RPI Review

Goal

Write one evidence-based review record after implementation finishes. Assess the supplied task once, keep execution separate from outcome, and route each finding to the stage or later work that can resolve it.

Flow

1. Resolve one task artifact set: current plan, phase details, latest plan critique, changes record, and relevant research. Use the supplied paths or the stable task slug and date. Stop if multiple unrelated sets remain ambiguous.
2. Create one record at .copilot-tracking/reviews/logs/{{YYYY-MM-DD}}/{{task_slug}}-review.md using templates/review-log.md. Do not create review modes or plan a second review pass.
3. Confirm plan markers, phase details, changes evidence, handoff prose, blockers, remaining work, follow-up items, and validation state are reconciled. Then compare the complete supplied boundary: requirements, acceptance criteria, phase and task completion evidence, critique dispositions, descriptive implementation-time updates and decisions, validation, blockers, remaining work, and plan ## Follow-Up Items. Confirm significant or divergent implementation decisions preserve confirmed user intent and are reflected in the current plan. Navigate by markers and headings, not line numbers.
4. Use generic bounded subagents for independent lenses only when they reduce a specific review uncertainty. Give each a narrow question, exact read boundary, and no source-write authority. Do not use a dedicated RPI review worker or fixed review-worker allowlist.
5. Record one complete set of substantive, severity-graded RV-xxx entries. Keep execution status separate from outcome: execution is Complete, Partial, or Blocked; outcome is Conformant, Conformant with justified divergence, Defects found, Residual work, or Not accepted.
6. Route each actionable gap once: defects suitable for later implementation to rpi-implement, significant or divergent decision gaps to rpi-plan, material evidence gaps to rpi-research, and residual work to a distinct follow-up item. Route unresolved plan follow-up items distinctly without treating them as defects or adding them to active plan scope. A later rpi-implement invocation does not require this Review to run again.
7. Return the review record, separate execution status and outcome, validation evidence, findings, and recommended destinations.

Success criteria