equivalence-partitioning
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [NO_CODE]: The skill is composed entirely of Markdown, YAML, and JSON configuration files. No scripts (e.g., Python, JavaScript) or binary executables are included, eliminating the risk of direct malicious code execution.
- [SAFE]: The skill includes extensive instructions for the AI agent to prioritize sourced facts over assumptions and to clearly distinguish between inferences and recommendations. It explicitly forbids the agent from editing code or target systems, which mitigates the risk of accidental or malicious system changes.
- [INDIRECT_PROMPT_INJECTION]: The skill provides a framework for analyzing external material (prompts and requirements), which is an inherent injection surface. 1. Ingestion points: User-provided material for partitioning analysis as defined in
prompts/equivalence-partitioning.md. 2. Boundary markers: Requires a mandatory "six-part input audit" to identify and isolate missing, conflicting, or assumed information before processing. 3. Capability inventory: The skill is functionally limited to text-based design recommendations and has no tools or permissions for file system writes, network operations, or subprocess execution. 4. Sanitization: Relies on strict instructional constraints and the absence of execution-capable tools to prevent embedded instructions in the analyzed material from causing harm.
Audit Metadata