Problem

OpenClaw already forwards several OpenAI Responses controls like max_output_tokens, reasoning.effort, and service_tier, but it does not currently forward OpenAI's text.verbosity setting.

That makes it harder to use an important OpenAI-native control for answer length/style without resorting to prompt hacks.

Why this matters

text.verbosity is distinct from reasoning depth:

• models can think deeply internally
• but still return a short external answer

This is useful for people who want "think more, say less" behavior from OpenAI models.

Proposed support

Allow model params such as:

agents: {
  defaults: {
    models: {
      "openai/gpt-5.4": {
        params: {
          textVerbosity: "low"
        }
      }
    }
  }
}

and forward that to OpenAI Responses payloads as:

{
  "text": { "verbosity": "low" }
}

Prefer supporting both alias styles:

• textVerbosity
• text_verbosity

Scope

• OpenAI Responses payload shaping
• config/model params passthrough
• tests for payload injection / invalid values / precedence

I already have a patch prepared for this and will open a PR shortly.