gauntlet-loop
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied goals, constraints, and reference artifacts to drive the iterative workflow, presenting a standard indirect injection surface for agentic systems.
- Ingestion points: User-provided inputs are defined in the 'Outcome brief' (SKILL.md, Execution workflow step 1) and 'Quality bar' (SKILL.md, Execution workflow step 2).
- Boundary markers: Prompt templates in SKILL.md use structured headers (e.g., Goal, Constraints, Quality bar) to delimit user-controlled data.
- Capability inventory: The skill encourages using system tools for artifact inspection, including automated tests, linters, builds, and browser automation (SKILL.md, Tool-use requirements).
- Sanitization: The instructions do not define specific sanitization or escaping mechanisms for interpolated user data.
- [DYNAMIC_EXECUTION]: The workflow involves a 'Builder' agent generating code or content which is then executed or inspected by an 'Evaluation harness'. This dynamic execution is the primary intended purpose for software development and testing tasks.
- Evidence: SKILL.md sections 'Build the evaluation harness before polishing' and 'Execution workflow' step 6 describe generating and running artifacts.
Audit Metadata