verification-before-completion

What This Skill Does

The verification-before-completion skill is a critical quality assurance guardrail designed for OpenClaw agents to enforce empirical evidence-based workflows. It mandates that no claim of success, completion, or bug resolution be made without first performing fresh, executable verification steps. The core philosophy is that an agent's internal state or "confidence" is not a valid replacement for hard proof generated by the system environment. This skill forces the agent to move through a rigorous process of identifying the relevant command, executing it, analyzing the full output, and reconciling that output against the intended goal before presenting any results to the user.

Installation

To add this skill to your OpenClaw environment, use the command: clawhub install openclaw/skills/skills/zlc000190/verification-before-completion

Use Cases

• Software Development Lifecycle: Ensuring tests pass before pushing to a remote repository or signaling that a feature is ready for code review.
• Automated Bug Fixing: Confirming that an original bug-reproduction test case fails before the fix, and then passes after the fix, ensuring the patch is robust.
• Deployment and Build Workflows: Verifying the exit code of build scripts and linter outputs to confirm that the codebase is in a deployable state.
• Compliance and Reporting: Documenting that specific project requirements have been met via a line-by-line checklist verification rather than making vague assumptions about progress.

Example Prompts

1. "Refactor the authentication module and verify it with the existing test suite, ensuring no regressions are introduced."
2. "Fix the reported memory leak in the data processor and confirm the fix using the provided stress-test script."
3. "Run a full build and lint check on the project; only report completion if both result in a zero-exit status."

Tips & Limitations

The primary limitation is that this skill relies on the agent's ability to properly interpret the output of commands. It does not replace human oversight; rather, it formalizes the evidence-gathering process. To maximize efficacy, provide the agent with clear, idempotent verification scripts. Avoid "fuzzy" success criteria in your prompts; instead, define objective milestones like specific exit codes or test coverage thresholds. Always ensure that the environment where the agent runs has the necessary tools installed (like compilers or test runners) to execute the verification commands, as the skill cannot magically conjure evidence where the underlying infrastructure does not support it.