schema-query

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • PROMPT_INJECTION (LOW): The skill translates user input into SQL commands via Claude, creating a surface for prompt injection. 1. Ingestion points: User natural language queries via CLI arguments. 2. Boundary markers: None specified in the documentation. 3. Capability inventory: Local DuckDB access and network requests to Anthropic API. 4. Sanitization: None mentioned; the system relies on LLM-side safety filters.
  • COMMAND_EXECUTION (LOW): The skill executes shell commands using uv run and just to perform queries and schema exploration. It also executes dynamically generated SQL against the cdm_store.db database, which could lead to unintended data access if the LLM is successfully manipulated.
  • DATA_EXFILTRATION (SAFE): While the skill accesses a local database and sends data to Anthropic for processing, it uses official API endpoints and does not show evidence of exfiltration to unauthorized domains.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:43 PM