The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides Luau code snippets and templates intended for execution within the Roblox Studio environment via a bridge tool (MCP) or standalone scripts. This is the primary and intended function of the skill.
[EXTERNAL_DOWNLOADS]: The skill references official Roblox documentation, community guides from the Roblox DevForum, and configuration files from the official Roblox GitHub repository. These sources are authoritative and well-known within the platform's development ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a surface where user-provided descriptions are used as prompts for procedural model, mesh, and material generation tools.
Ingestion points: User requirements passed to generate_procedural_model, generate_mesh, and generate_material prompts (references/full.md).
Boundary markers: The skill includes explicit instructions to "inspect and reuse," "read back properties," and obtain "explicit consent" before asset insertion (SKILL.md, references/full.md).
Capability inventory: Subprocess execution via execute_luau, file operations via asset generation and image storage tools, and network operations via upload_image (SKILL.md, references/full.md).
Sanitization: There is no evidence of automated input sanitization; the skill relies on manual human-in-the-loop verification gates and read-back checks to mitigate injection risks.
[DATA_EXFILTRATION]: While the skill mentions environment inspection (get_studio_state) and image upload capabilities (upload_image), it provides strict behavioral guidelines to only use permitted sources and obtain permission before uploading content, addressing potential data exposure concerns through process controls.

roblox-building