writing-style
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (emails, Slack messages, memos) to determine writing style. This creates a surface for indirect prompt injection where instructions embedded in those communications could attempt to influence the agent.
- Ingestion points: Sent emails, Slack messages, and past documents are analyzed as style references.
- Boundary markers: Not explicitly defined within the text, though it relies on external confirmation policies.
- Capability inventory: The skill implies the ability to "save or send" drafts through integration with other channel-specific skills.
- Sanitization: Instructions include a requirement to "Keep unrelated details from samples private" and to verify any native draft or send before claiming it happened. It also instructs the agent to follow a
<confirmation_policy>and the specific email or channel skill before sending.
Audit Metadata