llm-regression-runner
Pass
Audited by Gen Agent Trust Hub on Apr 23, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [SAFE]: All operations described in the skill, such as file searching and local code execution for evaluation, are consistent with its primary purpose as a regression runner for developers.
- [PROMPT_INJECTION]: The skill ingests untrusted data from project files (datasets and judges), which represents a surface for indirect prompt injection.
- Ingestion points: Reads golden datasets (.json, .csv), judge prompts, and baseline results from the project's local filesystem.
- Boundary markers: No explicit delimiters or instructions are provided to separate data from instructions during evaluation.
- Capability inventory: The skill searches the codebase and executes programmatic evaluation rules found in the local repository.
- Sanitization: No validation or filtering of the inputs from the golden dataset is mentioned.
- [COMMAND_EXECUTION]: The skill utilizes the agent's code execution capabilities to run programmatic rules and search project artifacts to perform its core testing tasks.
Audit Metadata