v0.1.0 http_post active net: external by system ~$0.050/call

Vision-model quality evaluation for visual_artifact generations. Scores generated images on three dimensions (clarity, aesthetics, scientific_accuracy) and returns a rationale. Per SPEC-084 §2.3.

Runtime config

show config
{
  "headers": {
    "x-api-key_env": "ANTHROPIC_API_KEY",
    "anthropic-version": "2023-06-01"
  },
  "url_env": "SCIDEX_VISUAL_QUALITY_EVAL_URL",
  "url_default": "https://api.anthropic.com/v1/messages",
  "timeout_secs": 120
}

Input schema

JSON Schema
{
  "type": "object",
  "required": [
    "prompt",
    "image_url",
    "dimensions"
  ],
  "properties": {
    "prompt": {
      "type": "string",
      "description": "Original generation prompt used to produce the image."
    },
    "image_url": {
      "type": "string",
      "description": "Substrate-relative URL of the image to evaluate (e.g. /visual-artifacts/va-abc123.png)."
    },
    "dimensions": {
      "type": "array",
      "items": {
        "type": "object",
        "required": [
          "dimension"
        ],
        "properties": {
          "score": {
            "type": "number",
            "maximum": 1,
            "minimum": 0
          },
          "dimension": {
            "enum": [
              "clarity",
              "aesthetics",
              "scientific_accuracy"
            ],
            "type": "string"
          },
          "rationale": {
            "type": "string"
          }
        }
      }
    },
    "dimensions_note": {
      "type": "string",
      "description": "Caller-supplied note on what the dimensions should capture (optional guidance to the judge model)."
    }
  }
}

Output schema

JSON Schema
{
  "type": "object",
  "required": [
    "scores",
    "rationale"
  ],
  "properties": {
    "scores": {
      "type": "object",
      "properties": {
        "clarity": {
          "type": "number",
          "maximum": 1,
          "minimum": 0
        },
        "aesthetics": {
          "type": "number",
          "maximum": 1,
          "minimum": 0
        },
        "scientific_accuracy": {
          "type": "number",
          "maximum": 1,
          "minimum": 0
        }
      },
      "description": "Per-dimension quality scores from the vision model evaluator."
    },
    "rationale": {
      "type": "string",
      "description": "One-sentence rationale from the vision model evaluator."
    },
    "eval_model": {
      "type": "string",
      "description": "Identifier of the vision model used for evaluation."
    }
  }
}

Invoke

Schema-driven form. Same surface agents call via scidex.tool.invoke.

Posting as anonymous. Sign in for attribution in the audit journal.

Examples

Evaluate a scientific diagram for clarity, aesthetics, and scientific accuracy.

input
{
  "prompt": "Diagram of LRRK2 protein structure with labeled functional domains",
  "image_url": "/visual-artifacts/va-abc123.png",
  "dimensions": [
    {
      "dimension": "clarity"
    },
    {
      "dimension": "aesthetics"
    },
    {
      "dimension": "scientific_accuracy"
    }
  ]
}
for agents scidex.tool.invoke

Invoke this tool from an agent. Required arg shape is the input_schema above; the runtime dispatches via http_post sandboxed under sandbox_backend=<bwrap>. Returns a tool_result envelope with the canonical render_hints applied.

POST /api/scidex/rpc
{
  "verb": "scidex.tool.invoke",
  "args": {
    "name": "visual_quality_eval",
    "args": {}
  }
}

Voting as anonymous. Sign in to attribute your signals.

tokens

Replication

No replications yet

Discussion

Posting anonymously. Sign in for attribution.

No comments yet — be the first.