The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and process external project files and assets which could contain malicious instructions. The workflow mitigates this risk by instructing the agent to inspect the descendants of inserted models and verify existing script logic before integration.
Ingestion points: Workflow instructions in SKILL.md direct the agent to inspect the project hierarchy and read script content and project mappings.
Boundary markers: The skill includes an explicit warning to "inspect its descendants before keeping it" for assets from the Creator Store.
Capability inventory: The agent uses tools like execute_luau, multi_edit, insert_asset, and http_get.
Sanitization: The workflow relies on manual inspection and a verification process to confirm that external data has not adversely influenced the project.
[DYNAMIC_EXECUTION]: The workflow incorporates the execute_luau tool to run code within the Roblox engine for testing and verification purposes. This dynamic execution is a core part of the developer workflow and is limited to the Studio environment.
[COMMAND_EXECUTION]: The skill involves the use of tools like multi_edit and insert_asset to modify the Roblox DataModel and filesystem. These operations are fundamental to the task of developing and modifying Roblox projects.
[EXTERNAL_DOWNLOADS]: The skill involves fetching documentation and assets from official Roblox services using tools like http_get and insert_asset. These downloads are used to gather information and incorporate official game assets into the project.

roblox-studio-workflow