skills/datadrivenconstruction/ddc_skills_for_ai_agents_in_construction/llm-document-extraction/Gen Agent Trust Hub
llm-document-extraction
Pass
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill performs document extraction by interpolating text from user-provided PDFs and images directly into prompts. While this is an indirect prompt injection surface, the implementation mitigates risk by requiring 'json_object' response formats and specific schemas, which prevent the LLM from wandering outside of its data extraction task even if malicious instructions are embedded in the documents.
- [EXTERNAL_DOWNLOADS]: All external dependencies are standard, reputable Python packages (openai, langchain, pdfplumber) used for their intended purposes. There are no patterns of downloading or executing unverified remote code or scripts from untrusted sources.
- [DATA_EXFILTRATION]: Network and filesystem permissions are used solely for the skill's primary function—reading local documents and interacting with authorized API endpoints (OpenAI and a user-provided Qdrant instance). No evidence of credential theft or unauthorized data transmission was found.
Audit Metadata