openrouter-skill
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides patterns for processing external, untrusted data which can be used to deliver malicious instructions to the agent.
- Ingestion points:
reference/multimodal.mddetails how to ingest images from external URLs and PDF files.reference/tool-calling.mddescribes ingesting results from external tools back into the prompt context. - Boundary markers: The implementation examples lack explicit delimiters (e.g., XML tags) or instructions that would tell the LLM to ignore potentially malicious content within the processed files or tool outputs.
- Capability inventory: The skill has access to network resources (OpenRouter, Langfuse), file system access for caching results in
.llm_cache, and the ability to execute tools, which could be leveraged if an injection is successful. - Sanitization: There is no evidence of sanitization, filtering, or validation of the text extracted from images or PDFs before it is used to construct the final prompt for the LLM.
Audit Metadata