Rpi Review Agent Skill

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

• 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.
• 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.
• 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.
• 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.
• 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.
• 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.
• Return the review record, separate execution status and outcome, validation evidence, findings, and recommended destinations.

Success criteria

• One review record exists at the canonical review path and includes all compared artifacts.
• One Review records the complete finding set for the supplied task boundary.
• The record separates execution state from outcome verdict.
• Findings are substantive, evidence-grounded, severity-graded RV-xxx records with an explicit destination.
• Defects, decision gaps, research gaps, and residual work are routed to distinct destinations.
• Descriptive implementation-time plan and detail updates, their rationale and evidence, material revision readiness, and plan follow-up items are explicitly assessed.
• Validation evidence is recorded or explicitly unavailable or skipped with a reason.
• Findings are routed clearly without creating closure, correction, full, targeted, or amended review modes.

Constraints

• Do not implement fixes or mutate the plan, phase details, critique, research, or changes record in this stage. Review may create or update only its one canonical review record.
• Do not create per-phase review-worker outputs or depend on retired dedicated RPI review workers.
• Use plain-text workspace-relative paths in the review record.
• Use references/review.md for the review method, outcome vocabulary, routing detail, and conversation protocol.

Conversation guidance

Use references/review.md as the authority for the state-first opening, materiality gate, continual-update template, marker meanings, pre-question context, and closeout behavior. Persist review-owned state before an opening or potential material update; chat is a concise projection, never a second history or delivery log. Preserve the read-only boundary, separate execution status from outcome, standalone versus parent continuation, conditional compaction, and linked Markdown table. For every relevant existing artifact, use the two-cell row | actual/workspace-relative/path.ext | Short description |, using that artifact's actual workspace-relative path as both link text and destination; omit unavailable files and render the table immediately before the final ## Next Steps section. End with ## Next Steps: state the exact eligible user command, active-parent action, blocker-clearing action, follow-up choice, or that no user action is required. When compaction is warranted, tell the user to run /compact before the next RPI command; otherwise omit compaction guidance.

Stop rules

• Stop as Blocked if a reviewable artifact set cannot be formed or evidence is insufficient for a credible verdict.
• Stop as Not accepted when material defects or unaccepted decision gaps remain.
• Complete a partial review only when the record names the evidence boundary and routes the missing work.

Handoff

Return the review record, execution status, outcome, severity summary, validation coverage, and the next recommended RPI stage or distinct follow-up item. A standalone review advises the exact /rpi-* command only when a finding needs that destination and does not invoke it. In rpi-quick or confirmed automatic RPI Agent mode, return the review record to the parent for automatic continuation.

Final response

Return review execution status separately from outcome, findings, validation coverage, blockers or open items, routed follow-up, and conditional compaction advice when warranted. Follow Conversation guidance for standalone or parent-orchestrated continuation, the linked artifact table, and final next steps.