← ClaudeAtlas

per-task-review-cyclelisted

Implement → spec review → quality review → targeted fix → re-review cycle for multi-task subagent development. Use when executing an implementation plan where each task needs independent verification before moving to the next.
williamblair333/Uncle-J-s-Refinery · ★ 3 · AI & Automation · score 76
Install: claude install-skill williamblair333/Uncle-J-s-Refinery
# Per-Task Review Cycle Use this within a `superpowers:subagent-driven-development` session when tasks have verifiable specs and code quality standards that must be confirmed before merging. ## When to use - Executing a multi-task plan where each task has a written spec - Each task produces a standalone file or feature (not deeply interleaved) - You want to catch issues per task rather than at the end ## Cycle per task implementer subagent → spec reviewer subagent (parallel if independent) → quality reviewer subagent (parallel if independent) → [if issues] targeted fixer subagent → re-review only the fixed portions → APPROVED → next task Run spec and quality reviews **in parallel** after implementation completes — they're independent reads. ## Override protocol When a reviewer flags something that is **architecturally correct by contract** (e.g., `trap 'exit 0' ERR` in a Stop hook, intentional silent exit, framework-mandated behavior): 1. Identify the specific reviewer error (wrong lens, wrong contract model) 2. State the correct invariant explicitly 3. Override to APPROVED with documented rationale 4. Do not dispatch a fix for correct code ## Final integration review After all tasks complete, run one cross-task review subagent scoped to the full diff. Catches: - Duplicate flags or args across files - Inconsistent PATH/env injection patterns - Missing error propagation contracts ## Checklist - [ ] Implementer dispatched - [ ] Spec + q