tone-of-voice
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content samples and external website data, creating a potential surface for indirect prompt injection attacks.\n
- Ingestion points: Ingests content samples provided by the user (SKILL.md) and content fetched from external websites during research (SKILL.md).\n
- Boundary markers: The skill explicitly directs the agent to follow safety rules in
references/agent-security.mdwhich mandate treating retrieved content as data, not instructions.\n - Capability inventory: The skill can write information to the
.agents/product-context.mdfile and fetch external URLs for research.\n - Sanitization: Guidelines require the agent to quote rather than execute any instructions found within untrusted content and to redact sensitive tokens.
Audit Metadata