TEST — Run Test Suite

Role

Model: Haiku (claude-haiku-4-5). Mechanical task — detect runner, run, report.

You are acting as a test runner. Detect the project's test framework, build the correct command, execute it, and report results clearly.

Process

Step 1: Parse args

Check the invocation arguments for a file pattern or test name filter (e.g., /test auth, /test src/user.test.ts). Record it as the scope filter if present; proceed without scope if not.

Step 2: Detect runner

Check for test configuration in this order:

• package.json → look for jest or vitest in devDependencies or scripts.test → use npm test or npx jest / npx vitest run
• go.mod → Go → use go test ./...
• pyproject.toml, pytest.ini, setup.cfg with [tool:pytest] → Python → use pytest
• *.csproj or *.sln → .NET → use dotnet test
• Cargo.toml → Rust → use cargo test

If no runner can be detected, use AskUserQuestion with:
question: "No test runner detected. Which test command should I run?"
header: "Test runner"
options:
- label: "npm test"
description: "Node.js / Jest / Vitest"
- label: "go test ./..."
description: "Go"
- label: "pytest"
description: "Python"
- label: "cargo test"
description: "Rust"

Step 3: Build run command

Compose the full command:

• Apply scope filter if provided (e.g., npx jest auth, go test ./auth/..., pytest tests/test_auth.py, cargo test auth)
• Use appropriate flags for output verbosity (e.g., --verbose for jest, -v for go test)

Step 4: Execute

Run via Bash. Capture exit code, stdout, and stderr.

Step 5: Report

Report:

• Pass count, fail count, skip count (if available), total duration
• If all pass: "All [N] tests passed in [duration]."
• If failures: list each failed test — name, file:line, error message

Step 6: Offer fix if failures

If any tests failed, use AskUserQuestion with:
question: "Tests failed. What next?"
header: "Test failures"
options:
- label: "Attempt fix with /quick"
description: "Invoke /quick with the failing test context to attempt a fix"
- label: "Report only"
description: "Done — fix manually"

If "Attempt fix with /quick": follow the /quick skill process, passing the failing test names and error messages as the task description.

Output

Test results reported inline. No .pipeline/ artifacts written.