Turn human feedback into agent-ready revision context.
Review AI-drafted specs, plans, docs, maintained knowledge bases, and agent-created review forms as readable artifacts, then give agents structured comments and answers they can act on without guessing.
Commentary keeps humans responsible for approval while API and MCP surfaces make unresolved feedback, form answers, Brain health context, and evaluations usable for the next revision.
- AI-drafted docs
- Forms answers
- Structured comments
- API/MCP handoff
Generated work moves faster than approval context.
Agents can draft a spec quickly, but people still need to check assumptions, acceptance criteria, and rollout impact before implementation starts.
If feedback is scattered across raw diffs and chat, the next agent pass guesses what to change.
A human review loop agents can consume.
Draft Reviews, Forms, Knowledge Brain, and developer APIs keep generated artifacts readable for reviewers and structured enough for tools to fetch next actions.
- Readable generated artifactsReview the spec or plan before an agent turns it into code, tickets, or another patch.
- Structured unresolved commentsKeep objections and requested changes explicit instead of scattering them across chat.
- Human form answersLet agents create review forms, then read permitted answers through API or MCP instead of scraping the UI.
- Brain evaluation contextLet agents inspect changed Brain files, health findings, requested revisions, and evaluation ledgers.
Agents need review context, not scattered instructions.
The best handoff preserves the reviewed artifact, the selected section, and the human decision together.
Chat-driven revision
- The agent guesses which paragraph the feedback applies to.
- Acceptance criteria are buried in separate conversations.
- Reviewers cannot easily confirm the exact follow-up patch.
Review-driven revision
- Feedback is attached to the rendered section that needs work.
- Unresolved comments form a clear next-action queue.
- Latest and Diff modes make the agent revision easy to recheck.
Commentary turns human review into agent-ready context.
Teams comment on the generated document, then agents can fetch the explicit feedback and apply the next revision.
A human and agent review loop.
Keep the person responsible for acceptance in the loop while agents revise from structured feedback.
- 01
Open the generated spec, draft, or PR.
- 02
Review the rendered decision and leave section comments.
- 03
Let the agent fetch actionable feedback and permitted Forms answers through API or MCP.
- 04
Recheck the revision in Latest and Diff modes.
Give agents structured review feedback.
Start with developer surfaces for API/MCP handoff or Draft Reviews for generated work before a PR exists.