The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes Python code in SKILL.md within the ToolEvaluator class that utilizes the requests library to perform GET operations. Specifically, the _test_performance method makes network requests to an api_endpoint defined in the tool_config object. While this is intended for performance benchmarking, it allows the agent to make outbound connections to arbitrary, non-whitelisted domains provided at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing external data without sufficient guardrails.
Ingestion points: The tool_config dictionary is passed to methods like evaluate_tool and _test_performance in SKILL.md.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the tool_config data.
Capability inventory: The skill possesses network request capabilities via the requests library.
Sanitization: There is no evidence of URL validation, domain whitelisting, or input sanitization before the api_endpoint is used in a network call.

Tool Evaluator