What is the 'Iron Law' in this skill?

The Iron Law requires that the AI never claims success or task completion without providing fresh, real-time evidence within the same interaction.

Does it rely on cached test results?

No, the skill explicitly mandates that tests be run fresh for every claim, even for 'trivial' changes, to ensure that previous passing states haven't regressed.

Can this skill handle code reviews?

Yes, it includes anti-sycophancy measures that require the agent to verify reviewer feedback against the actual codebase before implementing any suggestions.

What tools does this skill use to verify work?

It utilizes the Read, Bash, Glob, Grep, Edit, and Write tools to interact with the file system and execute validation scripts or tests.

How does this prevent AI hallucinations?

By banning phrases like 'should work' and requiring active command execution to prove claims, it forces the AI to check the actual state of the code before making a statement.

Verification Discipline

Name: Verification Discipline
Author: josephneumann

byjosephneumann

•

Security & Testing

Enforces a strict evidence-based protocol to ensure every completion claim is backed by fresh, verified command output.

The Verification Discipline skill transforms Claude's behavior from speculative to empirical by mandating a five-step verification checklist for every task completion or finding. It eliminates 'hallucinated success' by banning non-committal language and requiring the agent to execute real-time tests or scripts before reporting status. Designed for high-stakes production environments, it prevents common pitfalls like relying on cached test results or blindly accepting reviewer feedback, ensuring that every code change is independently validated against the current state of the repository.

Key Features

01Mandatory 5-step verification checklist for every completion claim

02Real-time command execution and exit code validation

03Strict filter banning non-committal language like 'should work' or 'probably fine'

04Anti-rationalization logic to prevent reliance on cached or historical data

052 GitHub stars

06Anti-sycophancy measures to independently verify peer review feedback

Use Cases

01Validating documentation changes to ensure they haven't inadvertently broken build processes

02Peer-reviewing code changes by independently verifying claims through local execution rather than memory

03Confirming bug fixes actually work by running fresh tests before reporting success

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add josephneumann/claude-corps verify

For use in Claude.ai and ChatGPT