| name | auto-test |
| description | Run a project's automated tests end-to-end and return a single structured report (ASCII dashboard + machine-readable trailing lines + path to .test-runs/<UTC-ts>/run.log). Trigger when the user asks to run, execute, or check the status of a project's tests โ phrases like "run the unit tests", "run the tests", "execute the test suite", "ai check tests for me", "run npm test and tell me what failed", or "how are the tests doing?". This is the meta-skill that delegates to layer-specific runners (Phase 1 ships only the `unit-test-runner` child); future phases will add `browser-test`, `integration-test-runner`, and `coverage-reporter` to the same orchestration. Do NOT trigger on "tail the log", "show me the latest test output", or "what does .test-runs/latest contain?" โ those are plain `Read` of `<project>/.test-runs/latest/run.log`. Do NOT trigger on lint, typecheck, build, deploy, or coverage-only requests โ those belong to other (separate) skills. Do NOT trigger on browser-only / Playwright-with-page.goto navigation requests until the Phase 2 `browser-test` skill ships. |
| allowed-tools | Bash, Read, Glob, Grep |
auto-test
Single entry-point for "run the tests" intents in a JS / TS / Bun project.
Phase 1 wires:
detect framework โ init run folder โ exec runner (stdout streamed to log)
โ parse runner output โ run.json
โ finalize (summary/manifest/run.log + latest)
โ render ASCII dashboard
The output is a single dashboard block plus four trailing
KEY=value lines (LATEST=..., LOG=..., JSON=..., EXIT=...) that
loop / agent consumers can parse without ANSI handling.
When to use
Activate when a user (or a higher-level orchestrator like a claude-bridge
loop) wants to run the test suite of the project at the current
working directory and get a verdict back. Concretely the trigger phrases:
- "run the unit tests", "run the tests", "run npm test"
- "execute the test suite", "kick off tests"
- "ai test this for me", "check that tests still pass"
- "what failed in the last run?" (re-runs the suite โ not a log read)
Do not activate when:
- The user wants to read an existing run log โ
Read <project>/.test-runs/latest/run.log directly.
- The request is for lint / type-check / build โ separate concerns.
- The request is for browser-only / Playwright navigation tests โ Phase 2
browser-test skill (not yet shipped).
- The request is for coverage thresholds only โ future
coverage-reporter skill.
- The project is Python / Rust / Go / Ruby โ Phase 3 multi-runtime
detectors (not yet shipped).
If detection fails (framework=unknown), this skill exits 2 with a clear
"could not detect a JS/TS test framework" message โ do not loop or
retry; surface the error to the user.
How to use
SKILL_DIR="<install-path>/skills/auto-test"
"$SKILL_DIR/scripts/orchestrate.sh" "$PROJECT_ROOT"
The dashboard rendered to stdout looks like:
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
โ auto-test ยท run 20260506T0500Z ยท duration 782ms โ ALL โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ Project : my-app โ
โ Framework : bun (auto-detected) โ
โ Command : bun test โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ UNIT โ 12 passed โ 0 failed โ 0 skipped โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ ARTIFACTS โ
โ log /abs/path/.test-runs/20260506T0500Z/run.log โ
โ json /abs/path/.test-runs/20260506T0500Z/run.json โ
โ manifest /abs/path/.test-runs/20260506T0500Z/manifest.json โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโค
โ EXIT 0 all 12 tests passed โ
โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
LATEST=/abs/path/.test-runs/20260506T0500Z
LOG=/abs/path/.test-runs/20260506T0500Z/run.log
JSON=/abs/path/.test-runs/20260506T0500Z/run.json
EXIT=0
When tests fail, a FAILURES section lists up to the first 3 failing
cases (name + first 60 chars of error_msg). Beyond 3 the dashboard
emits (โฆ N more โ see run.log); the full list is in
<run-dir>/run.json under the failures array.
Examples
User: "run the tests" โ activate; resolve project root from cwd; emit
dashboard; reply with the box + trailing LATEST= etc.
User: "tail the latest test log" โ do not activate;
Read <project>/.test-runs/latest/run.log directly.
User: "what failed in the last run?" โ activate (re-runs the suite โ the
question implies the user wants the current state, not the log).
User: "run the playwright e2e against the dev server" โ do not
activate; that needs Phase 2 browser-test.
Files
SKILL.md
scripts/
orchestrate.sh โ meta orchestrator: delegates to unit-test-runner/run.sh + render
render-dashboard.sh โ read manifest/summary/run.json โ ASCII box (Phase 1 compact form)
references/
exit-codes.md โ 0 / 1 / 2 contract (matches PRD ยง7 + ARCHITECTURE ยง(g))
tests/
test-orchestrate.sh โ E2E on a fixture (skips on missing bun)
test-render-dashboard.sh โ pure-fn unit test against a pre-baked run dir
run-all.sh โ convenience runner
goldens/
sample-run/ โ handcrafted summary/manifest/run for renderer test
See also
docs/PRD.md ยง7 โ meta-skill terminal report wireframe (we ship the compact Phase-1 form).docs/ARCHITECTURE.md ยง2 โ auto-test row.docs/ARCHITECTURE.md ยง(b) โ run folder contents this skill consumes.docs/ARCHITECTURE.md ยง(g) โ exit-code contract.skills/unit-test-runner/ โ the only Phase-1 child orchestrated.skills/test-log-centralizer/ โ folder layout + finalize contract.
