slowmist-agent-security-framework

AI skill security review involves evaluating proposed installations for obfuscation, credential access, code execution, and network exfiltration patterns. This framework applies threat-aware checks and red-flag scanning to generate a risk rating with actionable findings before agent activation proceeds.

Threat modeling for AI agent installations is a verification-first security workflow that assesses untrusted content like GitHub repositories, URLs, and on-chain addresses. It identifies supply-chain compromise, social engineering, and prompt injection risks to produce risk-ranked reports for safe agent actions.

Auditing a GitHub repository for supply chain risk requires assessing repository trust through metadata and evidence-based red-flag scanning. This framework evaluates untrusted repositories to identify potential compromise factors and generates a risk level to determine if human approval gates are needed.

You can analyze blockchain addresses for on-chain AML risk factors using this framework's untrusted input analysis feature. It identifies risk factors associated with on-chain addresses and produces risk-ranked reports to guide safe next actions for your AI agent.

This security review framework supports human approval gates for higher-risk cases by enforcing a verification-first workflow. It requires following review guides, red-flag pattern checks, and trust-tier assessments to output a risk rating that triggers human approval when elevated risks are detected.

You need a security review for MCP installation when evaluating skills and MCPs for malicious patterns before activation. This framework checks for obfuscation, credential access, code execution, and network exfiltration risks to prevent AI agents from acting on unsafe external services.