[voltcrash]

Problem

CodeRabbit's reply to a review request says “Review triggered.” before editing the same comment with the result. PR watches currently wake the agent for both, although the acknowledgement has nothing to act on. Fixes #32275.

Change

Skip only coderabbitai replies (including the [bot]-suffixed login) whose first 1,000 characters contain CodeRabbit's reply marker and an “Action(s) performed” section starting with “Review triggered.” or “Full review triggered.” Still advance the watermark, so progress is consumed without spending a comment wake. Outcome edits, other replies, and other authors keep their existing behavior.

Scope and approval

This qualifies for CONTRIBUTING.md's very small, focused obvious-bug exception: the existing watch reports a queued-review acknowledgement as actionable feedback. The fix only filters that acknowledgement in the server evaluator; it adds no settings, contracts, client changes, or provider-specific behavior. The linked issue has not been maintainer-triaged. No changes to edit debouncing, wake labels, snippet rendering, or check recovery.

Verification

• The two focused CodeRabbit regressions fail on current upstream main because the acknowledgement appears in remarks. With the fix, all 42 tests in vp test run apps/server/src/orchestration-v2/pullRequestWatch.test.ts pass. Coverage includes both acknowledgement variants, finished/rate-limited outcome edits delivered once, mixed comments, same-time arrivals, other authors, other replies, both bot-login forms, and emoji summaries. The predicate still bounds its input to 1,000 characters.
• vp run check:changed passed: format, lint, both unused-code scans, generated GraphQL, three affected package typechecks, and 44 tests across the evaluator and agent-context budget files.
• Original runtime proof: replayed the actual CodeRabbit reply's edit history through the evaluator using Node, without starting a server:

                         upstream main      fixed
Review triggered.        remarks, wakes=1   no change, wakes=0
same acknowledgement     no change          no change
Review finished. edit    remarks, wakes=2   remarks, wakes=1
same outcome             no change          no change

Both versions advance the watermark to 00:24:57Z for the acknowledgement and 00:29:47Z for the outcome on 2026-11-07. The reactor already persists watermark-only changes.

Modernization (2026-11-13): rebased onto upstream main ff1eead6f411e02a23288236ad7b16d72ff8b147 without conflicts. Range-diff preserves all four non-merge commits; the final diff remains the evaluator filter and its two regression cases. The focused regressions and check:changed were rerun. A direct evaluator replay against current main and the rebased head reproduced the same wake counts above. The existing Chromium base/head scenario evidence is retained; a new browser/native pass was not run for this mechanical rebase.

Limitations: no live remote watch or paid provider turn was started. The filter intentionally depends on CodeRabbit's observed reply format; unrecognised formats retain the existing wake behavior. Wake wording is unchanged.

Original implementation reviewed with review-loop: 1 round with GPT-6.1 Sol on xhigh. Reviewer verdict: approved.

Created with GPT-6.1 Sol (xhigh) in Codex through T3 Code.

[macroscopeapp]

Approvability

Verdict: Approved at 3c5e9d5

Macroscope's review found this PR approvable — PR watching skips CodeRabbit’s review-trigger acknowledgements while continuing to track comments and report other events. The narrowly targeted filter reduces unnecessary agent wakes without introducing an obvious material risk.

^{You can add or adjust custom eligibility rules. Learn more.}

[coderabbitai]

Note

Reviews paused

It 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 reviews.auto_review.auto_pause_after_reviewed_commits setting.

• Last reviewed commit: 769d347
• Latest commit, not reviewed: 2507208

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• Resume reviews
• Trigger review

No actionable comments were generated in the recent review.

Walkthrough

The PR watch evaluator excludes qualifying CodeRabbit review acknowledgements from reported remarks while retaining fresh-remark tracking. Tests cover watermark advancement, later outcome edits, and other comments that remain actionable.

Changes: CodeRabbit review acknowledgements

Layer / File(s)	Summary
Detect and filter acknowledgements apps/server/src/orchestration-v2/pullRequestWatch.ts	The evaluator recognizes matching CodeRabbit bot comments and excludes fresh acknowledgements from reported remarks. Fresh comments still determine cursor tracking.
Test acknowledgement and outcome handling apps/server/src/orchestration-v2/pullRequestWatch.test.ts	Tests cover watermark advancement without a wake, later outcome edits, comments with the same timestamp, and replies that remain actionable.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Low

Possibly related issues

• PR watch reports an edited comment as a new comment on every edit #30342: This PR filters a specific CodeRabbit acknowledgement, but does not add general edit coalescing for other bots or comment states.

Suggested reviewers: saphid

Merge Risk: Minimal · previous review 769d3

No merge-blocking risk was established for the acknowledgement filter.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title clearly and concisely describes the main change: preventing CodeRabbit review acknowledgement comments from waking the server watch.
Linked Issues check		Issue #32275 is open and directly linked. evaluatePullRequestWatch filters only coderabbitai review acknowledgements that match the specified CodeRabbit format within the first 1,000 characters. T…
Out of Scope Changes check		The changes are limited to the pull-request watch evaluator and focused regression tests. The predicate and tests directly support issue #32275. No unrelated settings, contracts, client changes, provi…
Description check		The description follows the repository template closely. It covers Problem, Change, Scope, and Verification, and it names the linked issue (#32275). The Problem and Change sections state the bug and t…

Full details: Linked Issues check

Explanation

Issue #32275 is open and directly linked. evaluatePullRequestWatch filters only coderabbitai review acknowledgements that match the specified CodeRabbit format within the first 1,000 characters. The evaluator still advances the watermark. Outcome edits, other replies, other authors, and non-matching formats remain actionable. Focused tests cover both trigger variants, outcome delivery, bounds, and non-qualifying comments.

Full details: Out of Scope Changes check

Explanation

The changes are limited to the pull-request watch evaluator and focused regression tests. The predicate and tests directly support issue #32275. No unrelated settings, contracts, client changes, provider behavior, or other watch behavior changes are shown.

Full details: Description check

Explanation

The description follows the repository template closely. It covers Problem, Change, Scope, and Verification, and it names the linked issue (#32275). The Problem and Change sections state the bug and the fix clearly. The Scope section justifies the small obvious-bug exception and cites the linked issue. The Verification section lists focused regression results, a replay table with wake counts, the check:changed run, and stated limitations. The author also states that no live remote watch or paid provider turn ran, and no new browser pass ran for the rebase. Minor gaps: the template asks for a conventional commit title, and the description does not state one. The description also refers to a 42-test count while the PR objectives cite 23 tests and 25 evaluator tests, which is a discrepancy the author may want to reconcile. The description also includes a model-attribution line, which the template requests. Overall the description is mostly complete.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: b3fdf2b
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: 64b1a646-efc4-47df-b5b5-9e26529f7bf0

Commits

Reviewing files that changed from the base of the PR and between 7ec6c15 and 769d347.

Files selected for processing: 2

• apps/server/src/orchestration-v2/pullRequestWatch.test.ts
• apps/server/src/orchestration-v2/pullRequestWatch.ts

Limit details: You’ve used all 10 included reviews currently available.

[juliusmarminge]

Evidence

Real Chromium web client, 1280×800, light theme; real server and replay agent, fresh isolated state per commit. GitHub API responses are synthetic; no live forge writes. Normal two-minute watch sweeps. The initial check-success wake precedes the reproduction on both sides.

Create a synthetic project and replay thread, enable its pull-request watch, and let the startup check-success wake finish. Supply a matching coderabbitai “Review triggered.” acknowledgement, wait for a normal watch sweep, and capture the thread. Edit the same comment to “Review complete. No actionable findings.”, sweep and capture again. Leave the comment unchanged through another observed sweep and capture the final state. Run the identical sequence on both commits.

Scenario: CodeRabbit acknowledgement stays idle; edited outcome wakes once

Proven: the scenario fails on base b3fdf2b10f and passes on head 769d3470ca.

	Base b3fdf2b10f	Head 769d3470ca
Result	❌ failed ASSERTION_FAILED: expected {"ackWakes":1,"ackTurns":1,"outcomeWakes":2,"outcomeTurns":2,"unchangedWakes":2,"unchangedTurns":2,"outcomeDelivered":1,"unchangedDelivered":1} to equal {"ackWakes":0,"ackTurns":0,"outcomeWakes":1,"outcomeTurns":1,"unchangedWakes":1,"unchangedTurns":1,"outcomeDelivered":1,"unchangedDelivered":1}	✅ passed
watch started
acknowledgement
finished outcome
unchanged sweep
failure

Before recording	After recording
Full sequence	Full sequence

Scenario, Synthetic forge driver, Runner config patch, Run instructions, Before runtime transcript, After runtime transcript.

AI-assisted; verified with the tests and tools listed above.

Not shown:

• A live @coderabbitai review request and the CodeRabbit service were not exercised: forge writes were excluded from this task, so matching acknowledgement and edit responses were supplied by a synthetic GitHub endpoint.
• The rate-limited outcome, Full review triggered wording, and [bot]-suffixed author variant were not captured; this proof covers coderabbitai, Review triggered., and a finished outcome.

[coderabbitai]

Tip

For best results, initiate chat on the files or code changes.

@juliusmarminge Thanks for the evidence. The reported head run shows no wake for the acknowledgement, one wake for the outcome edit, and no additional wake on the unchanged sweep.

The run used a synthetic GitHub endpoint. It did not exercise a live CodeRabbit request or capture the other variants you listed.

---

Learnings used

Learnt from: eimexdev
Repo: pingdotgg/t3code PR: 15427
File: apps/server/src/orchestration-v2/PullRequestWatchReactor.ts:136-139
Timestamp: 2026-10-04T02:34:08.441Z
Learning: In the pull-request activity producers under apps/server/src/pullRequest, initial review-thread comments already appear in PullRequestActivity.comments. GitHub merges reviewThreadConversation output; Bitbucket builds both lists from the same kept comment entries; GitLab notes include discussion comments; Forgejo builds both lists from its inline records; Azure returns no reviewThreads. In apps/server/src/orchestration-v2/PullRequestWatchReactor.ts, readRemarks can therefore seed its comment map from activity.comments and add continuation-page comments without separately merging initial thread.comments.

_{You are interacting with an AI system.}