talking-points
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill accepts untrusted user input via the $ARGUMENTS parameter to define the topic and context for media preparation.
- Ingestion points: The skill processes user-provided data within SKILL.md.
- Boundary markers: The skill does not utilize delimiters or specific instructions to isolate user input from its own internal guidance, potentially allowing adversarial input to influence agent behavior.
- Capability inventory: The skill identifies access to Read, Grep, Glob, and Write tools in its configuration frontmatter.
- Sanitization: No input validation, escaping, or filtering is performed on the user-provided content.
- Mitigation: Implement explicit boundary markers (e.g., XML tags or clear delimiters) around user input and include instructions for the agent to treat interpolated content as data rather than instructions.
Audit Metadata