# Human review checklist for AI agent output

- [ ] The output addresses the stated objective and acceptance criteria.
- [ ] Material claims are supported by attached or linked evidence.
- [ ] Facts, assumptions, and inferences are clearly distinguished.
- [ ] Calculations, queries, or code have been checked where practical.
- [ ] The agent stayed within its permissions and access scope.
- [ ] Sensitive or restricted information is handled correctly.
- [ ] Requested changes identify the exact artifact, quote, line, or decision.
- [ ] Every unresolved change has a clear owner or next action.
- [ ] Resolved comments point to the revision that addressed them.
- [ ] The final disposition is recorded: accept, revise, escalate, or stop.