extracting-keywords
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill describes the use of
xberg cache warm, which downloads embedding models from external sources to the local environment to enable the 'local' embedding provider functionality. - [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for processing untrusted document content (PDFs, text) which represents a potential attack surface where adversarial content in documents could attempt to influence agent behavior.
- Ingestion points: Files processed via the
xberg extractcommand as shown in the examples (e.g.,paper.pdf,report.pdf). - Boundary markers: None explicitly defined in the provided usage instructions.
- Capability inventory: The skill facilitates file system reading, network communication for model downloads, and integration with LLM embedding APIs.
- Sanitization: No specific sanitization or filtering of document content is mentioned before it is processed or passed to downstream tools.
Audit Metadata