standalone-test-repro
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- Indirect Prompt Injection Surface: The skill ingests data from test source files, project metadata, and documentation (e.g., src/tests/README.md, .csproj files) to guide the reproduction process. This creates a potential surface where adversarial content in these repository files could influence the agent's actions or the generated output. The skill includes a 'Scope Gate' to filter supported tests but lacks explicit prompt delimiters for interpolated data.
- Dynamic Code Generation and Execution: The skill programmatically constructs C# files and project definitions (e.g., Program.cs, StandaloneTestRepro.csproj) which are then compiled and executed using repository tools like the bootstrapped dotnet SDK. This runtime code generation is the primary function of the skill but represents a pattern where logic is dynamically assembled and run based on analyzed content.
- Managed Command Execution: The skill invokes system commands such as
dotnet buildandcorerun. It manages the environment for these commands by generating a .env file, including logic to avoid the inclusion of CI secrets and providing specific escaping for variable values, which is a notable safety practice. - Modular Skill Dependencies: The skill dynamically references and loads sub-skills from paths like ../standalone-test-repro-cmd/SKILL.md. This modularity depends on the integrity of the referenced skill files within the repository structure.
Audit Metadata