Index Documentation and Code Skill

Purpose

Create a searchable, structured knowledge base from existing documentation and codebase. This enables fast context lookup during task planning and implementation in brownfield projects through semantic search, documentation-to-code mapping, and quick reference systems.

Core Principle: In brownfield projects, context discovery is the bottleneck—indexing eliminates repeated searching.

Prerequisites

• Existing documentation in docs/ directory
• Structured, readable codebase
• Write access to .claude/index/ location
• Optional: document-project skill completed first

Workflow

1. Load Configuration and Validate

Check prerequisites and configuration:

# .claude/config.yaml
brownfield:
  docsPath: docs/
  codebasePath: src/
  indexLocation: .claude/index/
  includeCode: true
  maxDepth: 3

Validate:

• Documentation directory exists with files
• Index location is writable
• Check if index exists (offer refresh vs rebuild)

2. Index Documentation

Parse and structure documentation content:

1. Scan documentation files (.md files in docs/)
2. Extract structure:

- Headings and sections - Code blocks and examples - Keywords and concepts (API names, domain terms, technologies)

1. Build document index with sections, keywords, line numbers
2. Create cross-references between documents

Output: Documentation indexed (file count), sections extracted, keywords identified, cross-references created

See: references/templates.md#step-2-output for complete format and index structure

3. Index Codebase (Optional)

Map code to documentation:

1. Scan key code files:

- Models, schemas, types - Routes, controllers, services

1. Extract code elements:

- Exports, imports, functions - Map to line numbers

1. Build code-to-docs mapping:

- Find code mentions in docs - Create bidirectional links

Output: Code files indexed, exports/functions mapped, code-to-docs links created

See: references/templates.md#step-3-output for complete format and code-mapping.md for strategies

4. Build Searchable Index

Create fast lookup structures:

1. Create search index:

- Keyword → Documents - Concept → Sections - Code → Documentation

1. Generate quick reference:

- Data models with file locations - API endpoints with docs links - Patterns and where documented

1. Create glossary:

- Terms and definitions - Links to detailed documentation

Output: search.json (searchable index), quick-ref.md (quick reference), glossary.md (term definitions) created in .claude/index/

See: references/templates.md#step-4-output for complete file formats and examples

Using the Index

Quick searches: grep for keywords in search.json | Look up endpoints in quick-ref.md | Find term definitions in glossary.md

Integration: create-task-spec loads relevant docs | implement-feature looks up patterns | document-project refreshes index

See: references/templates.md#search-examples for complete search commands and #integration-examples for workflows

Index Maintenance

When to refresh: Documentation updated | New code modules | Architecture changes | Quarterly maintenance

Refresh vs Rebuild: Refresh=incremental update (faster, preserves customizations) | Rebuild=full re-index (slower, clean slate)

See: references/templates.md#index-maintenance for detailed refresh process and output examples

Best Practices

Index early (before implementation) | Keep docs updated (refresh after changes) | Use quick ref (faster than full docs) | Maintain glossary (project-specific terms) | Include code selectively (key files only)

Reference Files

Detailed documentation in references/:

• templates.md: All output formats (Steps 1-4), search.json structure, quick-ref.md format, glossary.md format, search examples, integration workflows, index maintenance, JSON output format

• indexing-format.md: Index structure details (comprehensive in templates.md)

• code-mapping.md: Code-to-docs mapping strategies

When to Escalate

• Documentation is sparse or missing (run document-project first)
• Codebase structure is unclear or chaotic
• Index becomes too large (>100MB)
• Search performance degrades

Part of BMAD Enhanced Brownfield Suite