security-threat-model

To secure an AI agent against prompt injection, treat all external inputs as untrusted and implement guards against confusable deputy attacks. This Skill enforces those defenses by analyzing code and memory modifications to block malicious instructions.

Threat modeling for agent operations is the process of proactively identifying and mitigating security risks like tool misuse and data exfiltration. You need it before deploying new tools or code that access user data to ensure secure operations.

You enforce least privilege by defining strict boundaries for tool usage, parameter validation, and data access. This Skill analyzes tool definitions to restrict access strictly to necessary data and establishes a clear approval process.

Yes, you can use a structured security-only review process for agent code changes. This Skill analyzes code, tools, and memory modifications against untrusted inputs and privileged actions using detailed threat model templates.

Yes, this threat modeling approach works for memory modifications by analyzing them against untrusted inputs and privileged actions. It ensures that memory updates do not introduce injection vectors or unauthorized data access.