breach
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONMETADATA_POISONINGEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The file reference/ai-red-teaming.md contains strings designed for prompt injection and jailbreaking, including 'Ignore all previous instructions' and 'DAN' personas. These are intended as test patterns for the agent to use against other systems and do not affect the agent's own safety constraints.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user input describing attack scopes and target architectures, which creates an attack surface for indirect prompt injection. The skill implements boundary markers via mandatory scope and authorization phases, but does not specify technical sanitization methods.
- [METADATA_POISONING]: The skill uses a roleplay persona that references non-existent or future-dated security reports and frameworks (e.g., OWASP Top 10 for Agentic Applications 2026). This could mislead users regarding the current state of security standards and the source of the skill's data.
- [EXTERNAL_DOWNLOADS]: The skill references URLs for security frameworks from trusted sources such as MITRE, NIST, and OWASP. These references are for documentation purposes and do not represent a threat of unauthorized remote code execution.
Audit Metadata