| 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.
- Adjacent files may be read for context, but findings and completion claims must stay limited to the owned review scope.
Deterministic Git Evidence Gate
Before producing findings or a completion verdict:
- define the
owned review scope
- run one or more path-scoped Git comparison commands for that scope
- record the exact command text that was used
- inspect the diff before reading implementation details deeply enough to form conclusions
Allowed Git evidence shapes:
git diff -- <owned paths> for working-tree reviews
git diff --cached -- <owned paths> when staged changes matter
git diff <base>...HEAD -- <owned paths> for branch or multi-commit review
git show <commit> -- <owned paths> for committed change sets
Rules:
- Never rely on a repo-wide diff when unrelated dirty files may exist.
- Never return
complete or not complete without quoting the exact path-scoped Git command(s) used.
- If the owned scope is ambiguous or the Git evidence cannot be isolated deterministically, downgrade the verdict to
unclear / partially verified or ask the user for a cleaner base.
Required Inputs
- the governing
.planr/plans/*.plan.md when one exists
- the relevant live scope entry from
.planr/status/current.json
- any explicitly referenced historical source doc still in scope
- the path-scoped Git diff for the owned review scope
- the changed implementation files and relevant tests in that scope
Required Workflow
- Read the governing plan and live scope context first.
- Define the exact owned review scope before reading implementation details deeply.
- Run and record one or more path-scoped Git comparison commands for that scope.
- Inspect every owned changed file and relevant test through the scoped diff plus direct file reads.
- Check acceptance criteria, checked phase items, completed
todos, and live-status claims against the actual diff and recorded verification.
- Review architecture, hard-cut cleanup, correctness, and test sufficiency.
- Produce findings-first output and persist a
.planr/review/*.review.md artifact only when it adds execution value.
Output Format
Use this structure:
Findings
Completion verdict
Review scope and Git evidence
Open questions or assumptions
Brief summary
Rules for the response:
- Findings come first, ordered by severity.
- Each finding should explain why the implementation fails the task, architecture, hard-cut direction, correctness, or cleanliness standard.
- The completion verdict must say one of:
complete
not complete
unclear / partially verified
- The
Review scope and Git evidence section must list the exact owned paths and the exact Git comparison command(s) used.
- If shared files required hunk-level review, say whether the owned hunks were isolated or remained ambiguous.
- If a plan doc was in scope, say whether its checked phase items were confirmed or overclaimed.
- If a persisted review artifact was created, include its path.
- If there are no findings, say that explicitly and still mention residual risks or testing gaps.
- Keep the summary short. The findings are the main output.
Additional Resource
- Read ../planr-shared.md first for shared CLI coverage and shared
.planr rules.
- For the detailed review checklist, hard-cut checks, and report templates, see reference.md