How does this prevent 'hallucinated' success reports?

It requires Claude to read and report the specific exit codes and output strings of verification commands before it is allowed to use words like 'done' or 'fixed'.

Does it support Test-Driven Development (TDD)?

Absolutely. It includes a specific pattern for regression tests that mandates a fail-then-pass cycle to prove a test is actually testing the intended fix.

Can this skill be used for linting and builds?

Yes, it specifically includes patterns for verifying build success (exit 0) and ensuring linters report zero errors before moving to the next task.

What is the 'Iron Law' of this skill?

The Iron Law states that no completion claims can be made without fresh, verified evidence from a command executed in the current session.

Verification Before Completion

Name: Verification Before Completion
Author: tianzecn

bytianzecn

•

セキュリティとテスト

Enforces a rigorous evidence-based workflow requiring command execution and output validation before any task is marked as finished.

The Verification Before Completion skill establishes a mandatory 'gate function' for Claude Code, ensuring that no work is claimed as fixed, passing, or complete without fresh empirical evidence. It moves beyond 'should work' assumptions by requiring Claude to identify specific verification commands, execute them, and analyze the full output before making success assertions. This skill is essential for maintaining high-quality codebases, preventing regressions, and ensuring that all commits and pull requests meet the defined requirements through a strict 'evidence before claims' methodology.

主な機能

01Anti-rationalization safeguards to prevent assumptions and 'partial' checks

02Standardized failure identification for tests, linters, and build processes

03Systematic verification of agent-delegated tasks via VCS diff analysis

041 GitHub stars

05Strict TDD Red-Green cycle enforcement for regression testing

06Mandatory gate function requiring fresh command output before status claims

ユースケース

01Confirming bug fixes through red-green cycles to ensure the root cause is resolved

02Verifying that unit and integration tests pass locally before creating a pull request

03Ensuring code satisfies a line-by-line requirements checklist before task completion

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add tianzecn/myclaudecode verification-before-completion

For use in Claude.ai and ChatGPT

Download Skill