
from repo-atlas20
Generate an in-repo persistent context atlas that maps a codebase: directory trees, entrypoints, architecture docs, and onboarding material so agents and engine
Repo Atlas is a procedural skill that generates a structured, in-repo atlas to help humans and LLM agents understand a codebase quickly. It provides a small generator script plus a workflow to create docs/atlas/ (repo-map, architecture, domain model, critical flows, gotchas, test matrix) without changing runtime code. The atlas is intended to be repo-specific and to include real file paths and examples rather than generic guidance.
Use Repo Atlas when onboarding contributors, preparing a codebase for LLM-driven automation, or creating persistent documentation for long-term maintenance. It is especially useful before asking an agent to modify or refactor code so the agent can rely on a curated map and avoid exploratory noise.
scripts/generate_atlas.py) to produce docs/atlas/repo-map.md and changelog snippets (the skill includes templates and examples).Designed for Claude Code and other agent frameworks that benefit from repo-level context and structured docs; works with local Python-based tooling for generation and verification.
Repo Atlas generates persistent context documentation (directory trees, entrypoints, changelogs, architecture docs) for any codebase so LLM agents can navigate it efficiently. The bundled generate_atlas.py script is well-structured Python 3 stdlib-only code that crashed in sandbox because it resolves REPO_ROOT relative to script location and tries to write to /docs/atlas — it's designed to run inside an actual repo, not in isolation. No security concerns; no network calls, no credentials, no destructive commands.
Solid, well-documented skill for a genuinely useful task. Script failure in sandbox is expected — it's repo-context-aware by design, not a bug. Clean security profile: pure stdlib, no network, no secrets.