Coding-agent work, independently verified
Agentic Harness
No setup required
See a verified result in about a minute
Run the real harness on a temporary practice project with a scripted worker. It uses no AI model or API key, and it never touches the project shown above.
Practice run — this demonstrates the workflow and evidence, not model quality.
New goal
What outcome do you want?
Choose a starting point, then describe the result in your own words. No special prompt format is required.
The assistant will plan the work, show progress, and only call it done when the evidence passes. For a question that only needs an answer, use your normal assistant chat.
A normal sentence is enough. Include any limit that matters to you.
Required: add a success check
Pre-filled from Setup. Edit it here to override the default for this run. This check runs independently, and the task is never verified done unless it passes.
Optional scope
Paths are enforced by the built-in model agent. Installed coding-agent CLIs receive them as guidance.
Describe the outcome you want. The app will explain anything else it needs.