agent-eval
Warn
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install a package directly from an external GitHub repository (joaquinhuigomez/agent-eval) using a specific commit hash. This repository is not associated with a trusted organization or well-known service, which constitutes the installation of unverified code.\n- [COMMAND_EXECUTION]: The core functionality of the skill involves executing shell commands provided in YAML task definitions (e.g., in the 'judge' field). This allows the tool to run arbitrary logic on the host machine.\n- [COMMAND_EXECUTION]: The skill exposes a surface for indirect prompt injection via its task processing pipeline.\n
- Ingestion points: YAML files in the 'tasks/' directory containing 'prompt' and 'judge' definitions.\n
- Boundary markers: None; there are no instructions to use delimiters or ignore embedded instructions when parsing task files.\n
- Capability inventory: The tool runs shell commands like 'pytest', 'npm build', and custom commands provided in the 'command' field.\n
- Sanitization: No evidence of sanitization or validation for the 'command' string in the configuration files is provided.
Audit Metadata