prometheus-configuration
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from existing codebases, repository instructions, and external documentation, which may contain embedded instructions.
- Ingestion points: The skill inspects available files, runtime versions, and existing conventions within a user's repository (SKILL.md, Workflow Step 1).
- Boundary markers: The instructions explicitly define a "SandBase boundary," directing the agent to keep core work local and only use external tools for documentation or requested inference.
- Capability inventory: The skill utilizes a suite of vendor tools (
sandbase_run,sandbase_discover, etc.) to interact with external APIs and can modify local configuration files. - Sanitization: While explicit sanitization steps aren't detailed, the workflow requires producing bounded, verifiable outcomes and obtaining user authorization before applying deployment configurations.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute specific vendor tools (
sandbase_run,sandbase_inspect) for capability discovery and task execution. These are used to interact with the SandBase API for documentation and model inference tasks.
Audit Metadata