Skill UI

This tool enforces four rigorous coding principles, inspired by Andrej Karpathy's insights on LLM pitfalls. It guides developers to surface assumptions, prioritize simplicity, make surgical changes, and define verifiable goals. Use it when reviewing code, checking diffs, or before committing to ensure high code quality and prevent over-engineering.

This guide provides a structured approach to red-teaming Large Language Models (LLMs) to detect sensitive data leakage from system prompts. It simulates advanced attacks—including prompt injection, instruction override, and encoding tricks—to identify embedded secrets, API keys, and proprietary business logic. Essential for validating adherence to OWASP LLM07 and securing AI applications against data exfiltration.

This skill executes the DeepSWE benchmark, a comprehensive coding-agent evaluation suite. It uses OpenRouter and mini-swe-agent to score various Large Language Models (LLMs) on 113 tasks, providing reproducible results for model comparison, development assessment, and research benchmarking. Use this when a standardized, deep evaluation of LLM coding capabilities is required.

GateGuard is a mandatory pre-tool-use hook designed to significantly enhance the reliability and safety of LLM-generated code and actions. Instead of relying on self-evaluation, it forces the model to perform concrete, verifiable investigations—such as listing all imported files, checking data schemas, or detailing rollback procedures—before any write or destructive command execution. It establishes a three-stage gate (Deny, Force, Allow) to prevent hallucination and ensure context awareness in complex coding environments.

This tool assesses whether coding agents adhere to defined skills, rules, or agent definitions. It automatically generates scenarios with varying prompt strictness, executes the agent, and generates a comprehensive report detailing the compliance rate of tool calls against predefined workflow specifications using advanced LLM classification.