Fix Agent
Turns review and QA findings into verified edits and opens the pull request.
- Typical run
- ~12 min
- Tools
- Git · GitHub API · Sandbox build, test, and run
- You provide
- A repository · The source review or QA run · A delivery mode
- Approval
- High-impact: every action waits for your approval card
◤ What it does
Turns review and QA findings into verified edits and opens the pull request.
The Fix Agent takes findings from a code review or a QA run, edits the repository in an isolated sandbox, and runs the project's own build, tests, and server before and after the change. A regression the fix would introduce blocks it. Changes land as a pull request for you to review, deletions included, unless you choose direct delivery.
Needs: GitHub repository write.
Limits: Changes land as a pull request unless direct delivery is chosen; an app that needs configuration the sandbox lacks is verified by build and tests only.
◤ Ask it
In your words. No prompts to engineer.
“Fix the findings from the last code review and open a pull request.”
“Fix the accessibility failures the QA Agent found on the checkout page.”
◤ Missions it runs in
- Audit my software
◤ Questions
About the Fix Agent
How do I know a fix did not break something else?
The agent runs your build, tests, and server before and after every change. If anything that passed before fails after, the fix is blocked and reported instead of shipped.
Can it push straight to my main branch?
Only if you choose direct delivery. The default is a pull request, and the action is gated behind an approval card either way.
◤ Works alongside
Agents that share its missions.
Code Review Agent
Reads a GitHub repository and reviews architecture, correctness, security, and tests.
QA Agent
Drives a real browser against your live app, and every finding carries evidence.
Analyst
Reasons over the evidence earlier steps gathered, as a named specialist, labelling every claim.
Evaluator
Re-fetches sources, recomputes figures, and scores every deliverable before you see it.