Large PR review
A checkout refactor implements one split task backed by:
example.txt
file~/.agents/artifacts/shop-api/checkout/spec.md
~/.agents/artifacts/shop-api/checkout/session-refactor-change-plan.md
~/.agents/artifacts/shop-api/checkout/session-refactor-task.md
The task scopes requirements and names the change plan for wave context. The spec remains requirement authority. The worker claims:
example.txt
fileAll checkout session behaviors preserved.
Treat that as a claim, because it is one. A non-implementer inspects changed files and reruns every applicable command. No Suspec review file.
Native judgment
Changed files:
src/checkout/session.tssrc/checkout/payment.tssrc/retry.ts- related tests
Command results:
example.txt
filenpm run test:integration -- active-session -> 1 passed
npm run test:integration -- expired-session -> expected 409, got 500
npm run test:integration -- missing-session -> 1 passed
npm run test:integration -- provider-failure -> 1 passed
AC-002 fails. Charge ordering changed on a money path. src/retry.ts is outside the task's affected
areas. Request changes on the PR. The implementer cannot accept.
suspec check ~/.agents/artifacts/shop-api/checkout/session-refactor-task.md \
--spec ~/.agents/artifacts/shop-api/checkout/spec.md
The checker reports task and spec facts. Clean structure is not clean code.
The implementer fixes AC-002 within scope, removes the retry edit, and records fresh output. A fresh reviewer reruns proof:
example.txt
filenpm run test:integration -- expired-session -> 1 passed
Earlier output is stale. Preserve the verified expired-session regression lesson through native memory when useful, then close the complete transient set.
Need a starting point? Install the skills