Repository navigation
fix(muse): failed Muse text generation explains Muse's error category - #29861
Conversation
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ApprovabilityVerdict: Approved at Macroscope's review found this PR approvable — Muse text-generation failures now show category-specific guidance for quota, sign-in and model errors while keeping raw provider messages out of the displayed detail. The change leaves generation defaults, permissions and successful requests unchanged. You can add or adjust custom eligibility rules. Learn more. |
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dismissing prior approval to re-evaluate 03d02e6
|
Note 🤖 Claude Opus 5.5 responding on behalf of saphid Reworked this to follow the documented error rule in docs/internals/effect-services.md "Errors". Muse's raw failure text no longer appears in the
Commit: 03d02e6 |
|
Note Written by Verified in a real dev environmentverified the signed-out muse failure on linux at head created a git project through the running app's cli and ran real codex the signed-out muse picker is disabled because muse returns no model catalog. enabled muse and added its documented nearby checks passed: a codex follow-up returned blacksmith passed all 9 focused muse text-generation tests and Not verifiedlive quota exhaustion, model-request failures, unknown failure categories, and successful authenticated muse generation: this vm has no signed-in muse account. the focused tests cover structured categories, but that is not live-provider proof. other generation entry points and native desktop, ios, and android clients were not exercised. both cold web startups took several minutes to render, but neither exceeded the ten-minute limit. |
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Note 🤖 Claude Opus 5.5 responding on behalf of saphid Delta since your verification at 03d02e6: one commit, 419e93b, answering CodeRabbit. A credential-proxy 429 quota refusal that has no |
Dismissing prior approval to re-evaluate 419e93b
Dismissing prior approval to re-evaluate 0a74022
Cherry-picked from 5fae2b5 (pingdotgg#35107) by Julius Marminge. Co-Authored-By: Julius Marminge <julius0216@outlook.com> Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Dismissing prior approval to re-evaluate b5a9f79
Resolve CommandPalette.browser.test.tsx to main's version: the branch's only change there was a cherry-pick of pingdotgg#35107, which main already has. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Dismissing prior approval to re-evaluate d8abe34
**v0.1.11-nightly.20261118.4877** · 24 changes from 13 contributors - **Paused tasks.** Running a paused one-time task manually no longer deletes it. - **Claude subagent results.** Resumed Claude subagents retain their earlier results instead of replacing them. ### Plus 20 more bug fixes. <details> <summary>Full changelog</summary> ## What's Changed * fix(server): agents read a Linear issue once when given its link by @StiensWout in pingdotgg/t3code#38425 * fix(server): All issues pages skip the signed-in account read by @StiensWout in pingdotgg/t3code#38431 * fix(server): recover JSON-stringified delegate targets by @MINIlaolao in pingdotgg/t3code#38436 * fix(mobile): retrying an issue filter only reloads repositories that failed by @StiensWout in pingdotgg/t3code#38423 * fix(web): pairing links never use the desktop app's t3code:// address by @juliusmarminge in pingdotgg/t3code#38442 * chore: vouch mvanhorn by @t3dotgg in pingdotgg/t3code#38455 * docs: vouched contributors can open small feature PRs by @t3dotgg in pingdotgg/t3code#38456 * fix(server): preview pages loaded through a service worker identify as Chrome, not HeadlessChrome by @Sean-R-Wilson in pingdotgg/t3code#38445 * fix(web): switching back to a thread no longer jumps through its HTML renders by @pouyarezvani in pingdotgg/t3code#38458 * fix(mobile): A document whose upload failed can be retried in the composer by @mwolson in pingdotgg/t3code#38443 * fix(server): running a paused one-time task by hand no longer deletes it by @juliusmarminge in pingdotgg/t3code#38463 * fix(server): codex text generation identifies as T3 Code by @juliusmarminge in pingdotgg/t3code#38464 * fix(server): review diff previews reuse the cached repository layout by @amv146 in pingdotgg/t3code#38451 * fix(web): Paste as Text pastes text when the clipboard also holds an image by @juliusmarminge in pingdotgg/t3code#38475 * fix(web): opening a file in the preview browser shows its current contents by @juliusmarminge in pingdotgg/t3code#38470 * fix(github): page closing pull requests past the first twenty on issues by @StiensWout in pingdotgg/t3code#38487 * fix(web): older comments no longer reappear after refreshing an issue by @StiensWout in pingdotgg/t3code#38489 * fix(web): typing into a simulator no longer answers a waiting question by @ipanasenko in pingdotgg/t3code#38476 * fix(server): keep-awake logs a hold only once acquired and reports why the inhibitor failed by @juliusmarminge in pingdotgg/t3code#38505 * fix(server): write the Windows service task as UTF-16LE so non-ASCII paths survive by @juliusmarminge in pingdotgg/t3code#38494 * fix(muse): failed Muse text generation explains Muse's error category by @saphid in pingdotgg/t3code#29861 * fix(server): imported Claude threads show prompts, not Claude Code markup by @dominikzabcik in pingdotgg/t3code#33403 * fix(server): a resumed Claude subagent keeps its earlier result by @ahalekelly in pingdotgg/t3code#34663 * fix(server): a preview navigation request issued before its frame exists no longer crashes the server by @Isaac-b-ux in pingdotgg/t3code#30946 ## New Contributors * @Sean-R-Wilson made their first contribution in pingdotgg/t3code#38445 **Full Changelog**: pingdotgg/t3code@v0.1.11-nightly.20261118.4864...v0.1.11-nightly.20261118.4877 </details> Upstream release: https://github.com/pingdotgg/t3code/releases/tag/v0.1.11-nightly.20261118.4877
…pingdotgg#29861) Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Julius Marminge <julius0216@outlook.com> (cherry picked from commit a1f8dcf)


Problem
When Muse fails while generating a commit message, PR text, branch name or thread title, T3 Code gives generic login advice even for quota exhaustion or a rejected model request.
Change
Keep the structured turn error and map it to fixed usage-limit, sign-in, model-request or generic detail. Quota takes precedence because Muse can label quota exhaustion as
modelError. When the completion has onlyreason, preserve and classify it through the same path. The raw message stays incauseand is never copied into the user-facing detail.The mapper also handles main's
MuseSdkErrorwrapper. Its private failure carrier usesSchema.TaggedErrorandSchema.is, with no diagnostic suppression. Startup and transport failures retain the generic detail.Scope
Two Muse files, plus the upstream command-palette test-isolation fix described below. The upstream test fix is included because this branch’s Test Web 2 failed on leaked localStorage; it is the five-line correction already merged on main. Related #29823, #29748 and #29800 handle other providers; #29159 changes Muse branch-name sanitizing rather than failure classification.
Verification
Current revision: b5a9f79. The two Muse files are unchanged from b822aac. The added five-line test-only repair is Julius Marminge’s 5fae2b5 (#35107), cherry-picked unchanged. It clears a last-used project leaked between browser test files, which can hide the contextual “New thread” row and shift the expected selection. Main had the same assertion failure before that upstream fix.
Linux Boat validation on the exact candidate tree
f0f13c299c3e81f49d4bc3acaf0fb76e8e606669, with all three source hashes matched locally and remotely:data-highlightedassertion as hosted CI (exit 1). The fixed test with the identical seed passes 5/5; the unmodified fixed test also passes 5/5.CI=true vp run check:changedexits 0: format, lint, knip, ten affected-package typechecks and 16 tests (Muse 9, CommandPalette 5, AgentContextBudget 2).Sidebar.drag.browser.test.tsx:55, then the file and full shard passed. Its cause remains unconfirmed; those runs are not verification of this tree.Fresh independent read-only review by GPT-6.1 Sol, Codex through T3 Code, runtime-reported high effort, found no actionable findings in the full three-file candidate or five-line repair (task
9d582029-3a13-480c-9d95-de7f6203c659). The reviewer inspected source and supplied logs without rerunning tests. The preceding Muse-only tree5d0f5fdwas also independently reviewed by Claude Opus 5.5; high effort was configured, but that reviewer reported low runtime effort.Hosted required checks on this head are satisfied: five passed and three skipped; Test Web 2 and the aggregate Check passed. Macroscope approved this exact head on eligibility (no code objects reviewed). CodeRabbit remains paused at the preceding revision; the fresh independent review above covers the current candidate. All review threads are resolved. Final maintainer approval remains. Runtime evidence uses the SDK host fake; no live failed Muse turn was exercised.
Original implementation and reason repair: Claude Opus 5.5 in Claude Code through T3 Code. Tagged-error revision and publication: GPT-6.1 Sol in Codex through T3 Code. Upstream test fix: Julius Marminge; integration and validation: Claude Opus 5.5. Fresh independent review: GPT-6.1 Sol/high in Codex through T3 Code.