the repainted block. + */ + #publishCommittedRows(): void { + for (const segment of this.#frameSegments) { + setNativeScrollbackCommittedRows( + segment.component, + Math.min(segment.rowCount, Math.max(0, this.#committedRows - segment.start)), + ); + } + } + /** * Prepare the composed frame for emission, in place. Rows below * `#preparedValidRows` are already prepared against the current frame (the diff --git a/packages/tui/test/autocomplete.test.ts b/packages/tui/test/autocomplete.test.ts index 5f2733d07..f3f21ce4c 100644 --- a/packages/tui/test/autocomplete.test.ts +++ b/packages/tui/test/autocomplete.test.ts @@ -119,6 +119,23 @@ describe("CombinedAutocompleteProvider", () => { expect(result?.items.map(item => item.value)).toContain("/tmp/"); }); + it("does not treat whitespace-only no-arg slash command arguments as file prefixes", async () => { + const baseDir = fs.mkdtempSync(path.join(os.tmpdir(), "autocomplete-quit-whitespace-")); + try { + fs.writeFileSync(path.join(baseDir, "copy-target.ts"), "export {};\n"); + const provider = new CombinedAutocompleteProvider( + [{ name: "quit", description: "Quit", allowArgs: false }], + baseDir, + ); + const line = "/quit "; + const result = await provider.getSuggestions([line], 0, line.length); + + expect(result).toBeNull(); + } finally { + fs.rmSync(baseDir, { recursive: true, force: true }); + } + }); + it("treats @ file-reference tokens as literal text inside slash command arguments without completions", async () => { const baseDir = fs.mkdtempSync(path.join(os.tmpdir(), "autocomplete-rename-args-")); try { diff --git a/packages/tui/test/editor.test.ts b/packages/tui/test/editor.test.ts index 1e0b27dfb..a4eb03ded 100644 --- a/packages/tui/test/editor.test.ts +++ b/packages/tui/test/editor.test.ts @@ -1,4 +1,7 @@ -import { afterEach, describe, expect, it } from "bun:test"; +import { afterEach, describe, expect, it, vi } from "bun:test"; +import * as fs from "node:fs/promises"; +import * as os from "node:os"; +import * as path from "node:path"; import { stripVTControlCharacters } from "node:util"; import { CURSOR_MARKER } from "@oh-my-pi/pi-tui"; import { CombinedAutocompleteProvider } from "@oh-my-pi/pi-tui/autocomplete"; @@ -424,6 +427,46 @@ describe("Editor component", () => { expect(editor.getText()).toBe("/help "); expect(editor.isShowingAutocomplete()).toBe(false); }); + + it("does not open file autocomplete after tab-completing no-arg slash commands", async () => { + vi.useFakeTimers(); + const baseDir = await fs.mkdtemp(path.join(os.tmpdir(), "slash-tab-no-arg-")); + try { + await Bun.write(path.join(baseDir, "visible-file.ts"), "export {};\n"); + const editor = new Editor(defaultEditorTheme); + editor.setAutocompleteProvider( + new CombinedAutocompleteProvider([{ name: "quit", description: "Quit", allowArgs: false }], baseDir), + ); + + let nextUpdate = Promise.withResolvers(); + editor.onAutocompleteUpdate = () => nextUpdate.resolve(); + editor.handleInput("/"); + await nextUpdate.promise; + + nextUpdate = Promise.withResolvers(); + editor.onAutocompleteUpdate = () => nextUpdate.resolve(); + editor.handleInput("q"); + vi.advanceTimersByTime(100); + await nextUpdate.promise; + + const chainedUpdates = Promise.withResolvers(); + let updateCount = 0; + editor.onAutocompleteUpdate = () => { + updateCount += 1; + if (updateCount === 2) { + chainedUpdates.resolve(); + } + }; + editor.handleInput(" "); + await chainedUpdates.promise; + + expect(editor.getText()).toBe("/quit "); + expect(editor.isShowingAutocomplete()).toBe(false); + } finally { + vi.useRealTimers(); + await fs.rm(baseDir, { recursive: true, force: true }); + } + }); }); describe("Unicode text editing behavior", () => { diff --git a/packages/tui/test/process-terminal-render.test.ts b/packages/tui/test/process-terminal-render.test.ts index d95bf6243..d9af2d757 100644 --- a/packages/tui/test/process-terminal-render.test.ts +++ b/packages/tui/test/process-terminal-render.test.ts @@ -71,4 +71,21 @@ describe("ProcessTerminal geometry reflow through the renderer", () => { expect(harness.terminal.columns).toBe(160); expect(harness.probe.last).toBe(160); }); + + it("recovers the full height when the grow-back report carries colon subparameters (#4748)", async () => { + // iOS soft keyboard under tmux-over-SSH: an in-band shrink lands (keyboard + // up), then the keyboard is dismissed and the grow-back report arrives with + // a spec-permitted `:`-subparameter and no accompanying OS resize. The + // parser must ignore the subparameter — dropping the report leaves the + // viewport pinned at the keyboard-present height. + harness = createProcessTerminalRenderHarness(100, 30); + await harness.feed("\x1b[?2048;1$y"); + await harness.inBand(15, 100, 300, 1000); // keyboard appears: 30 -> 15 rows + expect(harness.terminal.rows).toBe(15); + + await harness.feed("\x1b[48;30;100;600;1000:0t"); // keyboard dismissed: grow back + + expect(harness.terminal.rows).toBe(30); + expect(harness.terminal.columns).toBe(100); + }); }); diff --git a/packages/tui/test/streaming-scrollback-defer.test.ts b/packages/tui/test/streaming-scrollback-defer.test.ts index 6dfd7427a..7098aceaf 100644 --- a/packages/tui/test/streaming-scrollback-defer.test.ts +++ b/packages/tui/test/streaming-scrollback-defer.test.ts @@ -80,6 +80,23 @@ class CommittedRowsProbe extends SeamLineList implements NativeScrollbackCommitt } } +/** + * Extends the compose-time probe with the raw wire: every value the engine + * pushes through `setNativeScrollbackCommittedRows`, in arrival order — + * including the post-emit publish that lands *between* frames. Guards that + * run between frames (a controller deciding whether a displaceable block may + * still be retracted) read exactly this last value; if it lags the emit by + * one frame they retract rows that already entered immutable history. + */ +class CommittedRowsWireProbe extends CommittedRowsProbe { + received: number[] = []; + + override setNativeScrollbackCommittedRows(rows: number): void { + this.received.push(rows); + super.setNativeScrollbackCommittedRows(rows); + } +} + async function settle(term: VirtualTerminal): Promise { const nextTick = Promise.withResolvers(); process.nextTick(nextTick.resolve); @@ -604,6 +621,130 @@ describe("streaming scrollback — visual record", () => { } }); + it("publishes the post-emit committed count between frames — never one frame stale", async () => { + if (process.platform === "win32") return; + const term = new VirtualTerminal(40, 8); + overrideProbe(term, undefined); + const tui = new TUI(term); + const probe = new CommittedRowsWireProbe([]); + + try { + tui.addChild(probe); + tui.start(); + await settle(term); + + // Nothing has scrolled: the between-frames claim is 0 — no phantom rows. + expect(probe.received.at(-1)).toBe(0); + + // One frame grows past the viewport; its emit scrolls rows into + // native scrollback. No further render is requested — whatever the + // probe last received IS the claim a between-frames guard consults. + probe.setLines(rows("hist-", 20)); + tui.requestRender(); + await settle(term); + + // Compose ran before the emit advanced the boundary, so this frame's + // render() saw the pre-emit count. The emit must then push the fresh + // count: with compose-only propagation the last received value would + // still equal the stale compose view, and a guard would retract rows + // that just became immutable — stranding an orphaned copy in history. + const composeView = probe.committedRowsAtRender.at(-1)!; + const betweenFrames = probe.received.at(-1)!; + expect(betweenFrames).toBeGreaterThan(composeView); + // The fresh claim is the truth: exactly the rows above the window + // (tape = committed history rows + the 8-row grid). + expect(betweenFrames).toBe(tape(term).length - 8); + + // A frame that commits nothing must restate the boundary verbatim on + // every push — compose feed and post-emit publish alike. No regress, + // no phantom advance. + const wireLength = probe.received.length; + tui.requestRender(); + await settle(term); + expect(probe.received.length).toBeGreaterThan(wireLength); + for (const value of probe.received.slice(wireLength)) { + expect(value).toBe(betweenFrames); + } + } finally { + tui.stop(); + } + }); + + it("publishes the post-emit committed count on the full-paint replay path too", async () => { + if (process.platform === "win32") return; + const term = new VirtualTerminal(40, 8); + overrideProbe(term, undefined); + const tui = new TUI(term); + // Content taller than the viewport before the first paint: the initial + // frame takes the full-paint path, whose replay commits (frame - height) + // rows in one shot on a separate exit from the ordinary update emit. + const probe = new CommittedRowsWireProbe(rows("hist-", 20)); + + try { + tui.addChild(probe); + tui.start(); + await settle(term); + + // Compose fed the pre-emit count (0); the replay committed 12 rows. + // The full-paint return must publish the fresh count too — leaving + // it stale until the next compose is the same one-frame lag. + const composeView = probe.committedRowsAtRender.at(-1)!; + const betweenFrames = probe.received.at(-1)!; + expect(betweenFrames).toBeGreaterThan(composeView); + expect(betweenFrames).toBe(tape(term).length - 8); + } finally { + tui.stop(); + } + }); + + it("clamps each child's committed-count feed to its own extent", async () => { + if (process.platform === "win32") return; + const term = new VirtualTerminal(40, 8); + overrideProbe(term, undefined); + const tui = new TUI(term); + // A short header above a tall overflowing body: the engine's committed + // boundary sails past the header's 2-row extent. Both feeds are in the + // child's own coordinates and must saturate at what the child actually + // contributed — an unclamped count would make rows the header appends + // LATER read as already-committed, exempting them from ever painting. + const header = new CommittedRowsWireProbe(rows("hdr-", 2)); + const body = new CommittedRowsWireProbe([]); + + try { + tui.addChild(header); + tui.addChild(body); + tui.start(); + await settle(term); + + body.setLines(rows("body-", 20)); + tui.requestRender(); + await settle(term); + + // 22-row frame in an 8-row window: 14 rows committed, the boundary + // 12 rows past the header. Post-emit publish: the header's claim + // saturates at its own extent; the body receives the remainder in + // its own coordinates (boundary minus its start offset). + expect(tape(term).length).toBe(22); + expect(header.received.at(-1)).toBe(2); + expect(Math.max(...header.received)).toBe(2); + expect(body.received.at(-1)).toBe(12); + // Post-emit freshness holds per child in the multi-child layout: + // the body's compose view was still pre-emit, the publish delivered + // the advanced count. + expect(body.received.at(-1)!).toBeGreaterThan(body.committedRowsAtRender.at(-1)!); + + // The compose-time feed clamps identically: an idle frame restates + // each child's saturated claim on every push — never more. + tui.requestRender(); + await settle(term); + expect(header.received.at(-1)).toBe(2); + expect(Math.max(...header.received)).toBe(2); + expect(body.received.at(-1)).toBe(12); + } finally { + tui.stop(); + } + }); + it("never re-anchors a re-laying-out live block mid-run, repairs once at finalize", async () => { if (process.platform === "win32") return; const term = new VirtualTerminal(20, 4); diff --git a/packages/tui/test/terminal-appearance.test.ts b/packages/tui/test/terminal-appearance.test.ts index 88438c841..ea1067c1b 100644 --- a/packages/tui/test/terminal-appearance.test.ts +++ b/packages/tui/test/terminal-appearance.test.ts @@ -144,6 +144,24 @@ describe("ProcessTerminal OSC 11 appearance detection", () => { terminal.stop(); }); + it("replays already detected OSC 11 appearance to late subscribers", () => { + const { terminal } = setupTerminal(); + + process.stdin.emit("data", "\x1b]11;rgb:ffff/ffff/ffff\x07"); + process.stdin.emit("data", "\x1b[?1;2c"); + + const appearances: string[] = []; + terminal.onAppearanceChange(a => appearances.push(a)); + const detected = terminal.appearance; + + // Stop before asserting: a failing expect must not leak a live terminal + // (stdin listeners, kitty push) into subsequent tests. + terminal.stop(); + + expect(detected).toBe("light"); + expect(appearances).toEqual(["light"]); + }); + it("2-digit hex OSC 11 response is correctly normalized", () => { const { terminal } = setupTerminal(); @@ -582,6 +600,30 @@ describe("ProcessTerminal DECRQM + in-band resize (DEC 2026/2048)", () => { terminal.stop(); }); + it("applies a grow-back report whose fields carry colon subparameters (#4748)", () => { + // iOS soft keyboard dismissed under tmux-over-SSH: the pane grows back and + // the terminal reports the restored geometry in-band with a spec-permitted + // `:`-subparameter appended to a field. Mode 2048 allows subparameters on + // any field and requires clients to IGNORE them — dropping the whole + // report instead pins `rows` at the keyboard-present height, because no + // OS resize event accompanies the report to reconcile cached geometry. + Object.defineProperty(process.stdout, "columns", { value: 100, configurable: true }); + Object.defineProperty(process.stdout, "rows", { value: 40, configurable: true }); + const { terminal, received, resizeCount } = setup(); + process.stdin.emit("data", "\x1b[?2048;1$y"); // in-band active + process.stdin.emit("data", "\x1b[48;20;100;400;1000t"); // keyboard appears: shrink + expect(terminal.rows).toBe(20); + expect(resizeCount()).toBe(1); + + process.stdin.emit("data", "\x1b[48;40;100;800;1000:0t"); // keyboard dismissed: grow back + + expect(terminal.rows).toBe(40); + expect(terminal.columns).toBe(100); + expect(resizeCount()).toBe(2); + expect(received).toEqual([]); + terminal.stop(); + }); + it("tracks OS geometry on resize when the post-resize in-band report is missed", () => { // Real terminals always fire SIGWINCH (process.stdout dims refresh first), // but the matching DEC 2048 report can be dropped or arrive malformed. The @@ -658,6 +700,29 @@ describe("ProcessTerminal DECRQM + in-band resize (DEC 2026/2048)", () => { terminal.stop(); }); + it("reassembles a split grow-back report with colon subparameters without dropping or leaking it", () => { + // Same grow-back report, fragmented by the StdinBuffer flush window right + // after the subparameter colon. The partial pattern must accept `:`, or + // the prefix is rejected as garbage, the report never applies, and the + // `0t` tail leaks into the editor as literal keystrokes. + vi.useFakeTimers(); + Object.defineProperty(process.stdout, "columns", { value: 100, configurable: true }); + Object.defineProperty(process.stdout, "rows", { value: 40, configurable: true }); + const { terminal, received, resizeCount } = setup(); + process.stdin.emit("data", "\x1b[?2048;1$y"); // in-band active + process.stdin.emit("data", "\x1b[48;20;100;400;1000t"); // keyboard appears: shrink + expect(terminal.rows).toBe(20); + + process.stdin.emit("data", "\x1b[48;40;100;800;1000:"); + vi.advanceTimersByTime(50); // flush window elapses mid-report + process.stdin.emit("data", "0t"); + + expect(received).toEqual([]); + expect(terminal.rows).toBe(40); + expect(resizeCount()).toBe(2); + terminal.stop(); + }); + it("forwards a split report fragment as one escape sequence instead of leaking bare characters", () => { // The reported symptom: a fragment like `8;125;1156;1125t` (the tail of // `\x1b[48;125;1156;1125t`, missing a field) appeared as literal text in the diff --git a/packages/utils/package.json b/packages/utils/package.json index 9a805d384..ccaabb9a1 100644 --- a/packages/utils/package.json +++ b/packages/utils/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/pi-utils", - "version": "16.3.12", + "version": "16.3.14", "description": "Shared utilities for pi packages", "homepage": "https://omp.sh", "author": "Can Boluk", diff --git a/packages/utils/src/dirs.ts b/packages/utils/src/dirs.ts index c6fb9d9df..d0b63af69 100644 --- a/packages/utils/src/dirs.ts +++ b/packages/utils/src/dirs.ts @@ -22,6 +22,9 @@ export const APP_NAME: string = "omp"; /** Config directory name (e.g. ".omp") */ export const CONFIG_DIR_NAME: string = ".omp"; +/** Ordered main settings filenames: canonical write target first, legacy-compatible YAML fallback second. */ +export const MAIN_CONFIG_FILENAMES = ["config.yml", "config.yaml"] as const; + /** Version (e.g. "1.0.0") */ export const VERSION: string = version; diff --git a/packages/wire/package.json b/packages/wire/package.json index 94fda90cf..1245b516e 100644 --- a/packages/wire/package.json +++ b/packages/wire/package.json @@ -1,7 +1,7 @@ { "type": "module", "name": "@oh-my-pi/pi-wire", - "version": "16.3.12", + "version": "16.3.14", "description": "Shared wire protocol types for Oh My Pi packages", "homepage": "https://omp.sh", "author": "Can Boluk",