Agent Claude Code
Executes a single well-scoped, bounded coding task and reports back a terse summary. Use proactively for any discrete implementation step handed down by the orchestrator — one bead / one task at a time.
Agent Claude Code
Independently verifies one completed task against its bead's done-criteria in a fresh context — inspects the real diff, reads the changed files, runs the tests — and returns a pass/fail verdict with specific findings. Use after the implementer reports a task done, before the orchestrator closes the bead.