glmocr
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the execution of local Python scripts,
glm_ocr_cli.pyandconfig_setup.py, to interact with the external OCR API and handle initial environment configuration. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (images and PDFs) from local paths or remote URLs. Text extracted from these documents is presented to the agent, creating a surface for indirect prompt injection if the files contain adversarial instructions.
- Ingestion points: Files or URLs passed as arguments to
scripts/glm_ocr_cli.py. - Boundary markers: The skill returns the OCR output as a raw string within a JSON structure. It does not provide specific delimiters to help the agent distinguish between its instructions and the text content extracted from the document.
- Capability inventory: The skill uses the
requestslibrary for network access to the hardcoded Zhipu AI API endpoint. It also has file system read access for processing local images and write access for saving result files. - Sanitization: No specific sanitization or filtering is performed on the text extracted by the OCR model before it is returned to the user or agent.
Audit Metadata