| name | requesting-code-review |
| description | Use after meaningful implementation work to create an independent Yardlet review task |
Requesting Code Review
Purpose
Turn review into an explicit Yardlet queue item with bounded inputs and a
structured verdict. Review is independent work, not an implicit hidden action.
When to Request
Request review after a meaningful implementation slice, before integration, or
when the task contract explicitly requires independent acceptance review.
Prepare Evidence
Record:
- what was implemented;
- the governing intent, task, or specification;
- the exact changed files or local revision range;
- validation commands already run and their results;
- unresolved risks or assumptions.
Do not claim that the implementation is correct. The review task owns that
judgment.
Create the Review Task
Create or propose a normal Yardlet queue task with kind: review. Keep its
allowed scope read-only unless the contract explicitly creates a separate
repair task. Use code-reviewer.md as the review body.
The review must return one verdict per acceptance criterion with concrete file,
command, or runtime evidence. A failed criterion remains failed until a later
repair task changes the workspace and review is rerun.
Act on Findings
- Repair critical and important findings in a bounded implementation task.
- Challenge an incorrect finding with code or test evidence.
- Keep optional polish separate from blocking acceptance.
- Never mark the reviewed task complete solely because the implementer reports success.
