embedding-strategies
Pass
Audited by Gen Agent Trust Hub on Apr 1, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill documents strategies for ingesting and processing external data, which is a potential surface for indirect prompt injection intrinsic to the skill's purpose.\n
- Ingestion points: The
DomainEmbeddingPipelineclass inSKILL.mdcontains aprocess_documentsmethod that ingests document content from potentially untrusted sources.\n - Boundary markers: The current templates do not demonstrate the use of delimiters or specific safety instructions to the agent to treat ingested chunks as data rather than instructions.\n
- Capability inventory: The skill's capabilities are restricted to text processing, chunking, and embedding generation. It does not include high-risk tools such as shell access, arbitrary network requests, or sensitive file system operations.\n
- Sanitization: A
_default_preprocessfunction is included to clean text (removing excessive whitespace and special characters), which assists in embedding quality but is not designed as a security filter.
Audit Metadata