secure-agent

Structural security for AI agents. Load this skill before any sensitive operation.

What It Does

secure-agent gives your AI agent a security mindset grounded in structural controls — not behavioral promises. Based on the attack patterns documented at securingagents.com.

When loaded, an agent will:

• Run the Lethal Trifecta check before processing untrusted input
• Apply blast radius thinking before irreversible actions
• Stay current on active AI agent threats
• Know when to escalate vs. act

Who Should Use It

Any agent that:

• Processes user-supplied content (emails, documents, web pages)
• Has access to sensitive data or credentials
• Can make outbound requests or irreversible writes
• Operates as part of a multi-agent pipeline

How to Activate

In your agent's task or AGENTS.md, add:

Load and follow the secure-agent skill before beginning this task.

Or reference it directly if your platform supports skill loading.

Contents

| File | Purpose | |------|---------| | SKILL.md | Core security rules for the agent | | checklist.md | Pre-task self-audit checklist | | threats/latest.md | Current AI agent threat landscape (updated by Scout) |

Philosophy

Behavioral controls — system prompts, guardrails, output classifiers — fail against motivated adversaries. The only reliable defense is structural: design the blast radius before the agent gets credentials.

This skill helps agents apply that thinking to themselves.

---

Built by securingagents.com · Maintained by Scout 🔍