test-generate

Generate tests for $ARGUMENTS, then run them. This is a generate → run → verify loop, not just file generation.

1. Discover

• Detect the test framework already in use (look for jest.config*, vitest.config*, pytest.ini / pyproject.toml [tool.pytest], Go's _test.go convention, Cargo.toml [dev-dependencies], etc.). Never add a new framework or runner.
• Skim 1–2 existing test files near the target to mirror style, fixtures, setup/teardown, assertion style, and naming convention.

2. Generate

Cover:

• Happy path (expected usage)
• Edges (empty / null / boundary / malformed)
• Errors (invalid input, IO / network / DB failures)

Match existing patterns:

• Same framework, same fixture style, same naming convention.
• Reuse existing setup/teardown helpers; do not add new mock libraries or new test-runner config.
• Do not add test-only public methods or exports to production code to make a test possible — if the test wants a hook the prod code doesn't expose, stop and ask.

3. Run

Execute the project's real test command, scoped to just the new tests (-t <name>, --testPathPattern, pytest path/to/test.py, go test ./pkg, etc.). Report:

• Pass / fail counts.
• Any failure with the assertion message and file:line.

4. Verify the tests BITE

A test that passes is not the same as a test that exercises the code. For each new test, confirm:

• It would fail if the function under test returned the wrong value. Mentally swap a broken impl — does the assertion actually catch it?
• It isn't expect(true).toBe(true), expect(fn).toBeDefined(), or a mock asserting itself.
• It isn't passing only because the setup masked the real call.

Mark each new test biting or superficial (with reason). Superficial tests count as a failure to ship — either fix or delete.

5. Report

• Files added / modified.
• Pass / fail.
• Biting vs superficial breakdown.
• What was deliberately not tested, and why (third-party shims, trivial getters/setters, generated code). Documented gaps beat fake coverage.