review-story — PRD Review & Correction Pipeline

Review the following: $ARGUMENTS

Overview

Review and correction pipeline for already-implemented user stories. Takes a PRD (single story or full PRD), researches best practices (meta-code-style pattern), runs static analysis, then parallel AI code review + security audit, validates findings, then risk-tiered remediation with scope-guard. Stops after correction — no commit or push.

Key principles:

• Context compression at every phase boundary (Anthropic context engineering)
• Research-informed review — understand best practices before judging code
• Intent-aware — understand why the code was written before judging how
• Static-first — run deterministic checks (cheap, fast, high-precision) before AI review
• Fresh-context reviewers — subagents with no bias toward the code (Actor/Critic isolation)
• Signal over noise — target 2-4 high-value findings per file, suppress style opinions
• Scope discipline — review diff + 1 dependency hop, not the entire codebase
• Risk-tiered fixes — auto-fix low-risk issues, confirm medium-risk, escalate high-risk
• No commit — the user controls when and how to commit