visidata
Warn
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: VisiData utilizes Python evaluation functionality for core operations like derived columns (=), complex searching (z/), and row selection (z|). These expressions execute in the application environment and have access to internal objects. The tool also supports a .visidatarc startup file containing plain Python code, which represents a persistent dynamic execution surface.
- [COMMAND_EXECUTION]: The skill provides instructions for running the vd CLI, replaying command logs (.vdj files), and integrating the tool into shell pipelines for automated data processing.
- [CREDENTIALS_UNSAFE]: Documentation in formats.md provides examples of connection strings for IMAP, PostgreSQL, and MySQL that include plaintext password placeholders (e.g., 'imap://user:password@host').
- [PROMPT_INJECTION]: The skill is designed to ingest and process various external, potentially untrusted data formats. Evidence: 1. Ingestion: Supports dozens of file formats (formats.md). 2. Boundary markers: Not documented for input processing. 3. Capability inventory: Includes Python evaluation, file system access, and network operations (S3, IMAP). 4. Sanitization: No explicit validation or filtering logic is mentioned.
Audit Metadata