From 20bd4ab97b7105dd7e6b8ff99ed390effe011921 Mon Sep 17 00:00:00 2001 From: can1357 Date: Wed, 19 Aug 2026 01:42:50 +0200 Subject: [PATCH] chore(changelog): normalized [Unreleased] sections after merges and added missing entries - Repaired union-merge artifacts in packages/coding-agent/CHANGELOG.md (duplicated 17.3.6/17.3.7 blocks; promoted the new entries back to [Unreleased]) - Added missing [Unreleased] entries for PRs #8833, #8866, #8879, #8903, #8905, #8915, #8916, #8917, #8920, #8923, #8928, #8929, #8937 --- packages/agent/CHANGELOG.md | 1 + packages/ai/CHANGELOG.md | 18 ++-- packages/catalog/CHANGELOG.md | 14 +-- packages/coding-agent/CHANGELOG.md | 147 +++++++++-------------------- 4 files changed, 62 insertions(+), 118 deletions(-) diff --git a/packages/agent/CHANGELOG.md b/packages/agent/CHANGELOG.md index 1adc59e8c..e5d6c5da5 100644 --- a/packages/agent/CHANGELOG.md +++ b/packages/agent/CHANGELOG.md @@ -6,6 +6,7 @@ - Fixed `/compact` (and automatic compaction) resurrecting pre-`/clear` conversation turns: `prepareCompaction` now honors the latest `reset_boundary`, so a compaction after an in-place `/clear` only summarizes messages created after the reset ([#8718](https://github.com/can1357/oh-my-pi/issues/8718)). - Hardened compaction summarization against prompt injection: conversation history and previous summaries are now treated as untrusted, and embedded ``/`` boundary tags are neutralized before prompt assembly ([#8727](https://github.com/can1357/oh-my-pi/pull/8727) by [@koopmannleon19977-cmyk](https://github.com/koopmannleon19977-cmyk)). +- Compaction summarization input is now bounded to the summary model's context (windowed fold for oversized spans) and deterministic context-overflow 400s are no longer retried up to the full retry budget; artifact ids containing `503` no longer misclassify hard 400s as transient. ## [17.3.5] - 2026-08-16 diff --git a/packages/ai/CHANGELOG.md b/packages/ai/CHANGELOG.md index 0c85d1228..cb40dada2 100644 --- a/packages/ai/CHANGELOG.md +++ b/packages/ai/CHANGELOG.md @@ -2,29 +2,29 @@ ## [Unreleased] +### Changed + +- Fixed Gemini thought summaries occasionally leaking a raw `` ```thinking `` / `` ``````thinking `` fence delimiter into the reasoning block, so it no longer shows up as fence spam in the thinking display or persisted transcripts ([#8719](https://github.com/can1357/oh-my-pi/issues/8719)). +- Fixed the OpenCode Go login prompting for an "OpenCode Zen API key": the shared login flow now names the provider you selected, so connecting OpenCode Go asks for an OpenCode Go key (the `opencode.ai/auth` console is still shared, as documented upstream) ([#8738](https://github.com/can1357/oh-my-pi/issues/8738)). +- Fixed Anthropic-compatible endpoints with strict prompt validation (e.g. Z.AI GLM `api.z.ai/api/anthropic`, which rejects the whole request with `400 code 1213 "The prompt parameter was not received normally"`) failing sessions once a tool returned empty output on a vision-capable model: empty successful `tool_result` blocks now encode as `content: ""` instead of `content: []`, which both the official API and strict compatible endpoints accept. +- Fixed `retry.usageReservePct` (Reserve Margin) ignoring Claude Fable/Mythos weekly tier usage until it hit 100%, so a Fable model kept serving turns past the configured reserve; reserve health now honors the mapped tier row while credential-wide hard blocks still require confirmed exhaustion ([#8773](https://github.com/can1357/oh-my-pi/issues/8773)). +- Fixed `cursor-agent` streams stalling with "Provider stream stalled while waiting for the next event" when Cursor asked the client to approve a hosted WebFetch / web search (reproduced on `cursor-grok-4.6-xhigh` after "I'll fetch the page…"). Those `interaction_query` frames — including the newer WebFetch field 9 this proto did not name — were dropped, so the server waited forever and the idle watchdog aborted a live connection. Permission queries are now answered; hosted search/fetch is approved, unnamed permission fields get an `approved` reply on the same field number, and prompts this client cannot serve are rejected so the turn can continue. + ### Fixed - Fixed thinking effort selections being ignored for local Qwen 3.8+ models on llama.cpp and vLLM: the Qwen chat-completions dialects only toggled `enable_thinking`, so the chat template always reasoned at its `xhigh` default no matter which level was selected. The encoder now routes the requested effort onto the template's `reasoning_effort` kwarg (`chat_template_kwargs` for both Qwen dialects, plus the top-level field newer llama.cpp builds map natively). - - Fixed OpenAI Completions, Amazon Bedrock, and Cursor providers ignoring `onPayload` replacement payloads. The hook now transforms the actual request body sent upstream on these providers, matching the Anthropic/Gemini/OpenAI Responses replacement contract. `devin-agent` still does not fire the hook (its payload is a protobuf object). - Fixed Codex requests failing outright when the signed-in ChatGPT account is not entitled to the requested model; the exact model denial is now classified as an account-policy error so credential rotation can reach an entitled sibling account - Fixed Perplexity email-OTP login after its verification response renamed the encrypted session token from `token` to `challenge_token`. - -### Fixed - - Cloud Code Assist Gemini 3.6/3.7 Flash requests at `minimal` now send `thinkingLevel: LOW` on the aliased `-low` SKU instead of `MINIMAL`, which the API rejects with HTTP 400. - Answer Cursor `interaction_query` permission gates (hosted web search, Exa, unnamed field-9 WebFetch) so the Run RPC continues instead of sitting silent until the 300s idle watchdog. +- Fixed provider tool calls arriving with flattened array argument paths (e.g. Gemini's `questions[0].id`) being stripped and rejected by argument validation; well-formed flattened paths are now rebuilt into the nested arrays the tool schema expects ([#8886](https://github.com/can1357/oh-my-pi/issues/8886)). ## [17.3.7] - 2026-08-17 ### Changed - Send the `omp/` User-Agent on xAI chat (`xai` and `xai-oauth`) unless the request already set its own. -- Fixed Gemini thought summaries occasionally leaking a raw `` ```thinking `` / `` ``````thinking `` fence delimiter into the reasoning block, so it no longer shows up as fence spam in the thinking display or persisted transcripts ([#8719](https://github.com/can1357/oh-my-pi/issues/8719)). -- Fixed the OpenCode Go login prompting for an "OpenCode Zen API key": the shared login flow now names the provider you selected, so connecting OpenCode Go asks for an OpenCode Go key (the `opencode.ai/auth` console is still shared, as documented upstream) ([#8738](https://github.com/can1357/oh-my-pi/issues/8738)). -- Fixed Anthropic-compatible endpoints with strict prompt validation (e.g. Z.AI GLM `api.z.ai/api/anthropic`, which rejects the whole request with `400 code 1213 "The prompt parameter was not received normally"`) failing sessions once a tool returned empty output on a vision-capable model: empty successful `tool_result` blocks now encode as `content: ""` instead of `content: []`, which both the official API and strict compatible endpoints accept. -- Fixed `retry.usageReservePct` (Reserve Margin) ignoring Claude Fable/Mythos weekly tier usage until it hit 100%, so a Fable model kept serving turns past the configured reserve; reserve health now honors the mapped tier row while credential-wide hard blocks still require confirmed exhaustion ([#8773](https://github.com/can1357/oh-my-pi/issues/8773)). -- Fixed `cursor-agent` streams stalling with "Provider stream stalled while waiting for the next event" when Cursor asked the client to approve a hosted WebFetch / web search (reproduced on `cursor-grok-4.6-xhigh` after "I'll fetch the page…"). Those `interaction_query` frames — including the newer WebFetch field 9 this proto did not name — were dropped, so the server waited forever and the idle watchdog aborted a live connection. Permission queries are now answered; hosted search/fetch is approved, unnamed permission fields get an `approved` reply on the same field number, and prompts this client cannot serve are rejected so the turn can continue. ## [17.3.5] - 2026-08-16 diff --git a/packages/catalog/CHANGELOG.md b/packages/catalog/CHANGELOG.md index 689b21481..390d0adee 100644 --- a/packages/catalog/CHANGELOG.md +++ b/packages/catalog/CHANGELOG.md @@ -2,17 +2,18 @@ ## [Unreleased] +### Added + +- Regenerated the Cursor agent protobufs to model hosted WebFetch permission queries (`interaction_query` / `interaction_response` field 9) and the matching `ToolCall` variant (field 37). + ### Fixed - Fixed local Qwen 3.8+ models (llama.cpp, vLLM, loopback custom providers) exposing the generic `minimal..high` thinking ladder instead of the chat template's real `low`/`medium`/`xhigh` `reasoning_effort` tiers. The derived metadata now marks thinking as mandatory (the official 3.8 template raises on `enable_thinking: false`), vLLM-served Qwen routes through the `chat_template_kwargs` dialect (top-level `enable_thinking` is ignored by vLLM), and vLLM discovery lights up the reasoning dial for Qwen 3.8+ ids its `/v1/models` endpoint reports as non-reasoning. - Fixed `deepseek-v4-pro-0813` surfacing from Alibaba Token Plan discovery with `contextWindow`/`maxTokens` of `null`. The dated DeepSeek V4 Pro snapshot was missing from `ALIBABA_TOKEN_PLAN_DISCOVERED_MODEL_LIMITS`, so unlike its `deepseek-v4-flash-0731` sibling it fell through to unknown limits ([#8847](https://github.com/can1357/oh-my-pi/issues/8847)). - -### Fixed - - Cloud Code Assist Gemini 3.6/3.7 Flash no longer maps user `minimal` to wire `thinkingLevel: MINIMAL` when that effort is aliased onto the `-low` SKU. The request now sends `LOW`, which those SKUs accept. -### Added - -- Regenerated the Cursor agent protobufs to model hosted WebFetch permission queries (`interaction_query` / `interaction_response` field 9) and the matching `ToolCall` variant (field 37). +- Fixed SuperGrok (`xai-oauth`) Grok 4.6 hiding the thinking-level picker: the Responses effort-capable allowlist now includes `grok-4.6`, so `/model` can select the documented `low`/`medium`/`high`/`xhigh` ladder (`max` is rejected by api.x.ai). +- Marked CoreWeave runtime discovery as authoritative so stale bundled model ids that the endpoint no longer serves stop appearing as selectable models. +- ChatGPT Codex discovery that advertises only worker `-wm` SKUs now also registers the plain model route, so a configured `openai-codex/` keeps resolving instead of fuzzy-falling-back to the `-wm` SKU some accounts reject. ## [17.3.6] - 2026-08-17 @@ -23,7 +24,6 @@ ### Fixed - Raised the GPT-5.6 Sol/Terra/Luna context window on the Codex transport (openai-codex) from 372K to 1M tokens: OpenAI enabled the 1M window for subscription Codex on 2026-08-16, but the Codex model registry still reports the stale 272,000, so discovery now floors these SKUs at 1,000,000 instead of trusting the reported value ([openai/codex#38917](https://github.com/openai/codex/issues/38917)). -- Fixed SuperGrok (`xai-oauth`) Grok 4.6 hiding the thinking-level picker: the Responses effort-capable allowlist now includes `grok-4.6`, so `/model` can select the documented `low`/`medium`/`high`/`xhigh` ladder (`max` is rejected by api.x.ai). ## [17.3.5] - 2026-08-16 diff --git a/packages/coding-agent/CHANGELOG.md b/packages/coding-agent/CHANGELOG.md index b801e13b5..fdd5ae758 100644 --- a/packages/coding-agent/CHANGELOG.md +++ b/packages/coding-agent/CHANGELOG.md @@ -5,120 +5,67 @@ ### Added - Added `providers.cacheRetention` setting (`/settings` → Providers → Protocol) to control prompt-cache retention per request: `auto` keeps the provider default (Anthropic: 5m entries with idle keep-alive refreshes), `short` forces 5m, `long` restores 1h TTLs where supported and disables the keep-alive refresh loop, `none` disables prompt caching. + +### Changed + +- The `read` tool now materializes a local text file once per invocation instead of once per consumer. A ranged read of a file within the snapshot cap previously cost four opens and three UTF-8 decodes — an 8KiB binary sniff, a streaming scan for the rendered window, a whole-file read for bracket context, and another whole-file read to hash the snapshot — with two of those readers separately normalizing line endings; whole-file reads under the structural summarizer paid a fifth read. Byte counts and truncation boundaries are now measured on the buffered bytes, so they stay exact for content that is not valid UTF-8. Files above the snapshot cap keep streaming, since nothing on that path wants the whole file. Raw reads, which skip the tree-sitter parse that documented the old cost, no longer pay for it. + ### Fixed - Fixed Claude Code marketplace plugins ignoring the `enabledPlugins` switch in `~/.claude/settings.json` and `.claude/settings(.local).json`: a plugin turned off for a project no longer loads there, and a local-scope install enabled for a project loads even when its recorded `projectPath` is a different directory -### Fixed - - Fixed task and eval subagents discovering newly added agent definitions while resolving their role aliases from stale startup settings. Subagent preflight now atomically reloads persisted settings before agent discovery while preserving live runtime overrides. -### Fixed - - Fixed images returned by tools mounted under `xd://` rendering only as file links instead of inline terminal graphics. -### Fixed - - Resume Cursor idle-stall turns after completed MCP/todo tool results. The watchdog already closes the Connect stream, so unmarked blocks no longer need the `exec-resolved` marker to continue. -### Fixed - - Fixed the Web Search Provider Order settings summary showing providers excluded from web search ([#8884](https://github.com/can1357/oh-my-pi/issues/8884)). -### Fixed - - Fixed subagents aborting when external thinking exposes `think` as the required prelude before their remaining tools become callable ([#8909](https://github.com/can1357/oh-my-pi/pull/8909) by [@olegpulatov](https://github.com/olegpulatov)). -### Fixed - - Fixed session-title generation ignoring user `/skill:` invocations, so titles now see the skill name and args instead of only later assistant text. - -## [17.3.7] - 2026-08-17 - -### Changed - -- Send the `omp/` User-Agent on xAI chat (`xai` and `xai-oauth`) unless the request already set its own ([#8840](https://github.com/can1357/oh-my-pi/pull/8840) by [@Jaaneek](https://github.com/Jaaneek)). - -## [17.3.6] - 2026-08-17 - -### Added - -- Added `ExtensionAPI.registerFileWriteFallback(handler)` and `ExtensionAPI.registerFileDeleteFallback(handler)`, letting an extension supply a fallback writer or deleter that is consulted when a native `write`, `edit`, or `apply_patch` byte-write or unlink is denied with a permission error (`EPERM`/`EACCES`/`EROFS`) — for hosts that embed the agent inside a sandbox that denies direct filesystem access but exposes a privileged channel. The brokered path is symlink-resolved so a handler's allowlist sees the real destination, a destination that cannot be resolved is not brokered at all, and `req.sessionId` names the session that issued the mutation so a handler sharing the process-wide registry can enforce policy per session. See [`docs/extensions.md`](../../docs/extensions.md). -### Fixed - -- Extension bare imports of workspace members now resolve inside installed git-dependency monorepo plugins (the walk recognizes `workspaces` roots; installed node_modules copies still shadow members) - -### Changed - -- Updated the default model for XAI_API_KEY (xai) and SuperGrok OAuth (xai-oauth) to grok-4.6. Automatic model selection continues to prefer paid xai/grok-4.6 when only XAI_API_KEY is set, with xai-oauth/grok-4.6 still available explicitly. - -### Fixed - -- Fixed `omp stats` and `/stats` dashboards being unreachable from container hosts by accepting an explicit `--host` bind address while preserving the `127.0.0.1` default. -### Fixed - -- Fixed `omp completions ` hanging after writing shell completion scripts to stdout by invoking `postmortem.quit(0)` upon completion. Prevents lingering event loop handles (such as background timers or sockets loaded when inspecting command metadata) from pinning the process and blocking tools like `chezmoi`. -### Fixed - -- Fixed `omp --smoke-test` recursively deleting unrelated directories in `os.tmpdir()` (tmux/ssh sockets, editor state, build trees). The smoke broker now keeps its runtime dir under a private parent, and the dead-scope reclaim refuses any root that is not the `daemons` container and only prunes entries named like a 16-hex daemon scope key ([#8721](https://github.com/can1357/oh-my-pi/issues/8721)). -### Fixed - -- Fixed high CPU during multi-subagent / workflowz / orchestrate sessions: each live tool block (streaming args, a running partial tool, or a `task` subagent) armed its own 80ms spinner `setInterval` driving `requestComponentRender`, so N concurrent live blocks created N unsynchronized repaint timers that kept the render scheduler awake near-continuously. The per-block timers are now consolidated into a single shared spinner ticker that repaints every live block in one coalesced frame per glyph step, independent of block count ([#8731](https://github.com/can1357/oh-my-pi/issues/8731)). -### Fixed - -- Fixed `omp update` writing to the PATH launcher instead of the running binary on binary-only releases (major bumps or `omp.dist: "binary"`): a foreign symlink — e.g. an admin symlink into a shared install — now resolves to its real binary in every distribution channel, avoiding an `EACCES` on a root-owned link directory or a split-brain copy that shadows the shared install. Package-manager launchers keep their deliberate in-place takeover. ([#8732](https://github.com/can1357/oh-my-pi/issues/8732)) -### Fixed - -- Added `.css` to the built-in Biome server `fileTypes` so CSS files route through Biome's linter/asserter by default instead of requiring a full per-project `fileTypes` override. ([#8741](https://github.com/can1357/oh-my-pi/pull/8741)) -### Fixed - -- Fixed memory extraction sending its instructions, few-shot examples, and the user's message as a single user turn, which caused small local models to echo the examples instead of extracting facts; instructions now travel as a system turn and the raw text as the user turn -### Fixed - -- Fixed local title generation stopping on a stop string that appeared in the prompt instead of the generated tokens -### Fixed - -- Fixed the Subagents HUD role display and restored generated task labels by keeping spawn handles separate from UI descriptions. - -## [17.3.7] - 2026-08-17 - -### Changed - -- Send the `omp/` User-Agent on xAI chat (`xai` and `xai-oauth`) unless the request already set its own ([#8840](https://github.com/can1357/oh-my-pi/pull/8840) by [@Jaaneek](https://github.com/Jaaneek)). -### Changed - -- The `read` tool now materializes a local text file once per invocation instead of once per consumer. A ranged read of a file within the snapshot cap previously cost four opens and three UTF-8 decodes — an 8KiB binary sniff, a streaming scan for the rendered window, a whole-file read for bracket context, and another whole-file read to hash the snapshot — with two of those readers separately normalizing line endings; whole-file reads under the structural summarizer paid a fifth read. Byte counts and truncation boundaries are now measured on the buffered bytes, so they stay exact for content that is not valid UTF-8. Files above the snapshot cap keep streaming, since nothing on that path wants the whole file. Raw reads, which skip the tree-sitter parse that dominates other reads, are 30-45% faster (2.7MB file: 10.3ms to 5.7ms). - -### Fixed - -- Fixed a whole-file `read` of a file with a UTF-8 BOM minting a hashline tag hashed from BOM-bearing text. Because the patcher's live read strips the BOM, the next edit to that file only applied through stale-hash recovery and reported that the file had changed externally when it had not. - -## [17.3.6] - 2026-08-17 - -### Added - -- Added `ExtensionAPI.registerFileWriteFallback(handler)` and `ExtensionAPI.registerFileDeleteFallback(handler)`, letting an extension supply a fallback writer or deleter that is consulted when a native `write`, `edit`, or `apply_patch` byte-write or unlink is denied with a permission error (`EPERM`/`EACCES`/`EROFS`) — for hosts that embed the agent inside a sandbox that denies direct filesystem access but exposes a privileged channel. The brokered path is symlink-resolved so a handler's allowlist sees the real destination, a destination that cannot be resolved is not brokered at all, and `req.sessionId` names the session that issued the mutation so a handler sharing the process-wide registry can enforce policy per session. See [`docs/extensions.md`](../../docs/extensions.md). - -### Changed - -- Updated the default model for XAI_API_KEY (xai) and SuperGrok OAuth (xai-oauth) to grok-4.6. Automatic model selection continues to prefer paid xai/grok-4.6 when only XAI_API_KEY is set, with xai-oauth/grok-4.6 still available explicitly. - -### Fixed - -- Fixed `omp stats` and `/stats` dashboards being unreachable from container hosts by accepting an explicit `--host` bind address while preserving the `127.0.0.1` default. -### Fixed - - Fixed destructive `rm` escaping the critical-pattern approval check when anything separates the flags from the target, so `rm -rf -- /`, `rm --recursive --force /` and `rm -rf --no-preserve-root /` are now classified critical like `rm -rf /`. `--no-preserve-root` is treated as critical wherever it appears, since it is what defeats coreutils' own refusal to recurse on `/`. -### Fixed - - Fixed thinking-loop aborts (`AIError.Flag.ThinkingLoop`) walking `retry.fallbackChains` and switching to another model family on attempt 1, so a healthy planning turn on Grok 4.6 (SuperGrok / Cursor OAuth) no longer gets replaced by whatever the chain lists next. The loop guard now re-samples the same model with its `thinking-loop-redirect` notice, and no longer parks the model selector on a fallback cooldown. ([#8760](https://github.com/can1357/oh-my-pi/issues/8760)) -### Fixed - - Fixed the clipboard image-paste keybind attaching Finder's generated file icon instead of the copied image on macOS. Current Finder `Cmd+C` pasteboards advertise both a `public.file-url` and a generated 1024x1024 icon bitmap, so `arboard::get_image()` succeeded with the icon and `InputController.handleImagePaste` attached it before the file-URL branch was ever reached. The handler now probes `readMacFileUrlsFromClipboard()` before the bitmap representation, so an image file URL wins over the co-advertised icon; pure bitmap pasteboards (screenshots, browser copies) and non-image file URLs still fall through to the image/text paths ([#8769](https://github.com/can1357/oh-my-pi/issues/8769)). -### Fixed - - Fixed the Home Manager module (`programs.omp.settings`) breaking every launch on macOS with `Failed to acquire native file lock … Permission denied (os error 13)`. The declared config is now copied into `~/.omp/agent/config.yml` as a writable file via `home.activation` instead of a read-only `/nix/store` symlink, so OMP can acquire its config lock and persist runtime changes; `home-manager switch` still reapplies the declared settings ([#8775](https://github.com/can1357/oh-my-pi/issues/8775)). -### Fixed - - Fixed OpenCode MCP servers 401ing when config used OpenCode's `{env:VAR}`/`{file:path}` substitution (e.g. `Bearer {env:MCP_KEY}` headers); the OpenCode loader now expands those tokens the way OpenCode does instead of only `${VAR}` ([#8778](https://github.com/can1357/oh-my-pi/issues/8778)). -### Fixed - - Fixed `omp update` leaking Bun's raw `fetch()` error ("pass `verbose: true` in the second argument to fetch()") when a proxy environment variable (`HTTPS_PROXY`, `ALL_PROXY`, …) uses an unsupported scheme such as SOCKS; the update check now reports an actionable message naming the offending variable and the http/https proxy requirement ([#8784](https://github.com/can1357/oh-my-pi/issues/8784)). - Fixed worker subprocesses (memory embeddings, tiny-model titles, TTS/STT, JS eval, browser relay, LSP mux, daemon broker) running with their cwd pinned to the CLI install directory. They share the agent's foreground process group, and terminal cwd heuristics such as kitty's `new_tab_with_cwd` pick the newest process in that group, so new terminal tabs opened in `~/.bun/install/global/node_modules/@oh-my-pi/pi-coding-agent/dist` while any worker was alive. Workers now spawn with the absolute host entry and inherit the agent's cwd. +- Preserved MCP `ImageContent` tool-result blocks so vision-capable models and the TUI can inspect returned images instead of receiving only a text placeholder ([#8687](https://github.com/can1357/oh-my-pi/issues/8687)). +- Fixed a whole-file `read` of a file with a UTF-8 BOM minting a hashline tag hashed from BOM-bearing text. Because the patcher's live read strips the BOM, the next edit to that file only applied through stale-hash recovery and reported that the file had changed externally when it had not. +- Extension bare imports of workspace members now resolve inside installed git-dependency monorepo plugins (the walk recognizes `workspaces` roots; installed node_modules copies still shadow members) +- Fixed `omp completions ` hanging after writing shell completion scripts to stdout by invoking `postmortem.quit(0)` upon completion. Prevents lingering event loop handles (such as background timers or sockets loaded when inspecting command metadata) from pinning the process and blocking tools like `chezmoi`. +- Fixed `omp --smoke-test` recursively deleting unrelated directories in `os.tmpdir()` (tmux/ssh sockets, editor state, build trees). The smoke broker now keeps its runtime dir under a private parent, and the dead-scope reclaim refuses any root that is not the `daemons` container and only prunes entries named like a 16-hex daemon scope key ([#8721](https://github.com/can1357/oh-my-pi/issues/8721)). +- Fixed high CPU during multi-subagent / workflowz / orchestrate sessions: each live tool block (streaming args, a running partial tool, or a `task` subagent) armed its own 80ms spinner `setInterval` driving `requestComponentRender`, so N concurrent live blocks created N unsynchronized repaint timers that kept the render scheduler awake near-continuously. The per-block timers are now consolidated into a single shared spinner ticker that repaints every live block in one coalesced frame per glyph step, independent of block count ([#8731](https://github.com/can1357/oh-my-pi/issues/8731)). +- Fixed `omp update` writing to the PATH launcher instead of the running binary on binary-only releases (major bumps or `omp.dist: "binary"`): a foreign symlink — e.g. an admin symlink into a shared install — now resolves to its real binary in every distribution channel, avoiding an `EACCES` on a root-owned link directory or a split-brain copy that shadows the shared install. Package-manager launchers keep their deliberate in-place takeover. ([#8732](https://github.com/can1357/oh-my-pi/issues/8732)) +- Added `.css` to the built-in Biome server `fileTypes` so CSS files route through Biome's linter/asserter by default instead of requiring a full per-project `fileTypes` override. ([#8741](https://github.com/can1357/oh-my-pi/pull/8741)) +- Fixed memory extraction sending its instructions, few-shot examples, and the user's message as a single user turn, which caused small local models to echo the examples instead of extracting facts; instructions now travel as a system turn and the raw text as the user turn +- Fixed local title generation stopping on a stop string that appeared in the prompt instead of the generated tokens +- Fixed the Subagents HUD role display and restored generated task labels by keeping spawn handles separate from UI descriptions. +- Fixed `models.yml` custom-model providers declaring `auth: oauth` being rejected by validation with "apiKey is required", which forced a dummy `apiKey` that then shadowed the broker's OAuth tokens ([#8937](https://github.com/can1357/oh-my-pi/pull/8937) by [@usr-bin-roygbiv](https://github.com/usr-bin-roygbiv)). +- Provider-qualified model selectors (e.g. `anthropic/claude-opus-5`) now fail closed when the named provider is unavailable instead of silently re-binding to OpenRouter's same-named flat id and billing the aggregator ([#8832](https://github.com/can1357/oh-my-pi/issues/8832)). +- Fixed PlanYolo plan approval dropping all MCP tools: the post-handoff tool restore now accounts for MCP discovery that completed while planning instead of restoring a pre-discovery snapshot. +- Fixed parallel `web_search` calls hanging forever past the 60s timeout when the shared headless-browser daemon or page died mid-setup; browser fallback setup and teardown are now abort-protected ([#8865](https://github.com/can1357/oh-my-pi/issues/8865)). +- Fixed extension-package `.mcp.json` `${VAR}` env placeholders (stdio env/command/args/cwd, HTTP url/headers/oauth) reaching MCP servers unexpanded. +- Advisor blocker advisories raised inside the post-interrupt immune window now wake a new turn instead of parking as asides until the next user prompt. +- The exit banner only advertises `omp --resume ` when the session was actually written to disk, so the printed command no longer fails for sessions that ended before persistence ([#8860](https://github.com/can1357/oh-my-pi/issues/8860)). +- Fixed terminals that deliver Shift+Enter as a bare LF (or the legacy CSI `13;2~` form) getting a plain switch instead of summarize-and-switch in the `/tree` selector ([#8821](https://github.com/can1357/oh-my-pi/issues/8821)). +- Fixed OMP panicking at startup when the host environment contains a non-UTF-8 variable value; such entries are now skipped when copying the host environment into the shell ([#8925](https://github.com/can1357/oh-my-pi/issues/8925)). + +## [17.3.7] - 2026-08-17 + +### Changed + +- Send the `omp/` User-Agent on xAI chat (`xai` and `xai-oauth`) unless the request already set its own ([#8840](https://github.com/can1357/oh-my-pi/pull/8840) by [@Jaaneek](https://github.com/Jaaneek)). + +## [17.3.6] - 2026-08-17 + +### Added + +- Added `ExtensionAPI.registerFileWriteFallback(handler)` and `ExtensionAPI.registerFileDeleteFallback(handler)`, letting an extension supply a fallback writer or deleter that is consulted when a native `write`, `edit`, or `apply_patch` byte-write or unlink is denied with a permission error (`EPERM`/`EACCES`/`EROFS`) — for hosts that embed the agent inside a sandbox that denies direct filesystem access but exposes a privileged channel. The brokered path is symlink-resolved so a handler's allowlist sees the real destination, a destination that cannot be resolved is not brokered at all, and `req.sessionId` names the session that issued the mutation so a handler sharing the process-wide registry can enforce policy per session. See [`docs/extensions.md`](../../docs/extensions.md). + +### Changed + +- Updated the default model for XAI_API_KEY (xai) and SuperGrok OAuth (xai-oauth) to grok-4.6. Automatic model selection continues to prefer paid xai/grok-4.6 when only XAI_API_KEY is set, with xai-oauth/grok-4.6 still available explicitly. + +### Fixed + +- Fixed `omp stats` and `/stats` dashboards being unreachable from container hosts by accepting an explicit `--host` bind address while preserving the `127.0.0.1` default. ## [17.3.5] - 2026-08-16 @@ -188,10 +135,6 @@ - Fixed external thinking being offered on xAI reasoning-only Responses models (grok-4 family) that reject `reasoning.effort`, where the private scratchpad ran alongside native reasoning instead of replacing it. - Fixed the extension tool-call handler timeout rendering outside a titled section in `/settings` by registering its Extensions group on the Tools tab. -### Fixed - -- Preserved MCP `ImageContent` tool-result blocks so vision-capable models and the TUI can inspect returned images instead of receiving only a text placeholder ([#8687](https://github.com/can1357/oh-my-pi/issues/8687)). - ## [17.3.4] - 2026-08-14 ### Changed