codebase-overview

Files (1)

• SKILL.md

SKILL.md (first lines)

# codebase-overview

1. Glob top-level dirs + read the manifest.
2. Map each top-level source directory to a one-line purpose (skip node_modules/dist/build/.git/.next).
3. Trace primary data flow: entry point → routing → key components/screens → data layer (API/store/DB).
4. Flag performance-sensitive integration points by file:line — third-party script loading, router/navigation setup, large media/asset loading, analytics init.
5. Write/update `PROJECT/.claude/codebase-overview.md` with the findings — the reusable reference other tasks read instead of re-discovering. Record the architecture explicitly (pattern, boundaries, dependency direction, which layer owns transactions/errors/authz): detected-but-unrecorded is re-detected differently next session.
6. Run `/arch-check` once the map exists — it measures the tree against the architecture just recorded; a mixed pattern or an inverted dependency belongs in the overview as a known state, not as a surprise for the next feature.
7. If root `CLAUDE.md` lacks an architecture section, add ≤10 lines pointing to it (essentials inline, detail lazy-loaded from `codebase-overview.md`) — never let CLAUDE.md itself balloon past its existing budget.
8. Never modify source files — documentation only.

## Output

```text
CODEBASE OVERVIEW: [project name] — written to PROJECT/.claude/codebase-overview.md
DIRECTORIES: [dir → one-line purpose]
DATA FLOW: [entry → routing → components → data layer]
KEY COMPONENTS: [component → responsibility — only non-obvious ones]
PERF-SENSITIVE: [file:line — why it matters]
CLAUDE.md: [updated | already covered | needs manual review — reason]
```

From mtvrkan/senior-dev-kit, MIT.

More from mtvrkan/senior-dev-kit