The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill involves generating complex simulation configurations and prompts based on user-provided design briefs, which creates a functional surface for indirect prompt injection. This risk is managed through mandatory validation workflows and template auditing.
Ingestion points: User design briefs, application installation configurations, and natural language playtest actions described in SKILL.md.
Boundary markers: Explicit instructions to utilize validate_world and validate_world_for_publish to verify all candidate payloads before execution.
Capability inventory: A set of MCP tools for creating, patching, and publishing world data, as well as initiating playtests.
Sanitization: Guidance to audit {{...}} tokens against config.initFields and instructions to rely on platform-side validation and moderation for all asset uploads.
[COMMAND_EXECUTION]: The skill requires the agent to perform an HTTP PUT request to a signed upload URL for world cover assets. This network operation is a necessary part of the asset management workflow provided by the vendor's MCP and is triggered only by specific tool outputs.

worldos-authoring