m10-performance
Warn
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: MEDIUMPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION] (MEDIUM): The skill metadata uses an adversarial 'CRITICAL:' prefix in the description field. * Evidence: description: 'CRITICAL: Use for performance optimization. Triggers: ...' * Risk: This formatting is a known prompt injection tactic used to artificially elevate the importance of a skill or override the agent's internal task-selection logic. While the skill's content (covering Rust optimization tools like criterion, perf, and rayon) appears benign and educational, the use of high-severity system-command mimicry in metadata is an adversarial attempt to influence agent behavior.
Audit Metadata