The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands using unity-cli. It specifically mentions using cargo run to execute code within the current workspace as a fallback when no global binary is installed (as seen in references/runtime-checklist.md).
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and processing data from Unity scenes, which could contain instructions intended to influence the agent's behavior.
Ingestion points: Scene hierarchy, component values, and animator states are retrieved through unity-cli subcommands in SKILL.md.
Boundary markers: No explicit delimiters or instructions to treat ingested scene data as untrusted are included in the skill.
Capability inventory: The agent can execute commands via Bash and access the file system via Read, providing a path for potential exploitation of injected instructions.
Sanitization: The instructions do not define any validation or sanitization steps for the data retrieved from Unity before it is presented to the agent.

unity-scene-inspect