Refine Prompt

POST /v2/agents/prompt/refine

Refine Prompt

cURL

curl --request POST \
  --url https://api.velt.dev/v2/agents/prompt/refine \
  --header 'Content-Type: application/json' \
  --header 'x-velt-api-key: <x-velt-api-key>' \
  --header 'x-velt-auth-token: <x-velt-auth-token>' \
  --data '
{
  "data": {
    "analysisPrompt": "<string>",
    "demoFeedback": [
      {}
    ],
    "provider": "<string>"
  }
}
'

{
  "result": {
    "status": "success",
    "message": "Prompt refined successfully",
    "data": {
      "refinerResult": {
        "analysis_prompt": "## Objective\n...\n\n## Detection Logic\n...",
        "response_descriptions": {}
      }
    }
  }
}

Use this API to refine an existing analysis prompt based on feedback from demo runs. You provide the current prompt and a list of demos with the user’s feedback on what went wrong; the engine returns a refined analysis prompt and updated response descriptions.

This is the iteration step in the prompt-design loop: run Validate Prompt to expand a simple instruction, run the demos against the resulting prompt, then refine when the demos don’t pass.

Endpoint

POST https://api.velt.dev/v2/agents/prompt/refine

Headers

x-velt-api-key · string · required

Your API key.

x-velt-auth-token · string · required

Your Auth Token.

Body

Params

data · object · required

Show properties

analysisPrompt · string · required

Min 1 char. The current analysis prompt to refine. Typically the analysis_prompt returned by Validate Prompt.

demoFeedback · object[] · required

Min 1 entry. Demo feedback items.

Each entry:

Field	Type	Required	Description
demoId	string	yes	Min 1 char. Stable demo identifier.
demoTitle	string	yes	Min 1 char. Human-readable demo title.
demoHtml	string	yes	Min 1 char. HTML content the demo was run against.
demoExpected	string	yes	Min 1 char. The expected outcome for this demo.
feedback	string	yes	Min 1 char. Free-form feedback explaining what was wrong.

provider · string

LLM provider override: "gemini" or "claude".

Example Requests

Refine after a missed detection

{
  "data": {
    "analysisPrompt": "## Objective\nIdentify all broken links on the page.\n\n## Detection Logic\nCheck HTTP status codes for non-2xx responses.",
    "demoFeedback": [
      {
        "demoId": "demo-1",
        "demoTitle": "Mailto link with malformed address",
        "demoHtml": "<a href=\"mailto:invalid email@\">Email us</a>",
        "demoExpected": "detected",
        "feedback": "The agent missed this — malformed mailto links should also count as broken links, not just non-2xx HTTP responses."
      }
    ]
  }
}

Response

Success Response

{
  "result": {
    "status": "success",
    "message": "Prompt refined successfully",
    "data": {
      "refinerResult": {
        "analysis_prompt": "## Objective\nIdentify all broken or non-functional hyperlinks on the page, including malformed mailto and tel links.\n\n## Detection Logic\nFor http(s) URLs, check HTTP status codes for non-2xx responses. For mailto links, validate the email address format. For tel links, validate the phone number format.",
        "response_descriptions": {
          "title": "Brief description of the broken link",
          "severity": "critical for 5xx, high for 4xx and malformed mailto/tel, medium for timeouts, low for redirects",
          "targetText": "The anchor text of the broken link",
          "suggestion": "The correct URL, email address, or phone number"
        }
      }
    }
  }
}

Field	Type	Description
data.refinerResult.analysis_prompt	string	The refined analysis prompt.
data.refinerResult.response_descriptions	object	Updated per-field descriptions for AI response shaping.

Failure Response

{
  "error": {
    "message": "ERROR_MESSAGE",
    "status": "INVALID_ARGUMENT"
  }
}

Errors: INVALID_ARGUMENT (missing required fields, empty demoFeedback array, or any demo entry missing required fields).

{
  "result": {
    "status": "success",
    "message": "Prompt refined successfully",
    "data": {
      "refinerResult": {
        "analysis_prompt": "## Objective\n...\n\n## Detection Logic\n...",
        "response_descriptions": {}
      }
    }
  }
}