Search Results for "malicious"

HOL Guard is an open-source security layer designed to protect AI agents and their local tool ecosystems at runtime. It evaluates supported actions for secret exposure, prompt injection, unsafe commands, malicious packages, and risky MCP interactions. Safe operations can proceed automatically, while known threats are blocked and ambiguous actions can require human approval. The platform also examines plugins, skills, hooks, package installs, and agent configurations before they are trusted. It integrates with major coding agents including Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and GitHub Copilot CLI. ...

...Its infrastructure scanner fingerprints AI services and checks them against a large library of known vulnerabilities. Agent and skill analysis detects risks involving malicious instructions, unsafe code execution, privilege abuse, compromised tools, and insecure dependencies. Jailbreak evaluation tests language models against multiple prompt-based attack methods and enables cross-model comparison. The platform also provides a web interface, standalone scanning tools, APIs, Docker deployment, and integrations for incorporating security checks into AI workflows.

adversarial-spec is a framework focused on designing and testing systems using adversarial thinking to uncover weaknesses and improve robustness. It encourages developers to define specifications that anticipate failure modes, edge cases, and malicious inputs before implementing solutions. The project emphasizes proactive design, ensuring that systems are built with resilience in mind from the beginning. It provides structured approaches for identifying vulnerabilities and stress-testing assumptions. The framework can be applied across domains, including software development, AI systems, and security workflows. ...