The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads data from Unity behavior graph assets, which could serve as a vector for indirect prompt injection if a project contains malicious or untrusted assets. 1. Ingestion points: The skills behavior_graph_info, behavior_agent_get, and behavior_blackboard_list read data from local Unity .asset files. 2. Boundary markers: No explicit delimiters or warnings are defined in the instructions to separate data from instructions. 3. Capability inventory: The module includes the ability to create assets (behavior_graph_create), add components (behavior_agent_add), and modify blackboard variables (behavior_blackboard_set). 4. Sanitization: The behavior_blackboard_set skill validates input values against declared CLR types.
[COMMAND_EXECUTION]: The skill documentation describes using the package_install tool to install the com.unity.behavior package.
[SAFE]: The installation of com.unity.behavior targets an official package from a well-known and trusted service.

unity-behavior