What problem does it solve? AI coding agents repeatedly make the same mistakes in a repository because project rules, conventions, and pitfalls are scattered or undocumented. This Skill produces a small, verified agent-instruction block inside AGENTS.md so agents follow the repo's real rules from the first session. ## Core Features & Use Cases - Setup, adopt, refresh, record, and audit modes: Scaffold a new instruction block, migrate existing CLAUDE.md or AGENTS.md content into a managed block, re-verify it against the current codebase, capture observed agent mistakes as pitfalls, or prune stale lines. - Evidence-based verification: Every command claim is checked against the actual script or CI config, every path is validated, and deletions require one of four documented grounds with user approval. - Ledger-driven migration: Existing instructions are tracked line-by-line (retain, rewrite, relocate, automate, delete) so nothing is silently dropped during adoption. - Use Case: A team adopts AI coding assistants but agents keep running the wrong test command and editing a frozen legacy directory. Run this Skill to interview maintainers, scan the repo, and write a verified AGENTS.md block with the correct invocations and protected paths. ## Quick Start Ask the agent to set up the AGENTS.md project context block for this repository using the bmad-project-context skill.
