The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources (Unity project metadata) that could contain malicious instructions.
Ingestion points: Untrusted data enters the agent context via project_get_packages, project_list_shaders, project_get_tags, and project_get_player_settings which retrieve strings from the local Unity environment.
Boundary markers: There are no explicit delimiters or instructions within SKILL.md to warn the agent to ignore instructions embedded in the project metadata results.
Capability inventory: The skill includes the ability to trigger a build process (build_player) and modify project configuration (project_add_tag).
Sanitization: No evidence of sanitization or validation of the retrieved metadata strings is provided in the instruction set.
[COMMAND_EXECUTION]: The build_player skill initiates a build process through BuildPipeline.BuildPlayer. Although this is the primary purpose of the skill and includes a note about output path restrictions to prevent external file writes, it represents an impactful system-level operation that could be abused if the agent is influenced by indirect injections.

unity-project