Can Claude skip verification for minor changes?

No, the skill enforces an 'Iron Law' with no exceptions, ensuring that even small changes are validated by fresh evidence.

What does this skill prevent?

It prevents Claude from claiming a task is done or a bug is fixed without providing actual command output as proof, eliminating 'hallucinated' success.

How does the 'Red-Green' cycle work in this skill?

It requires seeing a test fail (proving the bug exists) and then pass (proving the fix works) before the task is considered verified.

Why is this better than standard AI coding workflows?

It builds absolute trust by replacing assumptions with objective evidence, ensuring that every commit or PR is truly ready for review.

Verification Before Completion

Name: Verification Before Completion
Author: yusoofsh

byyusoofsh

セキュリティとテスト

Enforces a strict evidence-based workflow by requiring fresh verification output before any task is marked as complete or successful.

概要

This skill acts as a rigorous quality assurance gate for Claude Code, preventing premature or unsubstantiated claims of task success. It mandates a 'verification-first' approach where every assertion—from passing tests to successful builds—must be backed by fresh command-line evidence and direct output analysis. By eliminating assumptions and 'should-be-fixed' statements, it ensures high-integrity development, reduces regressions, and maintains trust through a non-negotiable red-green verification cycle.

主な機能

Active prevention of vague language like 'should' or 'seems to'
Red-green cycle validation for bug fixes and regressions
Real-time exit code and failure count analysis
Independent verification of delegated agent task reports
Mandatory command execution before any success claims
0 GitHub stars

ユースケース

Validating that a bug fix actually resolves the issue via regression testing
Ensuring all unit tests pass before committing code to a repository
Confirming build success and linter cleanliness before opening a pull request

概要

主な機能

Active prevention of vague language like 'should' or 'seems to'
Red-green cycle validation for bug fixes and regressions
Real-time exit code and failure count analysis
Independent verification of delegated agent task reports
Mandatory command execution before any success claims
0 GitHub stars

ユースケース

Validating that a bug fix actually resolves the issue via regression testing
Ensuring all unit tests pass before committing code to a repository
Confirming build success and linter cleanliness before opening a pull request