lyt-knowledge-check
Pass
Audited by Gen Agent Trust Hub on Jul 9, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data including images, CSV/XLSX spreadsheets, and PDF documents provided by the user.
- Ingestion points: SKILL.md specifies support for reading user-uploaded screenshots, tables, and documents.
- Boundary markers: The instructions include a logical filter requiring the agent to prioritize concepts over author opinions, though it lacks technical delimiters (e.g., XML tags or specific 'ignore' markers) for the content itself.
- Capability inventory: The skill possesses no dangerous capabilities; there are no subprocess calls, file-write operations, or network requests identified in the provided configuration.
- Sanitization: No specific sanitization or escaping of the external file content is described.
- [PROMPT_INJECTION]: The skill includes a 'Conclusion Stability Protocol' which is a robust defense against direct prompt injection. It explicitly instructs the agent to categorize user disagreements and forbids changing its diagnostic conclusion based on user tone or persistence alone. This hardens the agent against manipulation attempts aimed at bypassing its evaluation logic.
Audit Metadata