← ClaudeAtlas

llm-docs-generatorlisted

Operating playbook for generating and maintaining version-pinned documentation packs with the llm-docs CLI. Use when building, refreshing, or verifying an agent-ready docs pack for a specific installed library version, or when maintaining this repository.
Zyepher/llm-docs-generator · ★ 5 · AI & Automation · score 75
Install: claude install-skill Zyepher/llm-docs-generator
# llm-docs-generator `llm-docs` is a deterministic engine: it parses sources, preserves structure, and records provenance. It makes no judgments. YOU make every judgment — which version, which source tree, what to exclude, how to slice, what the index says. A pack is only as correct as the decisions you feed the engine. The failures this playbook prevents are all judgment failures: unpinned versions, drafts swept in, a 250k-token monolith no agent can load, a pack with no index, a pack that broke when it moved. Run this workflow in order. Every step states WHY so you can adapt it, not cargo-cult it. ## Division Of Labor (never violate) - You resolve version, source authority, source tree, exclusions, and slicing. The CLI never reads a vendor nav config, never picks a source, never decides a version. - `generate --source` takes explicit local files or directories only. Never pass a package name, docs URL, repo URL, or discovery report to `--source`. - Run `llm-docs capabilities --json` before assuming any command, flag, mode, or output exists; anything not reported there is unavailable unless the installed CLI says otherwise. If a flag in this playbook is missing from the installed contract, treat it as not yet shipped and adapt. - `llm-docs agent doctor --json` is read-only environment diagnostics. Do not claim `agent install codex` or any other install/lifecycle command unless `capabilities --json` reports it as implemented. - Parser plugins: treat explicit parser plugin