What problem does it solve? Onboarding to a new or large repository forces every AI agent session to re-read the entire codebase, wasting context and producing inconsistent understanding. This Skill converts that exploration into a persistent, layered documentation system so agents locate the right module docs in one or two hops. ## Core Features & Use Cases - Progressive-Disclosure Knowledge Base: Builds a tiny always-loaded entry (CLAUDE.md/AGENTS.md) routing to a .rules/README.md index and on-demand per-module docs with load_when triggers. - Parallel Subagent Deep-Reading: Dispatches module-scoped subagents with strict templates to write L2 docs that cite code by symbol name only, never line numbers, and link rather than duplicate existing docs. - Verification & Sweep: Verifies every cited symbol against current source, fixes drift, and runs grep-based sweeps for leftover line numbers, broken links, and BOM issues. - Use Case: You inherit a 200k-line repo and keep re-explaining it to coding agents. Run this Skill once to produce a .rules/ tree, routing README, and updated CLAUDE.md/AGENTS.md entries that every future agent session reads first. ## Quick Start Ask the agent to explore this repository and build a .rules/ knowledge base with routing docs so future agents can navigate it without re-reading all the source.
