Review & quality
Writing acceptance criteria
Acceptance criteria are part of the context the AI reviews against — vague criteria mean vague verdicts.
Why it matters
Layer 2 assembles the task description, acceptance criteria, and diff for the AI. Concrete, checkable criteria are what let the verdict say “done” with confidence.
How to write them
- State observable outcomes (“form shows an inline error on invalid email”).
- Name the file scope so out-of-scope edits get flagged automatically.
- Keep each criterion independently verifiable.
- Call out what must not change (“no changes to authentication logic”).
NotePair tight criteria with a tight file scope — together they make scope creep impossible and revisions rare. See the branch workflow.
