capability:vision-input

{
  "id": "capability:vision-input",
  "_kind": "Capability",
  "_file": "capabilities/capabilities/vision-input.yaml",
  "_cluster": "capabilities",
  "attributes": {
    "displayName": "Vision input",
    "description": "Ability to accept image inputs (screenshots, photos, pasted clipboard\nimages) as part of a user/agent turn. Required by interaction primitives\nthat paste or attach visual content. Distinct from `text-streaming` or\nother text-only input modalities.\n",
    "appliesToNodeKinds": [
      "ModelVersion",
      "AgentRuntimeImpl"
    ],
    "category": "modality"
  },
  "outgoingEdges": [],
  "incomingEdges": [
    {
      "from": "interaction-primitive:paste-image",
      "to": "capability:vision-input",
      "kind": "requires_capability"
    },
    {
      "from": "agent-version:claude-code@1.x",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "Image input is backed by official Claude multimodal documentation, with lower confidence than the Claude Code-specific docs used elsewhere."
      }
    },
    {
      "from": "agent-version:copilot-cli@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "Public image-input evidence is strongest in the Copilot SDK rather than CLI docs."
      }
    },
    {
      "from": "agent-version:cursor@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "Cursor's agent overview documents image-capable reads and image generation."
      }
    },
    {
      "from": "agent-version:gemini-cli@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "Public Gemini CLI evidence describes multimodal CLI support."
      }
    },
    {
      "from": "agent-version:omp@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "Public README advertises image support; technical detail is thinner."
      }
    },
    {
      "from": "agent-version:openclaw@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "OpenClaw fallback metadata."
      }
    },
    {
      "from": "agent-version:opencode@1.x",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "Public OpenCode docs lack strong technical detail for image input."
      }
    },
    {
      "from": "agent-version:pi@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=0.0.0",
        "level": "full",
        "notes": "Pi fallback harness."
      }
    },
    {
      "from": "model:claude-opus-4-7@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=4.7.0 <5.0.0",
        "level": "full",
        "notes": "Opus 4.7 accepts image modality inputs per its modalities list. ",
        "evidenceSourceIds": [
          "evidence:anthropic-models-doc"
        ]
      }
    },
    {
      "from": "model:gpt-4o@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": "*",
        "level": "full",
        "evidenceSourceIds": [
          "evidence:openai-responses-api"
        ]
      }
    },
    {
      "from": "model:gpt-5@current",
      "to": "capability:vision-input",
      "kind": "supports",
      "attributes": {
        "versionRange": ">=5.0.0 <6.0.0",
        "level": "full",
        "notes": "GPT-5 accepts image modality inputs.",
        "evidenceSourceIds": [
          "evidence:openai-responses-api"
        ]
      }
    }
  ]
}