Computer Use Verification Skill Essential for All Agents

Warp introduced how to connect computer and browser usage capabilities to agents in its cloud software factory to reproduce bugs and verify fixes and new features. Computer use is most valuable when provided as a capability during the triage, implementation, and code review stages, rather than as a standalone agent.

The new verify-behavior skill distinguishes between using computer use for desktop and mobile native apps and browser use for web apps. Verification results prioritize video capture, falling back to screenshots when necessary.

The skill operates in two modes:

• reproduce: Reproduce reported bugs to confirm they actually occur.
• verify: Verify that fixed behaviors or new features match requirements.

Computer use is utilized in three stages:

• triage: Reproduce bugs before fixing them.
• implementation: Verify fixes and new features according to specifications.
• review: Provide evidence, such as videos or captures, that the code implements the expected behavior.

Especially for low-risk tasks like pure UI changes, videos demonstrating actual user-perspective behavior reduce code review burden and increase confidence. Additionally, agents can create debugging loops where they check execution results at each implementation step, compare them against specifications, and self-correct until satisfied.

This approach aligns well with spec-driven development using detailed PRODUCT.md files, but repeated computer use can increase costs, requiring usage management. The implementation is available in the open-source cloud-factory-demo and can be installed with the command npx skills add warpdotdev-demos/cloud-factory-demo --skill oz-cloud-factory-demo.