invariant-exploration-director
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill contains no executable code, scripts, or external dependencies. It focuses entirely on a methodology for categorizing creative variables into Invariants, Conditional, and Exploration categories to manage AI output stability and surprise.
- [INDIRECT_PROMPT_INJECTION]: The skill features a surface for indirect prompt injection as it ingests user-supplied creative requirements (ingestion point: Execution Step 1). It establishes boundary markers via a 'Generation Contract' (Execution Step 4) which dictates how the agent should interpret different variable types. However, no dangerous capabilities such as file system writes, shell execution, or network operations are present, and no tools are requested in the skill configuration, limiting the risk to the prompt reasoning layer.
Audit Metadata