The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to manage the Decentraland development environment, including npm run start to launch the Explorer, pkill and pgrep for process management, and claude mcp add to configure the MCP server within the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data originating from the scene runtime, creating an attack surface for indirect prompt injection.
Ingestion points: The agent reads scene logs via get_scene_logs (SKILL.md) and examines ECS entity details using get_entity_details (reference/interaction.md).
Boundary markers: There are no explicit instructions for the agent to use delimiters or ignore embedded instructions when evaluating these inputs.
Capability inventory: The agent has permissions to execute shell commands, modify the filesystem (src/), and change MCP configuration.
Sanitization: No sanitization of log output or entity data is mentioned before it is processed by the agent.
[DYNAMIC_EXECUTION]: The helper script scripts/screenshot.sh generates a Python script at runtime using a heredoc and executes it with python3. This script is used to decode base64 image data and parse JSON-RPC responses from the local MCP server.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading 3D assets from well-known sources like kenney.nl and itch.io (reference/assets.md). It also uses npm install and npx skills add to fetch vendor-aligned packages and additional skills from the decentraland organization.

unity-explorer-mcp