The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill directs the agent to read and process scene.json (specifically metadata fields like display.title, description, and worldConfiguration) to configure the deployment environment.
Ingestion points: Reads configuration and metadata from scene.json.
Boundary markers: The instructions lack explicit delimiting or instructions for the agent to ignore embedded commands within the ingested data.
Capability inventory: The skill uses shell execution capabilities for npx, sips, and magick commands.
Sanitization: No explicit sanitization or validation steps are provided for the data read from scene.json before it is used to influence agent behavior or command parameters.
[COMMAND_EXECUTION]: The skill suggests running command-line utilities (sips on macOS or magick identify) using file paths derived from the user's scene.json file. This creates a potential command injection vector if the agent does not properly sanitize the path variable before executing the shell command.
[SAFE]: All external URLs, content servers (e.g., worlds-content-server.decentraland.org), and Node.js packages (@dcl/sdk-commands) originate from the official vendor (Decentraland) and are used for their intended primary purpose. References to ENS domains (app.ens.domains) are to well-known infrastructure services.

deploy-worlds