What problem does it solve? AI coding agents landing in an unfamiliar repository lack a concise, project-specific onboarding guide, so they guess at build commands, conventions, and hard rules. This Skill inspects the repo and writes a tight AGENTS.md (or updates an existing one surgically) so future agent sessions start unblocked. ## Core Features & Use Cases - Repo discovery and drafting: Reads the README, package manifest, lint/test configs, CI workflows, directory layout, and git history to extract the commands and conventions that matter, then drafts a 200-400 word "Repository Guidelines" document ordered by leverage. - Surgical update path: When AGENTS.md already exists, it diffs repo state since the file's last-touched commit, classifies each line as keep/update/remove/missing, and applies targeted edits instead of a rewrite. - Directory-level scope: Invoked in a subdirectory, it produces a shorter local guide (120-250 words) capturing sibling file patterns, naming, and directory-specific tripwires. - Use Case: After scaffolding a new monorepo, run the skill at the root to produce an AGENTS.md citing @package.json scripts and CI gates, then run it again inside packages/api/ for a route-handler-specific guide. ## Quick Start Ask the agent to create an AGENTS.md onboarding guide for this repository, optionally passing a target directory or file path.
