Agent
Verify external-API/library/protocol facts against live documentation. Returns facts and source URLs, never reasoning chains or hedging. Used by the research skill to isolate doc-fetching from orchestrator context.
Agent
Code review subagent. Receives a diff plus dimensions and returns classified findings (material-gap | prose-clarity | implementation-detail) with a verdict. Never edits.
Agent
Multi-source verification subagent. Reads candidate sources, returns a Ground Truth Audit with confidence label and a concrete falsifier. Used by the triangulate skill to isolate heavy verification reading from orchestrator context.