paddleocr-text-recognition
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on executing the
paddleocrcommand-line interface to perform image-to-text conversion and PDF parsing. - [EXTERNAL_DOWNLOADS]: The skill installs the
paddleocrpackage using theuvpackage manager during the installation phase. This is a standard dependency for the tool's core functionality. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from external URLs and local files (images and PDFs). This creates an attack surface for indirect prompt injection if the extracted text contains malicious instructions aimed at the agent.
- Ingestion points: Reads content from
file_urlandfile_pathparameters inSKILL.md. - Boundary markers: Not explicitly defined in the provided instructions.
- Capability inventory: Executes the
paddleocrCLI and potentially writes output to a JSON file via the--outputflag. - Sanitization: None mentioned; the skill instructs the agent to show the full extracted content to users.
Audit Metadata