research-codebase

Documenting an existing codebase requires structured, read-only exploration to map component interactions and feature areas. This process deploys parallel sub-agents to locate files and gather concrete references, producing an evidence-backed technical map of the repository architecture.

Researching a GitHub issue involves decomposing the linked topic into targeted research areas and exploring the current git commit context. It generates a dated Markdown research document with YAML frontmatter, summarizing detailed findings and providing exact file path and line references.

Yes, you can generate codebase documentation without modifying source code by using read-only analysis. Specialized sub-agents locate files, explain how specific code works, and identify existing patterns safely, ensuring the repository remains completely unchanged during the research process.

Evidence-based synthesis for architecture discovery works by breaking a research question into targeted areas and gathering concrete references. It extracts exact file paths and line numbers during read-only code exploration, then combines these references into a comprehensive technical map.

Yes, you need a specific git branch and commit context to accurately map technical flows. The research process documents the repository implementation as it exists today, generating reports under flow/research/ that capture the exact state of the codebase from the current commit.

The main limitation of read-only codebase research is that it cannot fix or modify the code it analyzes; it strictly documents existing patterns and component interactions. Users needing code changes, refactoring, or active issue resolution must apply separate execution steps after the research phase.