| name | planr-review |
| description | Review agent-owned implementation scope in this repository against `.planr` plans or live status, path-scoped Git evidence, and acceptance criteria. Use for findings-first audits of completion, correctness, architecture, hard-cut cleanup, and test sufficiency. Not for implementing fixes (`planr-fix`), writing a new execution contract (`planr-plan`), or giving a verdict-only status answer (`planr-status`). |
Planr Review
Use this skill to audit whether scoped work is actually done and done cleanly.
A planr-review is invalid if it skips explicit owned scope or skips a path-scoped Git comparison for that scope.
CLI-First Rule
- Read ../planr-shared.md first.
- Use
./.planr/tooling/planr status show [--scope ...] for live scope context before dropping to raw .planr/status/current.json.
- There is no
planr.py review command today. The actual verdict still requires path-scoped Git evidence plus direct inspection of code and tests.
Owned Review Scope
Before reviewing any implementation, define the exact owned review scope.
- The default scope is only the files and hunks the agent changed in this task, or the exact files and hunks the user explicitly asked to review.
- Never assume the whole dirty worktree belongs to the review.
- If the repo is dirty and the agent cannot distinguish its own changed paths with high confidence, stop and ask for the review scope.
- If the same file contains both agent and non-agent edits, only review the agent-owned hunks when they can be isolated from conversation context, an explicit patch, or a known base commit.
- If mixed authorship inside a file cannot be isolated, do not guess. Mark the review
unclear / partially verified or ask for a clean comparison base.
