Research

This skill orchestrates deep codebase exploration by spawning sub-agents that search through multiple tiers of information: code maps, semantic search, git history, scoped grep, source files, and prior knowledge artifacts. It's designed for when you need comprehensive findings on a specific topic rather than quick lookups. The implementation is backend-agnostic, detecting whether you're running Codex sub-agents, Claude native teams, or background tasks, then dispatching accordingly. Interesting bit: it enforces iterative retrieval where agents score relevance and extract new search terms from high-scoring files for up to three refinement cycles. Results land in markdown files under .agents/research/. The auto flag skips approval gates for fully autonomous workflows.