{
  "capabilities": {
    "streaming": true
  },
  "defaultInputModes": [
    "text"
  ],
  "defaultOutputModes": [
    "text"
  ],
  "description": "Spatial-aware research agent built on compute-grounded reasoning (CGR). Code computes selected facts over explicit spatial representations for field work analysis. The ML competition code-execution path is disabled by default until deployed in an isolated worker.",
  "name": "Spatial Atlas",
  "preferredTransport": "JSONRPC",
  "protocolVersion": "0.3.0",
  "skills": [
    {
      "description": "Analyzes factory, warehouse, and retail environments from images, videos, PDFs, and documents. Spatial reasoning with structured scene graphs, safety inspection, and formatted reporting.",
      "examples": [
        "Analyze warehouse layout for safety violations"
      ],
      "id": "fieldwork-research",
      "name": "Multimodal Field Research",
      "tags": [
        "spatial",
        "multimodal",
        "vision",
        "fieldwork",
        "research"
      ]
    },
    {
      "description": "Generates a Kaggle-style ML pipeline and a submission file. Generated-code execution is disabled by default and needs an isolated worker.",
      "examples": [
        "Train a model for the spaceship-titanic competition"
      ],
      "id": "ml-engineering",
      "name": "ML Engineering",
      "tags": [
        "ml",
        "kaggle",
        "data-science",
        "code-generation"
      ]
    }
  ],
  "url": "http://127.0.0.1:9019/",
  "version": "1.0.0"
}