What problem does it solve? Implementation plans are typically self-reviewed by the same agent that wrote them, so defects like nonexistent file paths, forward task dependencies, and missing test coverage are only discovered after the implementation has already been paid for. This skill dispatches a fresh, review-only agent that verifies the plan against the real repository and returns a machine-parseable verdict before any code is written. ## Core Features & Use Cases - Independent codebase-verified review: A fresh agent reads every file the plan touches, greps for claimed-missing symbols, and checks scope, codebase fit, dependency order, and test coverage using a structured rubric. - Finding triage and plan revision: Each finding is accepted, declined with reasoning, or marked unresolved; accepted fixes are edited into the plan and verified, with deferred work triaged as absorb, file, or drop. - Machine-parseable verdict with graceful degradation: Emits a PLAN-REVIEW output block (clean, proceed-with-warnings, blocked, degraded) that orchestrating flows can gate on, and never blocks the build when the reviewer itself fails. - Use Case: On the notion-dev superpowers build path, the ticket flow invokes this skill with a plan file and context packet; the reviewer catches that Task 4 modifies a file that does not exist, the plan is revised, and the run proceeds with a clean verdict. ## Quick Start Ask the agent to review the implementation plan at plans/my-feature.md against the repository before starting implementation.
