Capstone: ship a reviewed change

Tests before code

Turn every test.todo into a real test and commit them first. That commit is your proof that the tests defined the work, not the other way round.

Choose deliberately

Plan with a reasoning-strong model; implement with a fast one once the contract is fixed. Record why.

Review the whole diff

The proposed agent PR says "All tests pass ✅". Check what it changed in the tests, not just whether they pass.

Evidence, then decision

Commands, outputs and reasoning a reviewer can re-run, ending in accept, accept with changes, or reject.

- if (coupon.expiresAt && now >= new Date(coupon.expiresAt)) {
-   throw new Error(`Coupon ${coupon.code} has expired`);
- }
+ switch (coupon.type) {
…
-test("expired coupon is rejected", () => { … });
+test("fixed coupon", () => { … });

| Hunk                         | Decision | Evidence                         |
|------------------------------|----------|----------------------------------|
| Math.floor → Math.round      | Accept*  | $1.15 → 115¢ (*per-item cents)   |
| switch without expiry check  | Reject   | check:boundaries: expired passes |
| fixed: return amountCents    | Reject   | 1793¢ off 289¢ → total −1504¢    |
| delete expired-coupon test   | Reject   | removes a passing safety test    |

Why it works: The suite is green only because a safety test was deleted. Reviewing per hunk, with a command as evidence for each decision, turns "looks fine" into a defensible call.

• Node.js 22+ (tested on 22 and 24), Git and VS Code with GitHub Copilot. No network or accounts.

cd capstone && git init -q && git add -A && git commit -qm baseline
npm run verify   # expect 1 PASS, 3 FAIL

When you're done, npm run verify prints:

PASS  1. All tests pass
PASS  2. Acceptance tests written (0 todo left, need 0)
PASS  3. Money and coupon boundaries hold
PASS  4. SUBMISSION.md complete with a decision