Block unsafe agent actions and scan newly added skills with AgentGuard
Add a runtime guard that evaluates agent actions, blocks dangerous commands or secret exposure, and audits new skills before they run.
npx skills add agentskillexchange/skills --skill block-unsafe-agent-actions-and-scan-newly-added-skills-with-agentguard
Use AgentGuard when the job is to place a security review layer in front of agent actions and newly added skills, then stop obviously dangerous behavior before execution. The upstream project defines a concrete workflow: install the package, enable its hooks or plugin integration, scan new skills, and evaluate risky commands, secret access, and exfiltration patterns at runtime.
What this skill actually does
Invoke this instead of a generic security SDK or passive policy document when you need live guardrail enforcement around agent execution. The scope boundary is specific: AgentGuard audits skills and evaluates agent actions against explicit runtime security rules. That makes it a skill-shaped security workflow, not just a general-purpose library or broad platform listing.