plan-review-integrator

Plan-Review Integrator

After a review panel generates 14 findings with mixed severity, someone has to cross-reference each one against the plan, classify what's a real blocker versus pre-existing debt, and edit the document without breaking coherence. This skill does it in minutes with full traceability.

A Claude Code skill that closes the gap between review reports and implementation plans. Instead of manually cross-referencing findings, classifying severity, and editing plan documents, this skill systematically processes every finding, validates it against domain context, and applies concrete updates with an audit trail.

Quick Start

You: Here's the review panel output for our API migration plan:
     review_panel_api_migration.md
     And here's the plan: docs/plan_api_migration.md
     Integrate the review findings into the plan.

Claude: [Reads both documents + project context]
        [Extracts 14 findings across 3 reviewers]
        [Cross-references each against the plan]
        [Classifies: 3 must-fix, 5 bundle, 4 defer, 2 informational]
        [Applies edits to plan document]
        [Produces traceability summary]

        ## Review Integration Summary
        - Total findings: 14
        - Must-fix: 3 (applied to plan)
        - Bundle into implementation: 5 (added to checklists)
        - Deferred: 4 (documented with rationale)
        - Informational: 2 (no action needed)

        | ID     | Severity | Summary                        | Category      | Action Taken                          |
        |--------|----------|--------------------------------|---------------|---------------------------------------|
        | R1-F01 | CRITICAL | Missing auth check in step 3   | Must-fix      | Added guard to Phase 2 code snippet   |
        | R1-F02 | HIGH     | Stale timeout config           | Bundle        | Added to implementation checklist      |
        | R1-F03 | MEDIUM   | Refactor legacy adapter        | Defer         | Pre-existing debt, tracked in TODO    |
        | ...    | ...      | ...                            | ...           | ...                                   |

        Would you like to adjust any classifications before finalizing?

Installation

Claude Code

# Clone the repo
git clone https://github.com/wan-huiyan/plan-review-integrator.git ~/.claude/skills/plan-review-integrator

# Or add as a project skill
git clone https://github.com/wan-huiyan/plan-review-integrator.git .claude/skills/plan-review-integrator

Cursor

Add to your .cursor/rules/plan-review-integrator.mdc:

---
description: Integrate structured review panel findings into implementation plan documents
globs:
alwaysApply: false
---

@plan-review-integrator/SKILL.md

What You Get

• Actionability-filtered findings — Low-signal noise dropped before classification; epistemic labels from upstream reviews weight effective severity
• Classified findings — Every finding sorted into must-fix, bundle, defer, or informational using epistemic-weighted severity (a [VERIFIED] HIGH outranks a [SINGLE-SOURCE] CRITICAL)
• Updated plan document — Concrete edits with dual rollback (rephrase, then revert) when edits break coherence
• Post-integration verification — Automated coherence, completeness, voice, and cross-reference checks
• Peripheral updates — ADRs, runbooks, and memory files updated where findings affect them
• Traceability summary — Full audit trail mapping each finding to its disposition and the action taken
• Persistent integration log — integration_log.jsonl enables cross-session learning from prior integration decisions
• VoltAgent specialist verification (optional) — When VoltAgent agents are installed, P0/P1 must-fix edits get domain-expert second opinions before application, catching wrong fixes that general context might miss

Comparison