tone-rewrite
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from various sources, including chat messages, feedback, and social posts, which creates a potential surface for indirect prompt injection.
- Ingestion points: Untrusted text is ingested via the main prompt instructions in
SKILL.md(e.g., chat messages, feedback, reviews, or social posts). - Boundary markers: The skill lacks explicit boundary markers (like XML tags or specific delimiters) for the input text, though it instructs the agent to extract and verify facts before rewriting.
- Capability inventory: The skill can interact with external note-taking tools (such as the Scratchpad connector or other connected notes apps) to save or update drafts.
- Sanitization: There is no explicit sanitization or filtering logic mentioned for the user-provided text, though the instruction to 'Add nothing they did not offer' and 'Keep the writer' serves as a conceptual guardrail against ignoring the original intent.
Audit Metadata