What problem does it solve? Non-trivial coding tasks often fail because the agent dives into edits without understanding the codebase, invents duplicate definitions, or declares success without evidence. This Skill enforces a disciplined written-plan workflow so changes are grounded in real code, self-critiqued, peer-reviewed, and verified with actual output. ## Core Features & Use Cases - Read-only exploration: Trace relevant files and data models, cite real code paths, and give every discovered constraint a verdict (work around or delete toward). - Explicit written plans: Name the single owning definition for each fact, group steps into items with inputs, deliverables, and core/dispatch delegation labels, and start convergent setup work in the background. - Self-critique and peer review: Reread the draft as a hostile reviewer for up to three passes, then hand the plan and diff to a fresh code-reviewer subagent. - Evidence-based verification: Require real test runs, observed behavior, or captured output instead of accepting a success claim. - Use Case: When asked to plan adding rate limiting to an API, the Skill maps middleware read-only, writes a plan reusing the existing Redis client, critiques and reviews it, then verifies by hitting the endpoint until a 429 response appears. ## Quick Start Ask the agent to plan how to implement a multi-file feature, and it will explore the codebase, write and critique a plan, review it, and verify the result with real output.
