← ClaudeAtlas

extract-structurelisted

Work out a repo's folder taxonomy - where the code root is, which folders are product features and which are shared infrastructure, and which files boot the app - into birdseye/.cache/structure.json. Use first when building or refreshing a birdsEye map.
tamalCodes/birdsEye · ★ 0 · Data & Documents · score 57
Install: claude install-skill tamalCodes/birdsEye
# Extract structure > **Dormant as of the zero-token rewrite.** birdsEye's `/birdseye:map` no > longer invokes this skill - the map is now a pure local AST parse (graphify) > with no model calls. This file is kept for a future opt-in `--with-llm` mode. Every other stage of birdsEye now goes through this one. The import graph, the routes, the docs and the whole Overview flowchart are all organised around the answer you give here: *what are this repo's modules, and which of them are features versus general-purpose infrastructure?* A script has already done the mechanical half - walked the tree, counted files, matched folder names against a list of conventional infrastructure names, looked for `pages/` directories. Your job is the half a script cannot do: look at the handful of folders it was unsure about and decide, the way a person joining the team would after ten minutes in the codebase. Write `birdseye/.cache/structure.json` and nothing else. Never modify the user's source tree. ## Non-negotiables - **Never fail. Always write the file.** A weird layout is not an error. If you genuinely cannot tell, ask the user (step 4) - but you still finish with a written `structure.json`. - **Only modules the scan found.** The scan lists every candidate folder. You may move a folder between `featureModules` and `sharedModules`, but you may not invent a module it did not list, and you may not split or merge folders. - **A feature owns screens or a flow a user walks through.*