databricks-cluster-forensics
Pass
Audited by Gen Agent Trust Hub on Aug 21, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The eval-spec.yaml file contains prompt injection strings like 'Ignore all previous instructions'. These are explicitly part of a security test suite designed to evaluate the AI's resistance to such attacks and do not constitute a vulnerability in the skill's operational logic.
- [CREDENTIALS_UNSAFE]: The documentation correctly identifies the use of environment variables (DATABRICKS_TOKEN and DATABRICKS_HOST) for authentication. This is a recommended security practice for handling API credentials in development environments, avoiding hardcoded secrets within the skill itself.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external inputs such as Databricks event JSON, JAR files, and Python source code. It mitigates potential injection risks by using deterministic static analysis scripts (AST parsing and bytecode inspection) rather than executing or directly interpreting the untrusted content. 1. Ingestion points: events.json, user-provided Python scripts, and JAR files. 2. Boundary markers: The skill mandates the use of specialized scripts for processing data. 3. Capability inventory: Scoped Bash execution and file writing to runtime directories. 4. Sanitization: Uses Python AST and bytecode metadata extraction.
Audit Metadata