What problem does it solve? AI coding agents like Claude Code, OpenAI Codex, and GitHub Copilot each load instructions, skills, hooks, and rules from different files and directories, so teams struggle to know which configuration is actually active, why a rule is being ignored, or how to port behavior between hosts. This Skill audits the real state of your development harness and runtime harness (Google ADK) and guides targeted fixes. ## Core Features & Use Cases - Harness Inventory and Classification: Scans AGENTS.md, CLAUDE.md, .claude/, .codex/, .agents/, .github/, tests, and CI to classify every documented artifact as active, present-but-disconnected, planned, or absent. - Ignored-Rule Diagnosis: Locates a failing rule, finds its enforcement mechanism, verifies discovery via /doctor, /skills, and /hooks, then strengthens the right layer with deterministic sensors like tests, lint, pre-commit, or CI. - Cross-Host Porting: Converts hooks, subagents, and skills between Claude Code, Codex, and GitHub Copilot formats, including hook input format differences like tool_input.command versus tool_input.filePath. - Use Case: A team notices Codex ignores a rule that works in Claude Code. The Skill inventories the repo, finds the rule only exists in .claude/, and ports it to AGENTS.md plus a .codex/hooks.json hook with a tested script. ## Quick Start Ask the agent to audit which harness rules and hooks are actually active in this repository and fix any that are documented but not enforced.
