analyze-codebase

name	analyze-codebase
description	Map an unfamiliar module's structure and dependencies, returning a summary
context	fork
allowed-tools	["Read","Grep","Glob"]
argument-hint	<module-path>

analyze-codebase — example SKILL (Domain 3, Task 3.2)

Frontmatter demonstrates all three tested options:

context: fork -> runs in an ISOLATED sub-agent context so the (verbose) findings

do NOT pollute the main conversation. Only a summary returns.

allowed-tools -> read-only (Read/Grep/Glob). The skill physically cannot Write,

Edit, or run Bash — a deterministic guardrail against side effects.

argument-hint -> if invoked as bare /analyze-codebase, Claude prompts for .

Skills are ON-DEMAND, task-specific workflows. Always-on universal standards belong in

CLAUDE.md instead. Want a personal variant? Put it in ~/.claude/skills/ under a

DIFFERENT name so teammates are unaffected.

Playbook

1. Glob the module path to list files (by NAME) — do not read everything.
2. Grep for entry points: exported names, route registrations, main/handler.
3. Read only the entry-point files; follow imports outward (incremental exploration).
4. For wrapper modules: list all exported names first, then Grep each across the repo.
5. Return a concise summary: purpose, key files, external dependencies, risk areas.