Quality and Verification
A spec without evidence is a document. A test without intent is decoration. Quality is what closes the gap between them.
This topic is about evidence. The spec states the target. The tests, checks, and review structure show whether the implementation actually hit it.
Prompt engineering works on the input. Quality and Verification examines the output against the recorded target. Intent Engineering needs both sides of that boundary.
Tests help catch behavior regressions. Acceptance Criterion IDs (AC IDs) keep the spec-to-test link visible when a spec gets edited. The checkpoint checklist prompts reviewers to inspect stale documents, missing archives, and PRs without evidence.
None of these practices are new. What changes with coding agents is the rate of change and how quickly informal review starts missing things.
Chapters
- Tests as Evidence, Not Ritual: test assertions and manual review as evidence against acceptance criteria
- Test Strategy and Convention: the test taxonomy, the convention document the agent reads, choosing cases from a criterion, and the AC registry
- AC IDs and Coverage: bracket-format IDs,
Test-type:field, dual@Tagtagging, the AC registry, positive and negative pairs - Keeping Documentation Up to Date: links from prose to code, drift signals, and checks during implementation and in CI
- Before, During, After: The Three Checkpoints: a short checklist linking preparation, implementation checks, and review to their detailed procedures
- PR Taxonomy: docs, structural, behavioral. Three review styles, one class per PR