Examples

Examples

One feature driven end to end. Every artifact below is committed in the repository's examples/ folder, so you can read what the extension writes before installing it.

The feature is 003-user-auth in a TypeScript service: session expiry plus scoping the order list to the requesting user. Three acceptance criteria, four functional requirements, two components, eleven cycles.

Artifact	Written by
tdd-profile.md	/speckit.tdd.setup
test-list.md	/speckit.tdd.plan
cycle-log.md	/speckit.tdd.run
verification.md	/speckit.tdd.verify

1. The profile

/speckit.tdd.setup detected vitest, ran each candidate command, and recorded only what worked:

runner: vitest
single: 'pnpm vitest run {file} -t "{name}"'
suite: pnpm test
coverage: pnpm test --coverage
mutation: 'pnpm stryker run --mutate "{files}"'
acceptance: pnpm playwright test {file}
contract: null
verified: [single, file, suite, coverage, mutation, acceptance]
suite_baseline: green
suite_seconds: 34

The body records the conventions the loop has to match (test file naming, the vi.fn() double style, factories in src/testing/factories.ts, the injected Clock port) and one exemplar test file to imitate. It also records what is missing: packages/legacy has no runner, so work there needs characterization tests and has no single-test command.

2. The test list

/speckit.tdd.plan turned three acceptance criteria into three outer-loop behaviors and nine inner-loop behaviors, each traced:

id	behavior	traces	kind	state
A2	An expired session is rejected with a 401 and no order data	AC-2	example	DONE
U1	Rejects a token whose expiry is in the past	AC-2, FR-2	example	DONE
U2	Accepts a token expiring exactly at the current instant	AC-2	example	DONE
U3	Round-trips any valid claim set through encode and decode	FR-2	property	DONE
U5	Current listing behavior for a legacy customer record	AC-1	characterization	BASELINE

Two things the plan phase caught before any code existed:

• AC-2 needed two behaviors. "Reject an expired session" split into U1 and U2, because one test on one side of a threshold passes with both < and <=.
• repository.ts had no tests. U5 was scheduled first, capturing what the code did for legacy records as a baseline, before U6 changed the query underneath it.

3. The cycle log

One entry per cycle. The red line is the point:

## Cycle 3: U2 accepts a token expiring exactly at the current instant

- test: `src/auth/session.test.ts::accepts a token expiring now` (new)
- red: `pnpm vitest run src/auth/session.test.ts -t "accepts a token expiring now"`
  -> `AssertionError: expected 'expired' to be undefined` (1 failed)
- green: `src/auth/session.ts:31` changed `<` to `<=`. Suite -> 126 passed
- refactor: extracted `isExpired(claims, now)` from the inline comparison; suite
  re-run green after the extraction
- commit: `9c2b117` (behavior), `5ee0a30` (structure)

A real command, its real failure output, recorded before the fix existed. Behavior and structure landed as separate commits, so a reviewer reads one for correctness and one for shape.

Two cycles show the honest handling of a test that passes on its first run:

## Cycle 10: U9 preserves the existing sort order

- red: passed on the first run. Deliberate mutant applied: removed the `ORDER BY` at
  `src/orders/repository.ts:44`, re-ran -> `AssertionError: expected [b, a] to
  deeply equal [a, b]` (1 failed). Restored, suite re-run -> 133 passed
- green: no production change needed. The behavior existed and is now pinned against
  the refactor in cycle 11

No fabricated red. The log says exactly what happened, and the audit can tell the difference.

The Notes and deviations section records a step that was too big: cycles 2 and 3 were originally one attempt whose red covered two behaviors, so it was reverted to the last green and split.

4. The verification report

Verdict: PASS_WITH_GAPS. Discipline held, every criterion is covered end to end, no test was weakened anywhere in the branch. Two gaps:

Two behaviors are LIKELY, not PROVEN. Their commits were amended, so git history can no longer corroborate that the test came first. The log says it did; the audit will not upgrade that on the log's word alone.

One surviving mutant is a real finding:

Mutant	Behavior	Survived	Judgment
session.ts:31 <= to <	U2	No	Caught by U2; the boundary is pinned
repository.ts:41 predicate inverted on the empty path	U8	Yes	Real gap. U8 passes with the filter inverted: empty either way
session.ts:44 changed a debug log string	none	Yes	Equivalent mutant, no observable behavior

U8 asserts that no orders yields [] rather than null, which is correct, and it passes even with the user filter inverted because the result is empty either way. The line runs, so coverage is satisfied. Mutation testing is what found it.

That finding is the reason the audit exists: U8 reads as a perfectly good test.

The report closes with What was not audited: packages/legacy (no runner), the mutation scope (4 changed files, not the repository), concurrency (no criterion, no test), clock skew, and load behavior. Stating this is not optional. A report that reads as exhaustive when it was not is worse than no report.

Reading order

1. tdd-profile.md for what "language agnostic" means concretely.
2. test-list.md for the plan.
3. cycle-log.md for the evidence, top to bottom.
4. verification.md for the grade.

Formats are documented in Stack Profiles, Test List Format, and Test Quality.