test-manager
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill features a goal-based UI testing profile that constructs prompts for AI agents by interpolating user-defined fields from test cases. This creates a surface where data processed by the skill could potentially contain instructions aimed at subverting the testing agent's constraints.
- Ingestion points: Test case fields (Objective, Preconditions, Runner Instructions) within CASES.md are processed at runtime.
- Boundary markers: The prompt template (goal-based-ui-runner-prompt.md) uses structured sections and a defined 'Execution contract' to guide agent behavior.
- Capability inventory: The skill provides instructions for operating UI browsers and recording business outcomes.
- Sanitization: The skill relies on clear instructional boundaries rather than programmatic sanitization of text inputs.
- [COMMAND_EXECUTION]: The Studio component uses child_process.spawn to launch the user's default web browser. This is used strictly for opening the local management interface at a loopback address.
- [DYNAMIC_EXECUTION]: Upon initialization, the skill writes studio.sh (macOS/Linux) and studio.cmd (Windows) to the workspace. These scripts are generated from local templates and are used to launch the Studio server using Node.js.
Audit Metadata