How does this skill prevent AI hallucinations in code?

The skill mandates the use of tools like 'Grep' and 'Read' to verify that every method call, service ID, and configuration key actually exists in the source files rather than assuming they work.

When should I use the Paper Test skill instead of just running tests?

Use it before deployment, when reviewing unfamiliar code, or when you don't have a debugger available. It is especially useful for catching edge cases and logic errors that simple unit tests might miss.

What is a 'Paper Test' in the context of this Claude skill?

A paper test is a systematic process where the AI mentally executes code line-by-line using specific, concrete values to trace state changes, verify logic branches, and identify bugs without needing a live runtime.

What kind of flaws can the Paper Test find?

It identifies logic errors (like off-by-one loops), null/undefined access, contract violations, performance anti-patterns (like N+1 queries), and missing error handling.

Paper Test Code Auditor

Name: Paper Test Code Auditor
Author: camoa

bycamoa

•

セキュリティとテスト

Systematically traces code logic line-by-line with concrete values to identify bugs, edge cases, and AI hallucinations.

Paper Test is a specialized Claude Code skill designed for rigorous code validation through manual logic tracing. Instead of just reading code, it simulates execution using specific test cases to uncover subtle logic errors, contract violations, and missing dependencies. It is particularly effective for auditing AI-generated code, verifying complex loops or conditionals, and checking external service integrations before deployment. By forcing a step-by-step trace and using tools to verify that methods and configurations actually exist, it ensures high-reliability code without requiring a full runtime environment.

主な機能

01Physical verification of external dependencies and method signatures using Read and Grep

0217 GitHub stars

03Comprehensive coverage analysis for untested logical paths and branches

04Line-by-line mental execution with concrete input values and state tracking

05Detection of AI hallucinations where non-existent methods or properties are used

06Validation of code contracts, interfaces, and inherited class relationships

ユースケース

01Debugging complex logic like recursion or nested loops without setting up a debugger

02Auditing AI-generated code to ensure it doesn't use invented APIs or incorrect signatures

03Verifying configuration dependencies and YAML mappings against actual source code

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add camoa/claude-skills paper-test

For use in Claude.ai and ChatGPT

Download Skill