Readiness And Verification — Claude Code Guide

Criteria, a verification plan, and DoD are different things

Acceptance criteria answer the question of which result must be true. A verification plan describes how to prove it. Task readiness, or task-level DoD, adds checks for scope, diff, documentation, risks, and human review.

A green test confirms only the scenario it checks; by itself, it does not mean the task can be accepted.

A formula for checking every criterion

• Expected state — what must become correct.
• Observable signal — how it will appear.
• Verification method — a test, command, API request, log, or manual step.
• Failure condition — which result blocks acceptance.

| Criterion | Signal | Check | Fail condition |
|---|---|---|---|
| One refund per key | One record | Regression test | Two records created |
| API compatible | Same response structure | API assertion | Client broken |
| No repeated event | One log event | Log check | Event created twice |

L1 verification for a specific task

A task verification plan may include a unit or regression test, an integration scenario, and, when needed, a manual API smoke check. First verify that the regression test reproduces the defect on the problematic version and passes after the fix.

1. Write the acceptance criteria.
2. Limit the permitted scope.
3. Create the verification plan before implementation.
4. Add or update the regression test.
5. Implement the fix.
6. Run local and integration checks.
7. Review the diff and record risks.

Task readiness checklist

• The changes remain within the agreed scope.
• There are no accidental formatting or unrelated changes.
• Relevant tests and integration checks pass.
• The public API is not broken.
• The logs match the expected behavior.
• Risks and limitations are documented.
• A human has read the diff.
• Another developer can reproduce the result.

Artifacts of a reproducible result

• TASK_SPEC.md with the goal, scope, criteria, and verification plan;
• regression test names and data for the smoke check;
• commands, expected results, and fail conditions;
• test results and logs;
• a summary of human review and residual risks.