byted-bytehouse-multimodal-search

Warn

Audited by Gen Agent Trust Hub on Mar 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/search_client.py file builds SQL queries using string interpolation (f-strings) for parameters like table_name and filter_condition. This approach is vulnerable to SQL injection if inputs are controlled by untrusted actors, potentially allowing the execution of unauthorized database commands.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by consuming external text and URLs to generate embeddings and perform searches.\n
  • Ingestion points: Processes text and URLs in scripts/embedding.py and search parameters in scripts/search_client.py.\n
  • Boundary markers: The implementation lacks markers or instructions to isolate or ignore malicious content within processed data.\n
  • Capability inventory: The skill can perform database read/write operations and make network requests.\n
  • Sanitization: There is no validation or sanitization of input data before it influences embedding generation or SQL construction.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 16, 2026, 02:58 PM