create-skill-test
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- Indirect Prompt Injection Surface: The skill is designed to ingest external data (specifically "Skill content" and "Scenario hypothesis") to generate YAML evaluation specifications. While standard for this task, this creates a surface where inputs could theoretically attempt to influence the agent's behavior.
- Ingestion points: The "Inputs" table in
SKILL.mdidentifies "Skill content" and "Scenario hypothesis" as required data sources entering the agent's context. - Boundary markers: The instructions do not specify the use of delimiters or specific safety instructions to isolate the processed input content.
- Capability inventory: The workflow involves the execution of local shell commands, Python scripts, and .NET projects (
dotnet run,python,sh) to validate the generated YAML files, as described in Step 10. - Sanitization: The documentation does not explicitly detail sanitization or escaping procedures for the input data before it is interpolated into the test specifications.
- Local Command Execution: The skill guides the agent to execute various local validation tools (e.g.,
check_eval_quality.py,SkillValidator.csproj). These are repository-local tools used to ensure the integrity of the generated tests, which is a common developer task but involves executing commands based on recently generated or modified configurations. - Dynamic Script Generation: The core functionality involves generating YAML specifications that define executable test environments (including setup and cleanup commands). This process follows established project schemas but is a form of dynamic configuration generation that warrants regular review of the generation logic.
Audit Metadata