review-feedback-orchestrator

# Review Feedback Orchestrator ## Inputs - Review feedback source (one of): - comments earlier in the chat, - pasted review text, - a markdown file path. - Optional scope hints (branch, folders, or files). ## Workflow 1. Collect review items from the provided source. - If source is ambiguous, summarize candidate items and proceed with the clearest set. - Normalize each item into: `id`, `summary`, `files`, `suggested_change`. 2. Validate every item before planning fixes: - Read the referenced files and nearby code. - Classify as one of: - `valid-defect` (real bug/risk), - `valid-improvement` (non-blocking but worthwhile), - `not-applicable` (incorrect or stale), - `needs-clarification` (insufficient evidence). - Add evidence for each decision (file paths, snippets, behavior). 3. Build a fix plan only for `valid-defect` and accepted `valid-improvement` items. - Group by dependency order and file overlap. - For each plan item include: objective, target files, acceptance criteria, tests/checks. 4. Dispatch implementation to subagents. - Use the most relevant coding/test subagent for each plan item. - Give each subagent: - exact review item IDs, - validation evidence, - required acceptance criteria, - required verification commands. 5. Track outcomes and reconcile with original review items. - Mark each item: `fixed`, `declined-with-rationale`, or `blocked`. ## Required Output 1. **Review Item Validation Matrix** - Columns: `ID | Verdict | Severity | Evidence | Rationale`. 2. **Implementation Plan** - Ordered checklist with owner subagent and acceptance criteria. 3. **Subagent Dispatch Prompts** - One prompt per implementation task. 4. **Completion Ledger** - Item-by-item status mapped back to the original feedback. ## Rules - Never implement a review item before validating it. - Keep declined items with explicit reasoning and evidence. - Prefer minimal-risk fixes and include regression checks. - If feedback conflicts, surface the conflict and propose a decision path.