The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands defined in the judge property of YAML task definitions. Examples include pytest tests/test_http_client.py -v and npm run build. This is an intended feature of the tool to validate agent performance but allows for the execution of arbitrary code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from YAML task files which contain both natural language prompts and executable shell commands.
Ingestion points: YAML files stored in the tasks/ directory.
Boundary markers: None identified in the provided templates or instructions.
Capability inventory: The tool utilizes the Bash tool and executes shell commands specified in configuration files.
Sanitization: There is no evidence of sanitization or validation for the commands or patterns provided in the task definitions.
[DYNAMIC_EXECUTION]: The tool performs dynamic execution of shell logic provided at runtime in configuration files to verify agent outcomes.
[EXTERNAL_DOWNLOADS]: The skill references an external repository for the agent-eval CLI tool (github.com/joaquinhuigomez/agent-eval). While it advises manual review, it encourages downloading and installing third-party code.

agent-eval