grobid-pdf-parsing
Pass
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: Fetches the GROBID source code from its official GitHub repository and provides instructions for pulling the official Docker image for deployment.\n- [DATA_EXPOSURE_AND_EXFILTRATION]: Makes network requests to a local GROBID server (localhost:8070) and interacts with the local file system to read PDF files and save parsed results in XML and JSON formats.\n- [COMMAND_EXECUTION]: Provides examples of using
curlanddockercommands to manage the document processing server and submit jobs.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted academic PDF files, which represents an ingestion point for external content.\n - Ingestion points:
GrobidClient.process_fulltextandbatch_processfunction inSKILL.md.\n - Boundary markers: Not present in the provided code snippets.\n
- Capability inventory: Includes local file system write access and network communication (requests to localhost).\n
- Sanitization: No explicit sanitization or filtering of the extracted text content is demonstrated in the examples.
Audit Metadata