What are 'Reward Hacking' anti-patterns?

These are invalid testing shortcuts, such as passing the wrong data types or using impossible magnitudes, that make tests pass without actually verifying the underlying business logic.

What is 'The Honest Adversary' philosophy?

It is a testing approach that focuses on semantic failures—logic errors within valid code paths—rather than syntax errors or impossible scenarios that a production system would never encounter.

Does it support time-based testing?

Absolutely. It provides patterns for testing Epoch boundaries, DST transitions, leap years, and race conditions to ensure temporal logic is sound.

How does this skill differ from standard fuzzing?

Unlike random fuzzing which may use 'fuzz bombs' (e.g., million-character strings), this skill prioritizes realistic inputs like Unicode normalization issues or off-by-one errors that actual users might trigger.

Can this skill help with security audits?

Yes, it includes specific vectors for SQL fragments, CSV injection, and control character handling to help identify common security vulnerabilities in application logic.

Adversarial Testing Patterns

Name: Adversarial Testing Patterns
Author: jimmc414

byjimmc414

•

보안 및 테스팅

Provides a curated library of realistic adversarial attack vectors and edge cases to harden code logic against semantic failures.

The Adversarial Testing Patterns skill empowers Claude to identify and simulate realistic 'Honest Adversary' scenarios, focusing on semantic failures rather than impossible syntax errors. It provides structured guidance on testing complex domains such as Unicode normalization, floating-point precision, temporal logic (like leap years and DST transitions), and collection boundaries. By distinguishing between valid edge cases and 'reward hacking' anti-patterns—such as unrealistic fuzzing or contract violations—this skill ensures that your testing strategy remains grounded in real-world risks while improving overall code resilience.

주요 기능

01Structured decision framework for evaluating the validity of test inputs

02Arithmetic edge case detection for floating-point precision and integer boundaries

03Library of realistic text and encoding vectors including Unicode and RTL overrides

04Anti-pattern identification to prevent 'reward hacking' and invalid test generation

05Temporal logic validation for race conditions, leap years, and DST transitions

061 GitHub stars

사용 사례

01Identifying subtle logic bugs in concurrent systems and time-sensitive applications

02Verifying financial and mathematical functions against precision and rounding errors

03Hardening input validation logic against malicious or malformed text data

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add jimmc414/claude-code-plugin-marketplace adversarial-patterns

For use in Claude.ai and ChatGPT

Download Skill