Agent Claude Code
Read-only code review of a strike-cli PR or diff. Correctness bugs first (concrete failure scenario), then needless complexity and missed reuse. Posts findings for the issue-handler to put on the PR — never edits code. Use after a PR is open or after review-fix pushes.
Agent Claude Code
Writes Go tests only for strike-cli using stdlib testing and repo conventions. Never modifies production code. Use when adding coverage for tools, permissions, protocol, session, config, providers, auth, engine, or TUI.
Agent Claude Code
Runs strike-cli verification only (make test/vet/build, optional race and package-focused go test). Read-only — report failures verbatim, never edit. Use after implementation or when the user asks to validate/test the repo.
Agent Claude Code
Builds and refines strike-cli's TUI — views, panels, modals, pickers, badges, dashboards, and layout/theme work in internal/frontend/tui, internal/frontend/tui/ui, and internal/frontend/tui/theme. Use for any change that touches how the TUI looks or lays out. Not for backend/engine work (internal/engine…
Agent
must not load as portable.
Agent
Review agent.
Agent
Stage and create git commits for already-validated work. No source edits, no push. Use via task tool or /agent commit when the user wants commits recorded.
Agent
Root-cause investigation for failing tests or bugs. Reproduce cheapest-first, may add temporary instrumentation then revert it. Returns diagnosis and minimal fix proposal — does not ship the fix unless asked.
Agent
Fast read-only codebase exploration. Find files, search code, answer where/how questions. Prefer for task tool when the parent only needs locations and short conclusions.
Agent
General-purpose subagent for multi-step research and execution when no specialist fits. Full tool access (subject to permissions). Use for bounded parallel slices via the task tool.
Agent
Coordinate subagents; don't solo large implementations. Plan → task to specialists → synthesize.
Agent
Own an open PR through CI and review. Watch checks, pull failed logs, fix branch failures, push, re-watch. Stop-and-ask on product ambiguity. Not a from-scratch feature implementer.
Agent
Run strike verification only (make test/vet/build, optional race). Report failures verbatim. Never edits code to go green.
Agent
Goal-backward requirements check. Verify acceptance criteria with evidence; report PASS/FAIL/UNVERIFIED. Never implements fixes.