Code mapping

<essential_principles>

Maps are hierarchical: L0 (architecture) links to L1 (domains) which link to L2 (modules) which link to L3 (source code). Always start at L0 and drill down as needed.

Anchors enable grep navigation: Headings contain [Lx:identifier] markers. Find any section with grep "\[L1:auth\]" docs/map/.

Size limits enforce abstraction:

• L0: 500 lines max (system overview)
• L1: 300 lines max (domain docs)
• L2: 200 lines max (module docs)

If a doc exceeds its limit, split it into the next level down.

Code is truth, map is derivative: When code and map conflict, the code wins. Update the map to reflect reality.

Bootstrap pattern: Agents read README.md first, then ARCHITECTURE.md, then relevant domain docs based on task.

</essential_principles>

<auto_routing>

Route based on user intent (match first):

User says	Action
"explore", "read", "understand", "navigate", "find", "where is"	Read workflows/explore.md, then follow it
"create", "generate", "new", "build", "document", "map"	Read workflows/create.md, then follow it
"validate", "check", "verify", "lint"	Run: uv run python scripts/code_map.py validate <map-dir> (path relative to this SKILL.md)

If intent is ambiguous, ask:

What would you like to do?

1. Explore - Navigate existing map to understand codebase
2. Create - Generate a new code map
3. Validate - Check map integrity

</auto_routing>

<quick_reference>

Directory structure:

docs/map/
├── README.md           # Entry point + domains index
├── ARCHITECTURE.md     # L0: System overview
└── domains/            # L1: Domain docs
    ├── auth.md
    └── api.md

Anchor format: ## [L1:identifier] Heading Text

Code link format: [`symbol`](path/to/file.py#L42)

Cross-reference format: [Domain Name](domains/name.md)

</quick_reference>

<reference_index>

Domain Knowledge (references/):

Reference	Purpose
format-spec.md	Format rules for anchors, links, and size limits
examples/	Input/output examples (l0-architecture.md, l1-domain.md, l2-module.md, readme.md)
domains.md	What domain docs should capture

</reference_index>

<cli>

Running the Scripts

The scripts use Python 3.11+ stdlib only (no external dependencies).

Script paths are relative to this SKILL.md file (not the working directory). Derive the absolute script path from this file's location:

• If this SKILL.md is at /path/to/code-mapping/SKILL.md
• Then the script is at /path/to/code-mapping/scripts/code_map.py

# Validate an existing map
uv run python scripts/code_map.py validate <map-dir>

# Generate map skeletons from source
uv run python scripts/code_map.py generate <src-dir> <map-dir>

Generate command features:

• Creates one .md file per source file (1:1 mapping)
• Uses <!-- TODO: description --> placeholders
• Preserves filled descriptions on subsequent runs
• Reports new sections, removed sections, and unfilled placeholders

</cli>

<workflows_index>

Workflows (workflows/):

Workflow	Purpose	Status
explore.md	Navigate existing map to understand codebase	Ready
create.md	Generate new map for a codebase	Ready

Scripts (scripts/):

Script	Purpose	Status
code_map.py	CLI entry point (validate, generate)	Ready
generate/	Map skeleton generation from source	Ready
validate/	Map validation (links, sizes, symbols)	Ready

</workflows_index>

<success_criteria> A well-executed code map:

• Has valid L0 ARCHITECTURE.md under 500 lines
• Has L1 domain docs under 300 lines each
• All file links resolve to existing files
• All code links have valid file + line number
• All anchors are unique across the map
• No orphan documents (everything linked from parent level) </success_criteria>