The Agent Skills Directory

[SAFE]: The skill and its associated Python script do not contain any malicious patterns, unauthorized access attempts, or safety bypass instructions. The logic is focused on performance measurement and metric aggregation using standard Python libraries.
[INDIRECT_PROMPT_INJECTION]: The evaluation framework defines an attack surface by ingesting and processing untrusted data (agent outputs and external ground truth).
Ingestion points: The evaluate method in scripts/evaluator.py processes raw agent outputs, task descriptions, and ground truth dictionaries.
Boundary markers: Absent; the evaluator processes strings directly for heuristic scoring without using delimiters to distinguish data from instructions.
Capability inventory: None detected. The scripts lack capabilities such as network access, file system modifications, or shell command execution, preventing the exploitation of the injection surface.
Sanitization: Absent; input is treated as raw text for keyword matching and level assignment.

evaluation