The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides patterns for code-based graders that use subprocess.run to execute system commands such as pytest on external repository paths.
[REMOTE_CODE_EXECUTION]: An evaluation example demonstrates the use of Python's exec() function to run agent-generated code strings to verify outcomes.
[PROMPT_INJECTION]: The skill is designed to process untrusted data which facilitates indirect prompt injection attacks.
Ingestion points: Untrusted data enters via outcome dictionaries, transcript lists, repo_path strings, and qa_case objects (SKILL.md).
Boundary markers: The instructions do not define delimiters or provide warnings to ignore embedded instructions within the ingested data.
Capability inventory: The skill utilizes subprocess.run and exec() to perform system-level operations based on the processed data (SKILL.md).
Sanitization: There is no evidence of sanitization, validation, or escaping of the external content before processing or execution.

agent-evaluation