What problem does it solve? Adding an AI-driven development workflow to a project that already has code, tests, and CI is risky: existing conventions, configs, and files like CLAUDE.md can be silently overwritten. This Skill scans the repository, extracts existing conventions, proposes an adoption plan, and applies it only after explicit user confirmation. ## Core Features & Use Cases - Repository Reconnaissance: Detects languages, frameworks, test runners, databases, CI systems, and existing rule files (README, CONTRIBUTING, .editorconfig, lint configs) to build a project profile. - Plan-Before-Apply Gating: Writes a PLAN.md listing every file to add, files it will not touch, and conflicts, then waits for an explicit "apply" before making any change. - Merge and Overwrite Modes: Never overwrites existing files without per-file confirmation; merges Claude Code hook settings per event and logs conflicts to CONFLICTS.md. - Use Case: You inherit a NestJS + Prisma backend with Vitest tests and GitHub Actions. Run the skill to detect the stack, extract conventions from CONTRIBUTING.md into a draft rules fragment, review the plan, and apply the 7-agent Harness workflow without touching source code or CI. ## Quick Start Tell the AI to run /harness-adopt in the root of your existing project to scan the codebase and propose a non-destructive Harness adoption plan.
