A Claude Code skill that produces navigable maps of how code actually works — not summaries of file contents.
Point it at any repo and say "explore this codebase." It will classify the architecture, trace data flows, identify fragility points, and produce a structured orientation map you can act on.
npx skills add quicksilversurfer/codebase-explorer
After installing, trigger it in Claude Code with natural language:
- "Explore this repo"
- "Help me understand this codebase"
- "I inherited this code — onboard me"
- "What does this project do?"
- "Audit this system's architecture"
The skill runs a structured 4-phase exploration:
Phase 0 — Archetype Detection: Classifies the repo (frontend, backend API, data pipeline, serverless, library, IaC, CLI, ML system) and selects the right exploration strategy.
Phase 1 — Orientation: Maps I/O boundaries, substrate constraints, implicit contracts, and development activity.
Phase 2 — Vertical Slices: Traces data from input to output using an archetype-specific strategy — visible values for frontends, record-through-stages for pipelines, request lifecycle for APIs.
Phase 3 — Smell Detection: Scans for architectural smells matched to the codebase type, including pipeline-specific patterns like drifting duplicates, implicit inter-stage contracts, and shared utility single points of failure.
Phase 4 — Risk Profile: Synthesizes findings into specific, actionable risk statements and explicitly marks what wasn't explored.
| Archetype | Exploration Strategy | Primary Output |
|---|---|---|
| Frontend Application | Trace visible value backward | Render-site trace |
| Data / Serverless Pipeline | Trace record through stages | Data lineage map |
| Backend API | Trace request lifecycle | Request lifecycle trace |
| Library / SDK | Trace public API inward | Abstraction layer map |
| Infrastructure as Code | Trace resource dependencies | Blast radius map |
| CLI Tool | Trace command execution | Command execution trace |
| ML System | Trace training data flow | Data + model lineage map |
| Audience | What they get |
|---|---|
| Developer inheriting code | Where to start, what's safe to change |
| Tech lead / architect | Risk map, coupling analysis, refactoring targets |
| Engineering manager | Change cost estimates, coordination needs |
| Product manager | What's hard vs. easy to change, in business terms |
| Security auditor | Attack surface, data flow, trust boundaries |
| New hire | Navigable mental model of the system |
| Consultant / contractor | Fast orientation before making changes |
SKILL.md # Main skill definition (353 lines)
references/
smell-catalog.md # 13 architectural smells with detection guides
documentation-templates.md # 7 output templates (including stakeholder briefs)
advanced-techniques.md # Fan-in handling, dead code detection, verification
The skill is pure Markdown — no code, no dependencies. When triggered, Claude Code loads SKILL.md as context, which guides the exploration methodology. Reference files are pulled in on demand when the main skill references them.
The skill improves Claude's output by providing:
- A repeatable protocol instead of ad-hoc exploration
- Archetype-specific strategies that prevent wasted effort
- Named concepts (shared fate groups, implicit contracts, data lineage maps) that produce more actionable output
- Audience adaptation that adjusts depth and vocabulary to the reader's role
MIT