discovery

Ground a task in the existing codebase by extracting key terms, grepping for prior art, and reading similar features to surface inherited patterns. This validates assumptions against existing code before specification, producing a discovery.md artifact with references and a handoff plan.

Codebase discovery is the process of scanning a repository to identify related code, conventions, and inherited gotchas before drafting specifications. You need it at the start of a task to validate assumptions against prior art and avoid duplicating existing patterns.

Find prior art and gotchas by grepping the repository for key terms extracted from your task, then reading files that implement similar features. This surfaces existing patterns, dependencies, tests, and inherited learnings to document before proceeding.

Yes, grep is used to scan files that implement related features and cite hits with paths and lines. Reading these similar features extracts conventions, dependencies, and tests, which are then documented as patterns to follow in the discovery output.

Document existing code surface and patterns by writing a discovery.md file containing sections for Existing Surface, Patterns to Follow, Inherited Gotchas, and Handoff. This artifact captures references, learnings, and a plan for the specification phase.

Validating assumptions against prior art prevents redundant work and unexpected conflicts by surfacing inherited gotchas and existing conventions early. Grounding the task in the codebase first ensures the specification accounts for established patterns and dependencies.