The Agent Skills Directory

Arbitrary File Write Capability: The rest_request.py script includes a save_raw feature that writes API responses to a file path specified by the raw_output_path input field. Because the script does not perform path sanitization or validation (e.g., checking for directory traversal or restricted paths), it could potentially be directed to overwrite or create files in sensitive locations if the execution environment is not strictly isolated.
Unrestricted Network Request Capability: Although the skill instructions focus on the ENCODE Project API, the underlying script accepts any base_url and path via the input payload. This generic design allows the tool to make requests to arbitrary external or internal network endpoints, which could be utilized for data exfiltration or service discovery beyond the intended scope.
Indirect Prompt Injection Surface: The skill is designed to fetch and process data from external REST APIs. If an external service returns content containing instructions intended to influence the agent's behavior, there is a potential surface for indirect prompt injection. While the script includes a _compact function that truncates long strings, the risk persists if the agent interprets the processed summaries as authoritative instructions.

encode-skill