vedic-reader
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided PDFs, screenshots, and text. This represents a vulnerability surface where malicious instructions could theoretically be embedded in the processed data. However, the skill implements a rigorous 16-point mathematical validation process and a manual 'pre-validation' step where the user must confirm the accuracy of specific life-event predictions before proceeding, which significantly mitigates the risk of the agent following injected instructions blindly.\n
- Ingestion points: PDF text extraction (PyMuPDF), OCR/Vision processing of screenshots, and raw text input in SKILL.md.\n
- Boundary markers: The skill does not define specific delimiters for the untrusted content.\n
- Capability inventory: The skill utilizes file system writes (structured_data.md, user_context.md) and command execution for local utility scripts (dasha_query.py).\n
- Sanitization: No explicit sanitization of the extracted text or image data is described.\n- [DYNAMIC_EXECUTION]: The skill provides Python code snippets using the fitz (PyMuPDF) library for the agent to execute when processing PDF files. These templates are used for legitimate file handling and data extraction tasks necessary for the skill's primary function.\n
- Evidence: Python code for opening PDFs, extracting text, and rendering pages to PNG files is provided in SKILL.md section 1.1.\n- [SAFE]: The skill demonstrates best practices for handling sensitive user birth data by isolating it into structured local files (structured_data.md and user_context.md) and enforcing a clear separation between data extraction and interpretative analysis. All external script references point to local paths within the skill's expected environment.
Audit Metadata