Submit a task
How to describe a goal, what the agent does with it, and what you get back.
A task is one plain-English request that ends in a Pull Request (or a failure log). You write the outcome; the agent figures out the steps.
Write a good task
Be specific about the outcome and how you'll know it's done. The agent does best with a clear acceptance bar.
Add rate limiting to the POST /api/login route: max 5 requests per
minute per IP, return 429 with a JSON error. Add a test. Open a PR.Tips:
- State the done condition. "Add a passing test", "make
pnpm typecheckclean", "the endpoint returns 429" — these give the agent something to check itself against. - Name files or areas if you already know them ("in
apps/api/src/routes"). It speeds things up but isn't required. - One outcome per task. Two unrelated changes are two tasks — easier to review, easier to roll back.
- Mention constraints you care about ("no new dependencies", "keep the existing test style").
What happens after you submit
- Plan. The agent breaks the goal into steps.
- Work. It edits files and runs your tests, builds, and linters in the sandbox.
- Check. It evaluates the result against your done condition. If a step fails and looks recoverable, it retries or replans.
- Open a PR. It commits to
flarecode/task-<id>and opens a Pull Request with a summary of what changed and why.
You can watch this live, or close the tab and come back to the PR.
What you get back
A Pull Request on your repo with:
- the branch and diff,
- a description of the change and the agent's reasoning,
- the cost of the task,
- CI status, once your checks run.
Review it like any other PR. The agent never merges to your default branch — you do, on your terms. (Auto-merge on green CI is an opt-in Pro feature, off by default.)
If a task fails
You get a failure log with what was attempted and where it stopped. No PR opens. Common causes and fixes are in Troubleshooting. Failed tasks aren't billed.