AI-Infra-Guard

AI-Infra-Guard (A.I.G) is an open-source AI red-teaming security testing platform developed by Tencent Zhuque Lab. It consolidates what teams previously assembled from separate scanners into one platform: Agent Scan, Skills Scan, MCP server scan, AI infrastructure vulnerability scan, prompt security evaluation, and LLM jailbreak evaluation, behind a web UI, CLI, REST API, and Docker deployment.

The infra scanner fingerprints AI component web services (such as Ollama or ComfyUI deployments) to detect known CVE vulnerabilities, while the MCP and skill scanners statically audit MCP services and agent skills for issues like tool poisoning before release. The skill-scan subproject ships SARIF 2.1.0 output and is bundled in OpenClaw's ClawScan/ClawHub security harness. One listing puts AI-infra coverage at 146 components with 2000+ rules (community-reported). The repository is Apache 2.0 licensed and, per GitHub repository details, carries roughly 6,800 stars with 60 releases (GitHub-reported, changing over time).

The honest picture: this is a security scanner for AI systems, not a general-purpose coding tool. Its value is concentrated in teams shipping agents, skills, or MCP servers to production. Community-reported figures and rule counts should be treated as approximate, and red-team results always need human review before being treated as pass/fail gates.

Deploy via Docker (docker-compose.yml in the repo) or run the CLI; web UI served at the backend port.