reducto
Pass
Audited by Gen Agent Trust Hub on Jun 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes
curlfor network communication with the Reducto API andjqfor parsing JSON responses. These actions are legitimate and necessary for the documented functionality of document processing. - [PROMPT_INJECTION]: The skill processes unstructured content from external PDFs and local documents, which constitutes a surface for indirect prompt injection.
- Ingestion points: Data enters the system via external URLs and uploaded local files processed by the
platform.reducto.aiservice. - Boundary markers: The skill does not provide specific instructions to the agent regarding the use of delimiters to separate the parsed document text from system instructions.
- Capability inventory: The skill leverages
curlfor data retrieval and transmission, and reads local file content. - Sanitization: No specific sanitization or 'ignore embedded instructions' prompts are included to mitigate potential adversarial text within parsed documents.
Audit Metadata