werkrevisor/docs/prs/2026-10-06-PR#30-no-stale-change-request.md

PR#30: a RED review no longer blocks later commits

WARNING: This document describes only the change applied in this PR. It may already be outdated once the next PR is merged. Historic PR documentation is not maintained along with new PRs; treat it as a snapshot.

The Problem

With [gate] yellow_gate = "approval", werkrevisor 0.19.2 submits a review REQUEST_CHANGES for a RED head. For a later head that is not RED, it submits APPROVED for GREEN, and a review COMMENT or nothing for YELLOW and ORANGE.

In Gitea, a review COMMENT does not replace an earlier REQUEST_CHANGES, and a new commit does not dismiss it. So the rejection of a fixed RED stays and blocks the merge, also after a human approval. This happened on PR#29: the RED of commit 4a3313a blocked the merge of commit 8116b42, until an admin dismissed the review.

The bot cannot dismiss its own review: Gitea answers "Must be repo admin", and the bot has only write access.

A RED head needs no review to be blocked: werkrevisor ends with exit code 20, so the required status check ai-review/gate fails for that head, and the next commit gets a new status.

Non-Goals

• No admin rights for the bot.
• No change of the branch protection: a human REQUEST_CHANGES still blocks the merge.

The Scenarios

Feature: a RED head blocks through its status only

Scenario#30.01: a RED head gets a review comment, no change request

So that a later commit that fixes the RED is not blocked by a stale rejection.

• Given the gate mode is approval
• When werkrevisor reviews a head with the verdict RED
• Then the bot submits a review COMMENT with the violated rules
  • and it submits no review REQUEST_CHANGES
  • and werkrevisor ends with exit code 20, so ai-review/gate fails

Verified by

• test_line_comments: test_red_gives_a_comment_review_and_a_failed_gate
• test_open_questions: test_open_question_gives_a_comment_review

The Solution

• In the gate mode approval, a RED head gets a bot review COMMENT with up to 3 violated rules instead of REQUEST_CHANGES.
• The README and docs/betrieb-git.javagil.de.md say that RED blocks through the failed status check, and that block_on_rejected_reviews now concerns human change requests only.
• werkrevisor 0.19.3.

Prerequisite PRs

• PR#29: larger full files and excerpts.