The Agent Skills Directory

Arbitrary File Write Capability: The script scripts/rest_request.py includes a save_raw feature that writes API response content to a local path. Because the raw_output_path is an optional input field that can be set by the agent, it allows writing to any location the agent has permission to access, which could be used to modify configuration files or shell profiles (e.g., ~/.bashrc).
General-Purpose Network Access: Although the skill is intended for EFO ontology resolution, the underlying rest_request.py script is a generic REST client. The base_url and path parameters are configurable via input, allowing the agent to perform GET or POST requests to any reachable network resource, which could facilitate data exfiltration if the agent is directed to sensitive internal resources.
Indirect Prompt Injection Surface: The skill ingests data from external API responses and processes them for the user. Since external data can contain natural language instructions, there is a potential surface for indirect prompt injection where a malicious API response could attempt to influence the agent's behavior.
Ingestion points: Data is fetched from external APIs via scripts/rest_request.py using the requests library.
Boundary markers: The skill does not define specific boundary markers or instruction-ignoring delimiters for the external data.
Capability inventory: The skill can perform network operations and write files to disk.
Sanitization: The _compact function in the script performs string truncation but does not filter for potential instructions or scripts within the response data.

efo-ontology-skill