| name | test-with-subagent |
| description | Use when tests need to be run with minimal output noise and delegated execution; routes test runs through the dedicated test-runner subagent and returns curated pass/fail evidence. |
Test With Subagent
Overview
Always run tests through the test-runner subagent to keep the main thread clean and high signal.
Core principle: Delegate test execution; keep only actionable results.
Policy
- ALWAYS use this skill for running tests.
- ALWAYS dispatch the
test-runner subagent before any direct test command.
- Do not bypass this flow unless the calling agent explicitly requires full raw logs for deep harness debugging.
Required Pattern
- Dispatch the
test-runner subagent (never run tests directly first).
- Give scope (
full suite or specific target like file/package/crate).
- Ask for curated output only: command source, command, PASS/FAIL, duration, relevant failures, short actionable excerpt.
- Use the returned signal to decide next step.
Prompt Template
Run tests using the test-runner workflow.
Scope: <full suite | specific target>
Context: <optional reason, e.g. pre-commit check or regression verification>
Return only:
- Test command source (AGENTS.md or inferred)
- Command executed
- PASS/FAIL
- Duration if available
- If failing: failing tests and a 5-15 line actionable excerpt
Failure Handling
- If failure is clear, fix code/tests and re-run via
test-runner.
- If failure is ambiguous, request one additional run scoped to the failing target.
- Escalate to direct/manual execution only when curated output is insufficient.
Red Flags
- Running raw
make test in the main thread before delegation
- Posting full unfiltered test logs to the calling agent
- Ignoring AGENTS.md command guidance
- Switching away from Makefile-first inference without a clear reason
