verify-before-done

What This Skill Does

The verify-before-done skill acts as an uncompromising quality gate for the OpenClaw agent. It enforces a strict "Verify, then Claim" workflow to prevent common AI failure modes such as false completion claims, unverified code assumptions, and premature task reporting. This skill mandates that for every claim of success—whether it be code compilation, test passing, or bug resolution—the agent must perform an active, real-time verification process rather than relying on prior memory or assumptions. By requiring explicit evidence (e.g., exit codes, specific test outputs, or diff reports) before acknowledging task completion, it significantly reduces technical debt and prevents the propagation of "ghost" fixes.

Installation

You can install this skill directly via the ClawHub CLI. Ensure your environment has the necessary permissions to execute verify-before-done protocols:

clawhub install openclaw/skills/skills/kjaylee/verify-before-done

Use Cases

• Continuous Integration/Deployment: Ensuring that builds or deployments are only marked as successful after an explicit check of the status code and deployment logs.
• Refactoring and Bug Fixes: Validating that a bug is actually resolved by executing a targeted reproduction test rather than assuming the fix works based on code changes alone.
• Agent Delegation: Cross-verifying reports from sub-agents to ensure that their "done" status is grounded in objective evidence before communicating the progress to the end-user.
• Quality Control Reports: Providing stakeholders with irrefutable evidence-backed summaries rather than ambiguous claims of progress.

Example Prompts

1. "I've modified the authentication middleware. Please apply the verify-before-done protocol to confirm the login flow is fully functional and report the exit code of the integration tests."
2. "After updating the React component, verify the build output for any linting errors and provide a snippet of the build success confirmation before pushing to the main branch."
3. "The sub-agent claimed the database migration is complete. Independently verify the schema state and provide the verification evidence before we proceed to the production deployment."

Tips & Limitations

• Adhere to the Gate Function: Always follow the identification, execution, reading, and verification steps in order. Skipping the execution step is the primary cause of hallucinated completions.
• Beware of Risk Words: Eliminate "I think," "probably," or "looks okay" from your internal thought process. These signal a lapse in rigor.
• Independence is Key: Never rely on a previous run's results. The environment changes; verify in the current state to ensure accuracy.
• Scope: This skill is focused on verification. If you find the verification step is complex, ensure you have appropriate tools installed to execute those specific tests. The skill does not replace the tests themselves, but enforces their execution.