- Added comprehensive unit tests for `TranscriptContainer` to verify uncommitted block tracking.
- Created integration tests ensuring `AssistantMessageComponent` correctly streams thinking and answer content into scrollback.
- Added tests verifying that expanded tool evaluation output records rows correctly without duplication after settling.
- Updated `AssistantMessageComponent` test suite to cover table streaming scenarios in the unsettled tail.
- Replaced commit-based stability checks with a unified `isTranscriptBlockFinalized` tracking mechanism.
- Removed deprecated provisional rendering configuration and flags across tool and renderer interfaces.
- Standardized native scrollback boundary logic to pin at the first unfinalized block using settled row verification.
- Updated and refactored test suites to validate block finalization and settled row boundaries instead of deprecated commit stability methods.
- Introduced an automated retry recovery system to track, manage, and persist recovered error states within agent sessions.
- Enabled compact transcript rendering for recovered auto-retry errors by removing heuristic commit machinery.
- Improved raw read tracking and provenance in the ReadTool to support refined file snapshot recording and hashline editing.
- Excluded recovered assistant messages from default model context and updated event controllers to handle retry recovery life cycles.
The cmux release regression tests install spies on CmuxSocketClient.prototype. Bun keeps those spies active across later browser-* files unless the file restores them explicitly, so browser-cmux-socket.test could stop exercising the real socket client depending on order.
Restore all Bun test mocks in afterEach after draining any test tabs, preserving mocked cleanup while preventing cross-file pollution.
Fixes#4499
Codex review of #4502 flagged that a bare `.catch(() => undefined)`
neutralizes the unhandledRejection but leaves the affected `runInTab`
call blocked inside `runCmuxCode` until timeout when the in-flight
code does not make another cmux socket request (e.g. `await
wait(60_000)`). `releaseTab` was signaling the run only by rejecting
an orphaned promise.
Wire the tab-close event all the way into the cmux run body:
- `PendingRun` gains a `closeAc: AbortController` that `releaseTab`
aborts BEFORE calling `pending.reject`. `wait(...)` (via
`waitForBrowserRun` -> `untilAborted`), in-flight cmux socket calls
(via CmuxTab's `#request` -> `untilAborted`), and facade proxies
(via `bindBrowserRunFacade`) all consume the composed signal, so
the run body unwinds within a microtask instead of blocking to its
own timeout.
- `runInTabWithSnapshot`'s cmux branch composes `closeAc.signal` into
the run's signal (`AbortSignal.any([opts.signal, closeAc.signal])`)
and now publishes `runCmuxCode(...)`'s outcome to the shared
`promise` via `.then(resolve, reject)` and returns `await promise`.
Both branches thus await the same promise, so `pending.reject`
always has an attached handler (removing the original crash) AND
the caller sees `Tab "..." was closed` immediately instead of
waiting on the run's timeout.
- Drop the defensive `promise.catch(() => undefined)` — the promise
is now actively consumed on both backends.
The new regression test adds a second case that exercises the
reviewer's exact scenario (`await wait(60_000);`) and asserts:
1. `pending.closeAc.signal.aborted` flips from `false` to `true`
across `releaseTab`, with the tab-close error as its reason.
2. The awaited `runInTab(...)` rejects with `Tab "..." was closed`.
3. No `unhandledRejection` fires.
Verified locally by temporarily removing `closeAc.abort(...)` in
`releaseTab` — the new assertions fail; restoring it makes them pass.
Fixes#4499
The cmux branch of `runInTabWithSnapshot` awaits `runCmuxCode(...)`
directly and never awaits/`.catch`es the `Promise.withResolvers()`
promise it stashes on `tab.pending`. When `releaseTab` walks pending
runs and calls `pending.reject(new ToolError("Tab ... was closed"))`
(a sibling subagent's `browser close --all`, session-scoped reap, etc.),
that orphaned promise had zero handlers and Bun surfaced the rejection
as `unhandledRejection`, which the CLI's top-level handler treats as
fatal — killing every other tab and subagent sharing the process, not
just the affected run.
Attach a no-op `.catch(() => undefined)` to the promise immediately
after creation. Inert for the worker branch (which still awaits the
same promise via `raceWithTimeout`, and attaching a second handler is
safe) and neutralizes the orphan on the cmux branch.
Adds a regression test that drives real `acquireBrowser` /
`acquireTab` / `runInTab` / `releaseTab` against a mocked
`CmuxSocketClient`, races `releaseTab` against an in-flight cmux run,
and asserts no `unhandledRejection` fires.
Fixes#4499
Skipped the workflow notice on raw selectors so bounded raw reads stay byte-for-byte, and taught resolveToolSearchScope to request pathOnly resolution so ast_grep/ast_edit scope-only calls no longer trip the inline-content cap on large artifacts.
Fixes#4482
When the error body already advertises OAuth endpoints, `/mcp add` and `/mcp reauth` use `authResult.oauth` directly and skip `discoverOAuthEndpoints`, so scopes advertised only in the RFC 9728 protected-resource metadata document never reach the grant.
Add exported `fetchResourceMetadataScopes(url, opts?)` that fetches the metadata doc and returns `scopes_supported` / `scopes` / `scope`. Hoist the shared `readMetadataScopes` reader out of `discoverOAuthEndpoints`. At all three call sites (wizard, `/mcp add`, `/mcp reauth`), when `oauth` is populated from the JSON body but `oauth.scopes` is empty and `authResult.resourceMetadataUrl` was advertised, fetch the metadata and merge scopes onto `oauth`.
Regression tests cover the resource-metadata fetch and its failure/empty-doc paths.
Refs #4467
shortenPath() the artifact.path leaked into the read-tool 'Artifact storage' and 'Unbounded raw read blocked' notices so $HOME never appears verbatim; details.resolvedPath keeps the absolute path for tooling.
Fixes#4482
Bash URL expansion and search/grep only need sourcePath; they now request pathOnly resolution so large artifacts stay usable for search/copy workflows while unbounded content materialization stays blocked.
Fixes#4482
Resolved artifact:// reads to backing files before selector handling, streamed bounded reads, and blocked unbounded raw reads for large artifacts with recovery guidance.
Fixes#4482
Extended ONNX Runtime CUDA sidecar repair to PI_TINY_DEVICE=auto and gpu on Linux x64, matching the CUDA-capable diagnostics path.
Added focused regression coverage for both generic accelerated settings.
Refs #4475
Deferred ONNX Runtime CUDA sidecar repair failure into the runtime metadata so loadTransformersRuntime keeps loading and loadPipelineWithDeviceFallback still gets its CUDA→CPU retry when NuGet is offline or the ort install script is unavailable. The failure surfaces through the CUDA diagnostics helper instead of hard-erroring the tiny worker.
Refs #4475
Downloaded missing ONNX Runtime CUDA provider sidecars when the compiled tiny-model side runtime is used with PI_TINY_DEVICE=cuda.
Preserved actionable CUDA worker diagnostics in tiny-models text output and added focused regression coverage.
Fixes#4475
@DylanBohlender's follow-up caught that MCPAuthorizationLinkPrompt.render
still ignored `width` and emitted `Copy URL: <full URL>` as one composed
row. On any viewport narrower than the row (~272 columns for a
Linear-shaped authorize URL), TUI#prepareLine's
`truncateToWidth(..., Ellipsis.Omit)` silently clipped the trailing
`code_challenge_method=S256` — the exact #4418 fingerprint reappearing
inside the remote-safety fix. A remote user on a narrow terminal
copying the rendered line would lose the S256 method again; the local
shortcut below cannot help them (localhost isn't reachable), and the
OSC 52 clipboard staged full URL isn't visible in their local browser.
Component-level fix: honor `width` in render.
- New `wrapUrlRows(label, url, width)` helper.
- When `label + " " + url` fits in `width`, emit one inline row.
- Otherwise emit the label on its own row and slice the URL into
chunks of `width - indent`, each on its own row.
- Floors the effective width at 16 columns so degenerately narrow
terminals still emit every character; browsers strip whitespace
when a multi-row selection is pasted into the address bar, so the
reassembled URL is byte-identical.
- `render(width)` now uses the helper for both the primary `Copy URL:`
row and the additive `Local shortcut (this machine only):` row.
Regression tests in
`packages/coding-agent/test/modes/controllers/mcp-authorization-link.test.ts`:
- Wide viewport (1000 cols): inline `Copy URL: <url>` layout preserved.
- Narrow viewport (80 cols) + Linear-shaped URL: every row's visible
width ≤ 80, and the chunks reassemble byte-for-byte to the URL —
explicitly asserting the trailing `code_challenge_method=S256`
survives.
- Launch shortcut also wrapped at 80 cols; every row fits.
- Degenerate viewport (4 cols): URL still reconstructs exactly; the
16-col floor governs chunk width.
- Full URL remains the primary target even when a launch URL is
present, and the shortcut row is omitted when launchUrl is absent
or identical to the full URL.
`/mcp reauth` and `/mcp add` consume `authResult.oauth` directly when the JSON error body carries endpoints, skipping `discoverOAuthEndpoints`. Leaving `oauth.scopes` empty there meant a challenge-only `scope="…"` still yielded a scope-less grant even though `AuthDetectionResult.scopes` recorded it.
Merge `challengeScopes` into the returned `OAuthEndpoints` inside `analyzeAuthError` so every consumer sees the same scope. Added a regression test where the JSON body advertises endpoints without `scopes` and the challenge is the only source.
Refs #4467
MCP servers such as JIT gateways advertise required scopes via the RFC 6750 `WWW-Authenticate` challenge (`scope="..."`) and via RFC 9728 protected-resource metadata (`scopes_supported` / `scopes` / `scope`), then reject follow-up requests with `insufficient_scope` when a bearer token was issued without them. OMP's discovery only picked up `scopes_supported` from the auth-server metadata document, so `/mcp reauth`, `/mcp add`, and the MCP add wizard silently minted scope-less tokens.
- Extract `scope`/`scopes` from the WWW-Authenticate challenge into a new `AuthDetectionResult.scopes` field via `extractOAuthChallengeScopes`.
- Thread a `protectedScopes` option through `discoverOAuthEndpoints` and its recursion; capture `scopes_supported`/`scopes`/`scope` off resource-metadata documents; use those scopes when the auth-server metadata omits them.
- Pass `authResult.scopes` from `analyzeAuthError` into every discovery call site (`/mcp reauth`, `/mcp add`, MCP add wizard).
- Add regression tests for insufficient_scope + resource_metadata, resource-metadata `scopes_supported` passthrough, and challenge-scope threading.
Fixes#4467
Collapsed duplicate byte-identical non-empty LSP text edits before overlap validation so rename_file accepts idempotent server output while still rejecting real conflicts.
Fixes#4458
An intermediate commit whose net effect is already on HEAD (redundant
change, or 3-way merged to HEAD by "theirs == ours") stopped the
sequencer with "The previous cherry-pick is now empty" and was
treated as a hard conflict. mergeTaskBranches aborted the whole range,
marked the branch failed, and dropped every remaining non-overlapping
commit.
Add cherryPick.skip and cherryPick.isEmptyError to the git namespace,
then in mergeTaskBranches' catch classify the failure before aborting:
loop --skip while the error stderr matches the "now empty" phrase so
consecutive empties advance the sequencer; fall through to abort/fail
on the first non-empty error (genuine conflict with unmerged files).
Fixes#4438
Codex review flagged that advertising `launchUrl`
(http://localhost:<omp-port>/launch) as the visible `Copy URL:` breaks
SSH/WSL/headless users: their local browser resolves the URL against
the local machine (no OMP listening) and fails before ever hitting the
provider. On terminals without OSC 8 support, they lose the manual
`/login <redirect>` path entirely.
Every OAuth-facing surface now shows the full authorization URL as the
primary copy target and offers `launchUrl` as an additional "Local
shortcut (this machine only)" line for wide-terminal local users who
want the truncation-safe convenience:
- MCPAuthorizationLinkPrompt renders `Copy URL:` with the full URL and
appends the local-shortcut row only when `launchUrl` differs. OSC 52
clipboard staging in the MCP onAuth handler switches to the full URL
(OSC 52 is a wire-level protocol — the terminal writes to the
caller's LOCAL clipboard even when OMP is on a remote SSH box).
- LoginDialogComponent.showAuth, selector-controller onAuth,
setup-wizard sign-in, and the auth-broker CLI mirror the pattern:
full URL first, launchUrl as an optional local shortcut.
- Setup wizard uses `wrapTextWithAnsi`, not truncation, so the RFC
7636 §4.3 downgrade bug that motivated launchUrl is unreachable
through it; still surfaces launchUrl for wide-terminal convenience.
Regression tests in
`packages/coding-agent/test/modes/controllers/mcp-authorization-link.test.ts`
now assert:
- Full URL is the primary `Copy URL:` line so SSH sessions can complete.
- launchUrl still appears beneath as `Local shortcut (this machine only): …`
when it differs from the full URL.
- No shortcut row when launchUrl is absent OR equals the full URL.
Two independent defects broke /mcp reauth against S256-only providers on
Windows boxes whose PATH no longer references System32:
1. openPath spawned bare rundll32 and swallowed the
`Executable not found in $PATH` throw with a bare `catch {}`, so the MCP
controller's outer try/catch was dead and the transcript unconditionally
claimed "Opening browser automatically...".
2. TUI#prepareLine silently truncates any composed row wider than the
viewport. MCPAuthorizationLinkPrompt rendered `Copy URL: <full URL>` as a
single ~271-column line whose trailing parameter is
code_challenge_method=S256. On the reporter's 270-col terminal the cut
landed inside that parameter, dropping the method while keeping
code_challenge — which RFC 7636 §4.3 treats as plain PKCE, which Linear
correctly rejects with "The plain PKCE method is not allowed. Use S256
instead."
OAuthCallbackFlow now hosts a `GET /launch` route on the same loopback
callback server it already runs; the route 302-redirects to the pending
authorization URL and is advertised as `OAuthAuthInfo.launchUrl` — a
~30-char copy target no viewport can meaningfully truncate. The MCP OAuth
fallback, /login, setup wizard, auth-broker CLI, and login-dialog all
prefer the launch URL for the visible copy target, keep the full URL in
the OSC 8 hyperlink for click-through, and the MCP flow additionally
stages the copy target on the clipboard via OSC 52 (same pattern the
setup wizard uses).
openPath now resolves rundll32.exe through %SystemRoot%\System32 (with a
C:\Windows fallback when SystemRoot is unset) and logs both synchronous
spawn throws and non-zero exits via the shared logger, so silent
misconfigurations show up in ~/.omp/logs/omp.*.log. The dead try/catch
around openPath in the MCP controller is removed.
Fixes#4418
Filtered TypeScript project-resolution diagnostics for files with no root marker ancestor while preserving syntax diagnostics. Added regression coverage for Bun scratch files outside project roots.\n\nFixes #4401
Added an embedText projection for remember() so stored transcripts can remain readable while embeddings, FTS indexing, and embedding-model rebuilds use marker-free text. Updated coding-agent retention to pass the marker-free projection and strip retained protocol markers from recall display.
Fixes#4395
Auto-retain now slices the transcript after the last retained user turn before storing an episode, preventing cumulative duplicate session transcripts in mnemopi banks.
Fixes#4396
- Added test case to verify that missing provider limits in sibling accounts are correctly marked as not reported.
- Verified that the usage breakdown correctly distinguishes between reported and unreported sibling provider limits.
Four tightly-scoped hot-path fixes covering the highest-impact items in the
reporter's CPU profile (13.1 s profiled / 30 s window):
1. `event-controller.ts:handleEvent` no longer fires a blanket
`statusLine.invalidate() + ui.requestRender()` before every session event.
The pre-render was a leftover from #4145 when `updateEditorTopBorder()`
still eagerly rebuilt the border; the lazy provider added in #4145 made
it redundant. It fired on every `message_update`/`tool_execution_update`
during streaming — the pre-render's frame ran while the handler was
awaiting, then the handler's own `requestRender` scheduled a second
identical frame. Every handler that mutates visible state already calls
`requestRender()`.
2. `shimmer.ts:shimmerSegments` iterates the segment string in place instead
of building a code-point array with `Array.from(seg.text)` every animation
frame. Runs of same-tier chars are emitted via a single `slice` per run
rather than accumulating into `runBuf`. Surrogate pairs stay atomic — the
code-point index still advances by 1 per emoji. Microbench over 30k
frames: 45 ms → 18 ms (2.53x), allocation rate down from ~N-per-frame to
a handful per frame. New tests cover mixed BMP+surrogate and all-emoji
inputs. `Array.from` was the #1 self-time hotspot in the reporter's
profile at 10.2%.
3. `Markdown.setText` gains an equality guard mirroring `Text.setText`
(returns `false` when `text === #text`). Providers re-emit identical text
on ticks with no delta (throttled frames, reconciled tool-execution
updates); each of those now short-circuits instead of dropping
`#cachedLines` and forcing a full lex + wrap on the accumulated paragraph
(the reporter's #3 hotspot at 8.4%). New test asserts render-reference
stability + return-value semantics.
4. `SPINNER_RENDER_INTERVAL_MS` aligned with `SPINNER_GLYPH_ADVANCE_MS`
(both 80 ms). The previous 33 ms cadence emitted ~2.4 paints per glyph
step; the differential-output dedup only skips the write, not the
compose walk. Visually identical (glyph advance was already 12.5fps),
halves paints during tool execution.
Skipped (out of scope for a bug fix, deserve dedicated PRs):
- Freezing streaming prefix on single `\n` boundaries — correctness-bound
to `\n\n` block separators (CommonMark loose-list continuation).
- Compose-phase idle gate + adaptive-backpressure moving-average — need a
component-level dirty flag; the 200 ms cap in `#scheduleRender` was set
for a reason (#4145 tail-latency guard).
Tests updated: two IRC-expiry tests in event-controller-message-start.test.ts
that were asserting the pre-render's second `requestRender` call now expect
one.
Fixes#4353
can1357