Detect and sanitize dangerous prompts and commands for AI agents.
npx skills add https://github.com/numbergroup/AgentGuard --skill agent-guard-numbergroup
Please help me install this Agent Skill. Skill: agent-guard Source: https://github.com/numbergroup/AgentGuard/tree/main Command: npx skills add https://github.com/numbergroup/AgentGuard --skill agent-guard-numbergroup
AgentGuard provides real-time detection and sanitization of dangerous prompts and commands to protect AI agents. It blocks Clinejection-style attacks, command injections, and Unicode bypass attempts, while enabling safe integration with multiple platforms.
Install the skill and start analyzing text for threats with agent-guard analyze.
High-intent search queries and answers about installing and using this skill.
Prompt injection detection uses pattern-based analysis and Unicode normalization to identify malicious commands. AgentGuard processes text locally, classifying threats via a configurable risk score to block injection attempts and sanitize dangerous inputs in real time.
Yes, AgentGuard supports multi-platform integration with both Claude MCP server and OpenClaw. It provides local processing for these integrations, enabling secure AI agent deployments with configurable rate limiting and security logging without requiring external API calls.
Blocking Unicode bypass attempts requires normalizing text encodings before evaluating prompt safety. AgentGuard applies Unicode normalization alongside pattern-based detection to catch obfuscated command injections and malicious prompts that attempt to bypass standard security filters.
Yes, you can configure the risk score used to classify prompt threats and decide actions. AgentGuard allows you to adjust the sensitivity of real-time threat detection, balancing security strictness against false positives when sanitizing dangerous prompts and commands.
Analyzing GitHub issues for security risks involves screening text inputs for injection patterns and command threats. AgentGuard provides GitHub issue screening and risk analysis, evaluating text locally to identify and sanitize dangerous prompts before they reach your AI agents.
zscoleDetects and sanitizes prompt-injection and command-injection threats in AI agent inputs.
PostalocityAnalyze application logs with regex patterns and generate threshold-based alerts.
EmeaAppGbbBuild and orchestrate AI agents with Microsoft Agent Framework.
xianmingyaoAnonymize text and images on-device across 8 languages and 21 privacy categories.
