The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes execute_luau to run code within the Roblox Studio environment. This tool allows for programmatic interaction with the game engine and state checks, which is a powerful capability within the target environment.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests data from external, potentially untrusted sources such as get_console_output and inspect_instance. An attacker controlling a Roblox experience could embed malicious instructions in game logs or object properties that the agent might inadvertently follow.
Ingestion points: The agent reads external data via get_console_output and inspect_instance (SKILL.md).
Boundary markers: There are no specific instructions or delimiters provided to help the agent distinguish between diagnostic data and instructions.
Capability inventory: The agent has access to execute_luau, user_keyboard_input, and user_mouse_input (SKILL.md), which could be misused if the agent is compromised by injected instructions.
Sanitization: No sanitization or filtering logic is specified for the ingested external data.

roblox-playtest-qa