AI Code Review: MR gets reviewed multiple times for the same commit

[echarrod]

Describe the bug

MR gets reviewed multiple times for the same commit, and comments multiple times.

To Reproduce

1. Self-hosted GitLab instance, running with config:

# -*- mode: yaml -*-
manifest:
  version: 1.0

automations:

  linearb_ai_description:
    if:
      - {{ not pr.draft }}
      - {{ not is.bot }}
    run:
      - action: describe-changes@v1
        args:
          concat_mode: append

  linearb_ai_review:
    if:
      - {{ not pr.draft }}
      - {{ not is.bot }}
    run:
      - action: code-review@v1
        args:
          guidelines: {{ guidelines }}

# Define variables
# Add conditions for PR approvals. For example - allow approval only for specific users
is:
  bot: {{ pr.author | match(list=['@project_35771297_bot_558e119f0f4d33f2fc8b315e66d53c86', '@group_14117954_bot_085a90599cd4653c9aa736ce6538ae23', '@project_35771297_bot_3baa505335b81595ea1f502563700a41']) | some }}

# Add `guidelines` to add your prompts to the review, you can have org wide instructions
# by placing the `REVIEW_RULES.md` in the `cm` repo or team level instructions by adding it to
# the team repo.
# Note: To when the file is in the `cm` repo, use this example: `../cm/REVIEW_RULES.md`
guidelines: {{ "./.cm/REVIEW_RULES.md" | readFile() | dump }}

Expected behavior

One comment per commit/push, with older comments being deleted

Screenshots

Additional context

Running on self-hosted GitLab instance, using image index.docker.io/gitstream/rules-engine:1.0.130

[vim-zz]

Thanks for your patience while we keep refining gitStream’s AI review experience. We have two changes rolling out this week that should address the pain points you’ve raised:

1. Trigger rule tightened: Only commit-push events will invoke an AI review. Other events e.g. label changes, PR description edits, review comments, etc. will no longer trigger a fresh AI analysis. This should cut down on the “extra” review cycles you’ve been seeing.
2. Smarter re-review logic: AI review findings will now persist until you explicitly resolve them in the code or dismiss them in the PR conversation.

• You’ll receive an automatic LGTM from the AI reviewer only when every outstanding finding has been addressed.
• Merely pushing new commits that don’t fix the flagged issues will no longer clear the findings list.

Both updates are already in staging and we expect them to be live in production later this week. Once they’re out, please let us know if you still notice any hiccups, or if there are other improvements you’d like to see!

Thanks again for helping us make gitStream better. 🙌