Demonstration codebase for building the AI Layer (CLAUDE.md hierarchy, hooks, skills, LSP, MCP, plugin) — companion repo for the 'How Claude Code works in large codebases' video.
Read-only subsystem explorer. Use it to map an unfamiliar service or package BEFORE editing — it explores with its own context window and reports back, so the main agent edits with the full picture instead of burning its context on discovery. The article's "split exploration from editing" pattern.
Use when creating or editing code under services/billing. Enforces Helpline's money-handling rules — integer cents only, seat limits, and BillingError for every violation.
Use after changing code, before claiming work is done — picks the correctly scoped pytest command instead of running the whole suite. Helps avoid the timeout-and-context-waste of full-suite runs on a one-service change.
Portable Helpline AI Layer: the scoped-tests skill, the self-improving CLAUDE.md review hook, the read-only explorer subagent, and the codebase-search MCP server. Install this and any repo gets the team's baseline Claude Code setup on day one.