verification-before-completion_skill

Overview

This skill enforces a strict verification step before claiming work is complete, making evidence the primary requirement for any success statement. It embeds the rule: run the verifying command, read full output and exit code, then report with evidence. Use it to prevent false positives, broken releases, and misplaced trust in agent reports.

How this skill works

Before any completion claim the skill requires identifying the exact verification command, running it fresh and complete, and inspecting output and exit status. If the output confirms the claim, include the evidence when reporting; if not, report the actual state with the verification output. The gate function halts claims until verification is performed.

When to use it

• Before committing, pushing, or creating a PR
• Before declaring tests, linters, or builds as passing
• Before marking bugs fixed or tasks complete
• Before moving to the next task or delegating results to others
• Whenever an agent reports success or modifies code

Best practices

• Always run the full verification command in the current environment; avoid relying on prior runs
• Capture and include command output and exit codes as evidence in your report
• Follow red-green cycles for regression tests: prove failure, then fix, then prove pass
• Avoid language implying success until verification is done; replace "should" with concrete results
• Treat agent success messages as prompts to verify, not as proof

Example use cases

• Run unit test suite and paste the summary with exit code before claiming all tests pass
• Execute the build command and show exit 0 output before merging a release branch
• Reproduce a reported bug, run the failing test to show it fails, apply a fix, then run tests again and show new pass evidence
• Run the linter and include the zero-error output before stating code is clean
• After an agent-made change, show the VCS diff and run verification commands before accepting the change