The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves asset metadata which can be manipulated to include malicious instructions.
Ingestion points: Asset search filters, names, and labels retrieved via asset_find, asset_get_info, and asset_get_labels (SKILL.md).
Boundary markers: No delimiters or instructions are provided to the agent to treat asset metadata as untrusted data.
Capability inventory: The skill provides powerful file system operations including asset_import, asset_move, and asset_delete (SKILL.md).
Sanitization: There is no evidence of metadata sanitization before it is processed by the agent.
[DYNAMIC_EXECUTION]: The skill supports workflows that involve writing and compiling code at runtime.
Evidence: The asset_refresh skill triggers Unity's script compilation process, allowing dynamically created or modified scripts to be executed within the editor environment (SKILL.md).
[COMMAND_EXECUTION]: The documentation provides examples of shell commands for local status polling.
Evidence: Example includes a curl command to poll localhost for compilation job status (SKILL.md).

unity-asset