map-codebase

SKILL.md

Reference Files

Workflow & Process:

Core Templates (always use):

Optional Templates (use as applicable):


Map Codebase

Analyzes existing codebases using parallel Explore agents to produce structured documentation in .planning/codebase/.

Objective

Spawns multiple Explore agents to analyze different aspects of the codebase in parallel, each with fresh context. Results are structured documents in .planning/ that provide a comprehensive map of the codebase state.

Core Output (always generated):

  • STACK.md — Languages, frameworks, key dependencies
  • ARCHITECTURE.md — System design, patterns, data flow
  • STRUCTURE.md — Directory layout, module organization
  • CONVENTIONS.md — Code style, naming, patterns

Optional Output (generate as applicable):

  • TESTING.md — Test structure, coverage, practices (if tests exist)
  • INTEGRATIONS.md — APIs, databases, external services (if external dependencies exist)
  • CONCERNS.md — Technical debt, risks, issues (for complex/brownfield projects)

When to Use

Use map-codebase for:

  • Brownfield projects before initialization — understand existing code first
  • Refreshing codebase map after significant changes
  • Onboarding to an unfamiliar codebase
  • Before major refactoring — understand current state thoroughly
  • When STATE.md references outdated codebase info — refresh understanding

Skip map-codebase for:

  • Greenfield projects with no code yet — nothing to map
  • Trivial codebases (<5 files) — not worth the overhead

Process

Step 1: Check Existing Documents

Check if codebase documents already exist in .planning/:

  • If yes: Offer to refresh (overwrite) or skip
  • If no: Proceed with creation

Step 2: Determine Template Set

Ask user which optional templates to generate (or use defaults):

  • Always: STACK.md, ARCHITECTURE.md, STRUCTURE.md, CONVENTIONS.md
  • Optional: TESTING.md (if tests exist), INTEGRATIONS.md (if external services), CONCERNS.md (if complex project)

Step 3: Load Project Context

Check for .planning/STATE.md to load existing project context if available. Helps agents understand project-specific terminology and patterns.

Step 4: Process Focus Area Argument

If user provided a focus area (e.g., "api" or "auth"), instruct agents to pay special attention to that subsystem while still providing holistic analysis.

Step 5: Spawn Parallel Explore Agents

Launch 4 Explore agents in parallel, each with "very thorough" exploration level:

Agent 1: Technology Stack & Integrations

  • Focus: Languages, frameworks, dependencies, external services
  • Outputs: Data for STACK.md and INTEGRATIONS.md

Agent 2: Architecture & Structure

  • Focus: System design, patterns, directory organization
  • Outputs: Data for ARCHITECTURE.md and STRUCTURE.md

Agent 3: Conventions & Testing

  • Focus: Code style, naming patterns, test infrastructure
  • Outputs: Data for CONVENTIONS.md and TESTING.md

Agent 4: Concerns & Technical Debt

  • Focus: Issues, risks, technical debt, potential improvements
  • Outputs: Data for CONCERNS.md

Step 6: Collect Agent Findings

Wait for all agents to complete. Collect and organize findings by document type.

Step 7: Write Core Documents

Write 4 core markdown documents in .planning/:

STACK.md: Languages, versions, frameworks, build tools, key dependencies
ARCHITECTURE.md: System design, patterns, component relationships, data flow
STRUCTURE.md: Directory layout, modules, naming patterns, organization
CONVENTIONS.md: Code style, naming conventions, patterns, documentation

Step 8: Write Optional Documents (if selected)

Generate only selected optional documents:

TESTING.md (if applicable): Test frameworks, structure, coverage, practices
INTEGRATIONS.md (if applicable): External APIs, databases, third-party services, config
CONCERNS.md (if applicable): Tech debt, issues, security, performance, improvements

Step 9: Provide Next Steps

Inform user codebase mapping is complete:

  • Review generated documents in .planning/
  • Use findings to inform refactoring or development
  • Update STATE.md with new insights if needed

Success Criteria

  • 4 core documents created in .planning/ (STACK, ARCHITECTURE, STRUCTURE, CONVENTIONS)
  • Optional documents created based on user selection
  • All documents have substantive, actionable content
  • Documents follow template structure
  • Parallel agents completed without errors
  • User informed of completion and next steps

Integration Notes

Command Integration:

  • Invoked via /map-codebase [optional: focus-area] command
  • Focus area argument is passed to agents for targeted analysis

Project Lifecycle:

  • Can run before initial project setup on brownfield codebases
  • Can run after initial project setup to refresh as code evolves
  • Can run anytime to refresh codebase understanding
Weekly Installs
16
GitHub Stars
9
First Seen
Jan 24, 2026
Installed on
claude-code15
opencode15
gemini-cli15
antigravity14
codex14
kimi-cli13