What problem does it solve? Implementation plans drafted by AI agents often contain fabricated file paths, missing requirements, or scope creep that only surface during execution. This skill blocks flawed plans from reaching the user by requiring three independent adversarial reviews to pass before any plan is presented. ## Core Features & Use Cases - Parallel Adversarial Review: Spawns three fresh Task() instances — Feasibility, Completeness, and Scope & Alignment reviewers — that independently verify the plan against the real codebase with read-only access. - Binary PASS/FAIL Gate: Every reviewer produces an evidence-backed verdict; any single BLOCKING issue fails the gate and forces plan revision, with a maximum of 3 iterations before escalation to the user. - Reviewer Isolation: Fresh instances on every iteration with no cross-reviewer visibility prevent anchoring bias and confirmation bias from the original planner. - Use Case: After an agent drafts a multi-file refactoring plan, the gate verifies every referenced file exists via glob, maps each user requirement to a work unit, and rejects scope creep before you ever see the plan. ## Quick Start Ask the agent to draft an implementation plan for your feature request and the gate will automatically review it before presenting it to you.
