The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructions in references/runtime-checklist.md advise the agent to use cargo run -- <args> to execute the unity-cli tool if it is not globally installed. This pattern requires the agent to compile and execute code directly from the current directory. In scenarios where an agent is working in an untrusted or compromised repository, this can result in arbitrary code execution during the build or run process.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface by ingesting data from the Unity runtime session.
Ingestion points: Console logs via read_console and test status via get_test_status in SKILL.md.
Boundary markers: There are no explicit markers or delimiters defined to isolate these inputs from the system instructions.
Capability inventory: The agent is authorized to execute complex shell commands via unity-cli in Bash.
Sanitization: No sanitization or validation logic is specified for the ingested logs or status data, allowing potentially malicious strings to influence the agent's logic.
[COMMAND_EXECUTION]: The skill relies on shell command execution via the Bash tool to interact with the Unity editor. While the tool scope is restricted to the unity-cli binary, the automation of input simulation (keyboard, mouse, gamepad) and media capture based on runtime feedback increases the risk profile if the Unity environment is controlled by a third party.

unity-playmode-testing