duckdb-adbc

Pass

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The instructions describe how to install the 'dbc' utility by downloading and piping a shell script to the command line from the vendor's domain. \n
  • Evidence: curl -LsSf https://dbc.columnar.tech/install.sh | sh\n- [COMMAND_EXECUTION]: For Windows users, the skill documentation suggests using PowerShell to download and execute a remote installation script while bypassing execution policies. \n
  • Evidence: powershell -ExecutionPolicy ByPass -c irm https://dbc.columnar.tech/install.ps1 | iex\n- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from remote database systems, creating a surface where instructions embedded in external database content could be processed by the agent. \n
  • Ingestion points: Data retrieved through the read_adbc function or ATTACH command from various ADBC-supported drivers like Snowflake, BigQuery, or PostgreSQL. \n
  • Boundary markers: No explicit delimiters are specified to isolate remote database output from the primary instruction stream. \n
  • Capability inventory: The skill enables execution of arbitrary DDL/DML statements via adbc_execute and interaction with the local DuckDB file system. \n
  • Sanitization: No specific mechanisms for filtering or sanitizing content from the external databases are documented.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 1, 2026, 10:14 AM
Security Audit — agent-trust-hub — duckdb-adbc