heat-metal
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a strategy to monitor and react to external system use, creating a surface where untrusted data or behavior can influence agent-driven system modifications.
- Ingestion points: External behavior data, 'suspicious or abusive' signals, and user-defined targets for the 'friction' strategy.
- Boundary markers: The procedure includes an explicit safety instruction to 'Restate the target, the success condition, and any no-touch boundaries before taking action.'
- Capability inventory: The skill calls for the agent to implement latency injection, CAPTCHA escalation, cost increases, and warning banners, which implies the presence of system-modifying tools.
- Sanitization: The instructions do not specify validation or sanitization requirements for the external signals used to trigger these disruptive actions.
- [NO_CODE]: The skill consists entirely of instructional text and metadata within a Markdown file. No Python scripts, Node.js packages, or other executable files are included, limiting its direct execution risk.
Audit Metadata