Pull Request Assistant
The Pull Request Assistant helps you address code review feedback directly within merge requests (GitLab) or pull requests (GitHub). Mention DAIV in a comment and it will apply changes, answer questions, or fix failing pipelines.
Note
On GitLab this works on Merge Requests, on GitHub on Pull Requests. This page uses "pull request" to refer to both.
How to use
Mention DAIV in any comment on a pull request:
| Text Only | |
|---|---|
Replace daiv with your DAIV bot's username if it differs.
DAIV reacts with an eyes emoji to confirm it received your request, then processes it and replies or pushes changes to the branch.
Where to comment
You can mention DAIV in two places:
- Inline on the diff — comment directly on specific lines of code
- General discussion — comment in the pull request conversation
DAIV has context of the diff and the comment's position, so inline comments give it precise context about what you're referring to.
What it can do
Apply code changes
Ask DAIV to modify code based on your review feedback:
| Text Only | |
|---|---|
| Text Only | |
|---|---|
| Text Only | |
|---|---|
Answer questions
Ask DAIV about the code to help you make review decisions:
| Text Only | |
|---|---|
| Text Only | |
|---|---|
Fix failing pipelines
When CI/CD fails, ask DAIV to investigate and fix it:
| Text Only | |
|---|---|
DAIV will inspect the pipeline logs, identify the root cause, and push a fix to the branch.
Conversation continuity
DAIV maintains context across multiple interactions on the same pull request. You can have a back-and-forth conversation — each new mention builds on previous context, so DAIV understands the full history of changes and discussions.
Tip
If DAIV starts drifting or gets stuck, you can use @daiv /clear to reset the conversation and start fresh.
Configuration
The pull request assistant is enabled by default. To disable it, add the following to your .daiv.yml:
Review reports
Each review is posted as a single comment on the pull request — findings grouped by severity (Critical / Important / Suggestions), open questions for the author, and a short list of recommended actions. On GitLab, reviews stack: a re-review posts a new report covering only the commits since the previous one (after a force-push, the next report covers the full change again and says so). If a review could not cover one of its dimensions — say the security detector failed — the report says so, and the next review re-covers that span rather than treating it as done. A review requested when nothing has changed since the last one answers with a one-line "already reviewed" note instead of a fresh report.
On GitHub the report carries no tracking marker: re-review scope comes from the conversation history on the pull request rather than from the reports themselves, so it is lost if that conversation is cleared — after which the next report covers the whole pull request again.
You can reply on the merge/pull request and mention DAIV to ask about a finding or have it apply a fix.
Custom review rules
Define team-specific review rules in .agents/review-rules.md at the repository root. Write them in plain language, one rule per bullet, with any path scope expressed in prose:
- Every external API call in
src/payments/**must set an explicit timeout. - Never log request or response bodies.
- New Celery tasks must be idempotent.
The code review agent checks for rule sources at the start of every review and, when any exist, runs a dedicated cr-custom-rules detector against the diff; each violation it posts cites the rule it enforces.
You don't need a dedicated .agents/review-rules.md: rules already written in your repository's AGENTS.md and .agents/AGENTS.md are picked up as a secondary source — the agent mines them for concrete, diff-checkable conventions. When sources disagree, .agents/review-rules.md wins. The detector is skipped only when none of these files exist.
Every custom-rule finding passes the same confidence gate and skeptical aggregation as built-in findings, so a noisy AGENTS.md will not flood the review.
Related pages
- Slash Commands & Skills — invoke
/code-reviewfrom a merge request, and write custom skills to extend it - Sandbox — the isolated container where DAIV reproduces and repairs failing pipelines
- Sessions — follow a running review and revisit what the agent changed
- Repository Config — tune review behaviour per repository with
.daiv.yml