Audit agent skills for security risks

Audit AI agents for prompt injection, information leakage, and jailbreaks

High-stakes agents where structured security audits catch jailbreak patterns that static linting misses.

• · agent definition
• · threat scenarios
• · input corpus

• · vulnerability report
• · remediation checklist
• · severity triage

• · LLM security testing framework
• · prompt mutation engine

Agent runnable; threat model documented

• · false positives from benign paraphrasing
• · adversarial examples unknown to test suite
• · non-reproducible findings

• · threat model documentation
• · test case library
• · severity scoring methodology