What is the core principle of this skill?

The core principle is 'Evidence before assertions,' meaning the AI must run verification commands and confirm the output before claiming work is fixed, finished, or passing.

What happens if a test fails during the verification process?

The AI is required to state the actual status with evidence of the failure, preventing it from rationalizing or ignoring issues to finish a task quickly.

Can this skill be used for manual verification?

Yes, while it focuses on automated commands, it also includes patterns for line-by-line requirement checklists and manual verification of agent-delegated tasks.

How does it improve AI-human collaboration?

It builds trust by ensuring the AI never claims success based on assumptions, requiring actual command output to prove that code changes are effective and safe.

Does this skill work with any programming language?

Yes, it is language-agnostic and applies to any development environment where verification commands (tests, builds, linters) can be executed.

Verification Before Completion

Name: Verification Before Completion
Author: abudhahir

byabudhahir

•

セキュリティとテスト

Enforces a rigorous evidence-based workflow that requires successful command execution and output validation before any task is marked as complete.

The Verification Before Completion skill serves as a critical quality assurance layer for Claude Code, preventing premature or inaccurate success claims by mandating hard evidence for every assertion. It implements a strict 'Gate Function' that requires identifying specific verification commands, executing them, and analyzing the raw output before the AI is allowed to express satisfaction or commit changes. By eliminating vague language like 'should work' and replacing it with fresh test results and build logs, this skill ensures high-integrity development, reduces regressions, and maintains a transparent, trust-based collaboration between the AI and the human developer.

主な機能

01Gate-keeping logic that validates exit codes and failure counts

02Standardized patterns for TDD red-green cycles and regression testing

03Mandatory execution of fresh verification commands before any success claims

04Holistic verification including build, lint, and requirement checklists

05Detection of 'Red Flag' phrases like 'should work' or 'probably fixed'

061 GitHub stars

ユースケース

01Ensuring production-ready builds by validating compilation and linter status

02Validating complex feature completions against a line-by-line requirements checklist

03Verifying bug fixes through rigorous regression tests before committing to the repository

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add abudhahir/superpowers verification-before-completion

For use in Claude.ai and ChatGPT

Download Skill