Agent Integration

ProjectAtlas is designed to be read at agent startup so you can:

• Pick the correct file quickly.
• Spot duplicated folder roles early.
• Keep structure clean as the repo grows.

Codex / AGENTS.md snippet

## Startup
1. Run `projectatlas map` (or ensure your build does).
2. Read `.projectatlas/projectatlas.toon`.
3. Scan `folder_tree[]` for where to work and `folders[]`/`files[]` for precise targets.
4. Check `folder_summary_duplicates[]` / `file_summary_duplicates[]` and flag drift.
5. Run `projectatlas lint --strict-folders --report-untracked`.
6. If lint fails, add missing Purpose headers or `.purpose` files (or remove stale items) before continuing.
7. Only then run deep indexing (code-index, LSPs) on the files you selected from the atlas.

Note: the non-source file list (`.projectatlas/projectatlas-nonsource-files.toon`) is agent-maintained input for
non-source summaries and is merged into the atlas. Agents still read only the generated atlas.

Codex skills

ProjectAtlas ships a Codex skill at .codex/skills/ProjectAtlas.md. Copy that file into your Codex workspace or keep it in place so Codex can load the ProjectAtlas workflow.

Claude / skills

Drop the skills/claude/ProjectAtlas.md into your Claude skills folder and reference it in your agent setup.

Lint and CI

ProjectAtlas lint is meant for local workflows. Many teams skip map generation in CI, but still run lint on PRs to surface missing headers. Use projectatlas map --force if CI must regenerate the map.