From fdced0bb92e8b0b4337e53ebe31e7d6d62942ebe Mon Sep 17 00:00:00 2001 From: iap Date: Sat, 21 Nov 2026 12:00:34 +0700 Subject: [PATCH 1/2] docs(agent): document bot-authored PR review limitation --- CONTRIBUTING.md | 1 + docs/agent/bot-reviews.md | 13 +++++++++++++ docs/agent/review-ops.md | 7 +++++-- 3 files changed, 19 insertions(+), 2 deletions(-) diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 2effc59..de857a9 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -269,6 +269,7 @@ adaptation ledgers in `docs/`. 3. After CodeRabbit reviews, reply in-thread to all findings with evidence (fix or justify). Continue until CodeRabbit's CI flow/check completes successfully. The maintainer or bot resolves the threads. 4. **Merge only when CodeRabbit is clean on the current head commit**, every Qodo finding is triaged (fixed or answered with evidence), every review thread is resolved, and CI + local gates pass. Green CI alone is insufficient. Treat "skipping"/"pending" as not clean. 5. Reply to findings with comments explaining what/why (not just resolving). Cross-check for false positives before applying fixes. +6. **Bot-authored PRs** (e.g. dependabot) cannot get a CodeRabbit verdict without a bot review seat; the check stays `skipping`. Do not waive the gate or retry — land the identical change in an ordinary PR, pass the full gate, and close the bot PR as superseded. See `docs/agent/bot-reviews.md`. Qodo config: optional repo-level Qodo/PR-Agent config may live at `.pr_agent.toml` or `.qodo/pr-agent.toml`; this repo ships none. If added, keep it to a one-line pointer here and in `AGENTS.md`. See `docs/agent/bot-reviews.md`. diff --git a/docs/agent/bot-reviews.md b/docs/agent/bot-reviews.md index 976b674..f0a382f 100644 --- a/docs/agent/bot-reviews.md +++ b/docs/agent/bot-reviews.md @@ -31,6 +31,19 @@ short form and must stay consistent with this file. (fixed or answered with evidence) on this head. Use `@coderabbitai review` for a small fix push within the same round. +## Bot-authored PRs (dependabot) + +CodeRabbit skips PRs authored by a bot without an assigned review seat: its +"Review skipped" note asks an org admin to assign a seat to the bot. Re-requests +and cooldowns cannot change that — do not spend review cycles retrying. The +merge gate still applies: a `skipping` CodeRabbit check is never clean, and +there is no waiver for bot PRs. The compliant remedy is to land the identical +change in an ordinary (human-authored) PR from a normal branch, run the full +gate on it, then close the bot PR as superseded with a pointer to the new PR. +Seat assignment by an org admin is the alternative. + +Observed on #210/#211 (Nov 2026), superseded by #235. + ## Gate (merge requires ALL) - CodeRabbit clean on the **current head commit** (`commit.oid == headRefOid`). diff --git a/docs/agent/review-ops.md b/docs/agent/review-ops.md index 87458e8..a0daefd 100644 --- a/docs/agent/review-ops.md +++ b/docs/agent/review-ops.md @@ -39,9 +39,12 @@ clean state before merge. The port contract outranks every bot. force-push, or large diff; use `@coderabbitai review` only for a small fix push within the same round. A CodeRabbit check of `skipping` means **not requested**, never clean. If it does not respond, check CodeRabbit's - documented commands rather than assuming it is clean. + documented commands rather than assuming it is clean. + A bot-authored PR (e.g. dependabot) has no review seat, so a request returns + "Review skipped" and the check stays `skipping`; do not retry — supersede it + with an ordinary PR per `docs/agent/bot-reviews.md`. - Two "no verdict" states look identical in the UI and are not: + Two "no verdict" states look identical in the UI and are not: - `skipping` - no verdict. - `in_progress` that never moves - **the review is paused**, not slow. After a From f5f13c0c99a20396bec3239afcdc42f1667629a6 Mon Sep 17 00:00:00 2001 From: iap Date: Sat, 21 Nov 2026 12:38:39 +0700 Subject: [PATCH 2/2] docs(agent): allow bot review retry after seat assignment --- CONTRIBUTING.md | 2 +- docs/agent/bot-reviews.md | 15 ++++++++------- docs/agent/review-ops.md | 6 ++++-- 3 files changed, 13 insertions(+), 10 deletions(-) diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index de857a9..3505952 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -269,7 +269,7 @@ adaptation ledgers in `docs/`. 3. After CodeRabbit reviews, reply in-thread to all findings with evidence (fix or justify). Continue until CodeRabbit's CI flow/check completes successfully. The maintainer or bot resolves the threads. 4. **Merge only when CodeRabbit is clean on the current head commit**, every Qodo finding is triaged (fixed or answered with evidence), every review thread is resolved, and CI + local gates pass. Green CI alone is insufficient. Treat "skipping"/"pending" as not clean. 5. Reply to findings with comments explaining what/why (not just resolving). Cross-check for false positives before applying fixes. -6. **Bot-authored PRs** (e.g. dependabot) cannot get a CodeRabbit verdict without a bot review seat; the check stays `skipping`. Do not waive the gate or retry — land the identical change in an ordinary PR, pass the full gate, and close the bot PR as superseded. See `docs/agent/bot-reviews.md`. +6. **Bot-authored PRs** (e.g. dependabot) cannot get a CodeRabbit verdict without a bot review seat; the check stays `skipping`. Do not waive the gate, and do not retry while the seat is missing: either have an org admin assign the bot a seat and request a fresh review, or land the identical change in an ordinary PR, pass the full gate, and close the bot PR as superseded. See `docs/agent/bot-reviews.md`. Qodo config: optional repo-level Qodo/PR-Agent config may live at `.pr_agent.toml` or `.qodo/pr-agent.toml`; this repo ships none. If added, keep it to a one-line pointer here and in `AGENTS.md`. See `docs/agent/bot-reviews.md`. diff --git a/docs/agent/bot-reviews.md b/docs/agent/bot-reviews.md index f0a382f..26e68cc 100644 --- a/docs/agent/bot-reviews.md +++ b/docs/agent/bot-reviews.md @@ -34,13 +34,14 @@ short form and must stay consistent with this file. ## Bot-authored PRs (dependabot) CodeRabbit skips PRs authored by a bot without an assigned review seat: its -"Review skipped" note asks an org admin to assign a seat to the bot. Re-requests -and cooldowns cannot change that — do not spend review cycles retrying. The -merge gate still applies: a `skipping` CodeRabbit check is never clean, and -there is no waiver for bot PRs. The compliant remedy is to land the identical -change in an ordinary (human-authored) PR from a normal branch, run the full -gate on it, then close the bot PR as superseded with a pointer to the new PR. -Seat assignment by an org admin is the alternative. +"Review skipped" note asks an org admin to assign a seat to the bot. Until a +seat is assigned, re-requests and cooldowns cannot change that — do not spend +review cycles retrying. The merge gate still applies: a `skipping` CodeRabbit +check is never clean, and there is no waiver for bot PRs. After an org admin +assigns the bot a seat, a fresh review request is allowed. Otherwise the +remedy is to land the identical change in an ordinary (human-authored) PR from +a normal branch, run the full gate on it, then close the bot PR as superseded +with a pointer to the new PR. Observed on #210/#211 (Nov 2026), superseded by #235. diff --git a/docs/agent/review-ops.md b/docs/agent/review-ops.md index a0daefd..d85b099 100644 --- a/docs/agent/review-ops.md +++ b/docs/agent/review-ops.md @@ -41,8 +41,10 @@ clean state before merge. The port contract outranks every bot. requested**, never clean. If it does not respond, check CodeRabbit's documented commands rather than assuming it is clean. A bot-authored PR (e.g. dependabot) has no review seat, so a request returns - "Review skipped" and the check stays `skipping`; do not retry — supersede it - with an ordinary PR per `docs/agent/bot-reviews.md`. + "Review skipped" and the check stays `skipping`; while the seat is missing, + do not retry. Either have an org admin assign the bot a seat and request a + fresh review, or supersede the PR with an ordinary one per + `docs/agent/bot-reviews.md`. Two "no verdict" states look identical in the UI and are not: