Overview
Transform dev-agent from returning pointers to returning actionable context. When an AI assistant asks "where is X?", we should return enough information to understand and work with the code—not just file paths that require additional reads.
Philosophy
- Structured data over prose - Code snippets, relationships, patterns
- Token-aware - Fit context within LLM limits using progressive disclosure
- Reduce round-trips - One query should return everything needed
Success Criteria
dev_search returns code snippets + imports (not just locations) ✅dev_refs shows what calls/is called by a functiondev_map provides codebase overview in one querydev_plan returns raw context, not heuristic task breakdowns
Tasks
Out of Scope
- LLM integration (we provide context, external LLMs reason)
- Effort estimation heuristics
- PR description generation
- Task breakdown logic (let the LLM do this)
Timeline
Target: v0.2.0 release
Overview
Transform dev-agent from returning pointers to returning actionable context. When an AI assistant asks "where is X?", we should return enough information to understand and work with the code—not just file paths that require additional reads.
Philosophy
Success Criteria
dev_searchreturns code snippets + imports (not just locations) ✅dev_refsshows what calls/is called by a functiondev_mapprovides codebase overview in one querydev_planreturns raw context, not heuristic task breakdownsTasks
dev_refs)dev_map)Out of Scope
Timeline
Target: v0.2.0 release