feat(review): surface review failures as a PR comment

[lapa2112]

What

A review that throws before it can post (e.g. a model context-size overflow on a large PR) previously died silently after the webhook's 200 — nothing appeared on the PR, and the only trace was in the worker logs. The primary/escalation pass can fail non-retryably, and the run terminated before postReview with no feedback.

This wraps the review run so any failure posts a ⚠️ Mimir review failed comment carrying the error reason. The notice is:

• idempotent — stored under a separate …#<n>::failed key, so a re-run updates one notice instead of stacking;
• non-destructive — its own <!-- mimir-review-failed --> marker keeps it distinct from the summary comment, so a failed run never clobbers a prior good review.

A structured review failed event is also logged for wrangler tail.

Why

Turning a silent no-op into an actionable, re-runnable notice: when a review fails, the maintainer sees why on the PR and can /review again, instead of assuming the bot is broken or the PR was skipped.

Changes

• lib/post-review.ts: new postReviewFailure() (update-or-create on a ::failed key).
• workflows/review-pr.ts: thin run wrapper around the review body that catches, logs, and posts the failure notice; re-resolving the (cached) installation client to post.

Test

• typecheck, lint, and the existing suite (91 tests) pass.

[mimir-code-assist]

Mimir review — 🟢 No blocking issues

Well-structured PR adding a post-review failure notification mechanism. postReviewFailure creates/updates a PR comment when a review crashes, keyed separately (::failed) so it never overwrites a good summary. postReview retracts any stale failure notice on success. The code is well-tested (7 test cases for the new function, 2 for the retraction integration). No blocking issues found; the existing minor findings (no integration test for the error wrapper, stale store keys) remain valid.

Findings: 0 critical · 0 major · 2 minor · 0 nit

_{💰 Review cost: $0.0136 — primary deepseek/deepseek-v4-flash $0.0136}

[gemini-code-assist]

Code Review

This pull request introduces error handling and failure reporting for the PR review workflow. It adds a postReviewFailure function to surface failures (such as model context-size overflows) as PR comments, and wraps the main workflow execution in a try-catch block to prevent silent failures. The review feedback is highly constructive and suggests cleaning up stale failure comments upon a successful run, sanitizing error messages to prevent markdown formatting issues, and rethrowing the caught error so the workflow platform can accurately track execution failures.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.