Agent completion verification compares durable evidence with the task's finish line. It can inspect changed files, diffs, tests, builds, rendered behavior, logs, external state, review standards, and whether required documentation or cleanup exists.
Verification should not rely only on the same model's explanation. Independent checks, direct product inspection, narrow review criteria, and visible failures reduce the chance that plausible text substitutes for a correct result.
ELI5
Agent completion verification checks real evidence that an AI agent finished the requested job instead of accepting the agent's own claim. The evidence can include tests, records, required files, constraints, or a review rubric tied to the requested outcome.
For example, a coding task is complete only when the expected behavior and tests pass without changing the tests to hide a defect. Less objective work may need a defined checklist, sampling, and human judgment.
