What problem does it solve? Adopting an AI agent harness into a mature repository risks overwriting existing conventions, CI setups, and AI-tool configurations. This Skill audits what the repository already knows about itself before installing anything, so integration happens without silent conflicts or guesswork. ## Core Features & Use Cases - Read-only codebase audit: Derives the real stack, commands, branch conventions, and CI behavior directly from manifests, lockfiles, and workflow files rather than README claims. - Conflict detection: Finds existing AI-tool configuration (AGENTS.md, CLAUDE.md, .cursor/rules, copilot instructions) and surfaces conflicts for the owner to reconcile instead of overwriting them. - Optional decision-history reconstruction: Recovers architectural decisions from PR/MR history and session logs, grading each as CONFIRMED, INFERRED, or UNKNOWN, and drafts ADRs only after explicit owner authorization. - Use Case: A team with a two-year-old TypeScript monorepo wants to adopt the agent harness. The Skill audits package.json scripts and GitHub Actions, flags an existing CLAUDE.md that conflicts with harness conventions, confirms findings with the owner, then hands off to the standard install flow. ## Quick Start Ask the agent to integrate the harness into this existing repository and audit its current setup first.
