generate-attack-vector
Pass
Audited by Gen Agent Trust Hub on Sep 24, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data to generate potential attack plans, which could allow maliciously formatted inputs to influence the agent's behavior.
- Ingestion points: The inputs
threat_surfaceandtargetin SKILL.md allow external data to enter the agent's context. - Boundary markers: The instructions lack delimiters or explicit directives to ignore instructions embedded within the target or surface descriptions.
- Capability inventory: No specific shell commands or file-writing tools are explicitly invoked in the skill, which limits the immediate risk of the generated vector.
- Sanitization: The skill provides no validation or filtering mechanisms for the provided inputs to prevent command or prompt injection.
Audit Metadata