ETL Pipeline

Pass

Audited by Gen Agent Trust Hub on Jun 3, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is designed to ingest data from multiple external sources, creating an attack surface for indirect prompt injection (Category 8).
  • Ingestion points: Data enters the agent's context through database queries (PostgreSQL, MySQL, MongoDB) and API integrations (Stripe, Salesforce, generic REST APIs) as defined in the Source Connectors section of SKILL.md.
  • Boundary markers: The provided instructions and templates do not include explicit boundary markers or delimiters to isolate untrusted external data from the agent's internal logic.
  • Capability inventory: The skill utilizes powerful tools including postgres_query, mysql_query, mongodb_aggregate, bigquery_load, and snowflake_api, which allow the agent to read from and write to production data environments.
  • Sanitization: While the Transformation Layer includes operational cleaning (handling nulls, deduplication), there are no specific security instructions for the agent to sanitize or ignore instructions potentially embedded within the data retrieved from external APIs or databases.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 3, 2026, 02:52 PM
Security Audit — agent-trust-hub — ETL Pipeline