ocr
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill incorporates robust defensive instructions to mitigate indirect prompt injection attacks. Standing Rule 2 mandates that all extracted text be treated as untrusted data, and it employs a security fence with unique nonces (e.g.,
<<UNTRUSTED-OCR-CONTENT nonce="...">>) to prevent the agent from inadvertently executing instructions embedded within processed documents. - [EXTERNAL_DOWNLOADS]: The documentation includes instructions for downloading model weights from Hugging Face using standard tools (
hf download). These downloads are for model data required for OCR and are performed from a well-known service using established workflows. - [COMMAND_EXECUTION]: The skill utilizes a specific set of CLI tools (
ocr,uv,docker) to perform OCR tasks and environment checks (ocr doctor). These tools are scoped to the skill's primary purpose and are managed through standard execution paths listed in the allowed-tools configuration. - [DATA_EXFILTRATION]: No patterns of data exfiltration or unauthorized sensitive file access were detected. The skill is designed for local image and PDF processing with output directed to the agent's standard output.
Audit Metadata