Agents panel splits streamed messages into many lines when connected to Claude Code via WebSocket

[sshtomar]

Describe the bug

When connecting an external code agent (Claude Code) via WebSocket, the agent’s streamed response is rendered as many separate lines/fragments instead of a continuous message. This makes the chat hard to read.

See attached screenshot.

Environment
OS: macOS 15.0 (Darwin 25.0.0)
Shell: zsh

Agent connection in Marimo: WebSocket ws://localhost:3017
Marimo version: 0.17.0
Browser: Chrome
Python: 3.12.11

Will you submit a PR?

• Yes

Environment

❯ marimo env               
{
  "marimo": "0.17.0",
  "editable": false,
  "location": "/Users/explorer/Inquiro/examples/notebooks/.venv/lib/python3.12/site-packages/marimo",
  "OS": "Darwin",
  "OS Version": "25.0.0",
  "Processor": "arm",
  "Python Version": "3.12.11",
  "Locale": "C",
  "Binaries": {
    "Browser": "141.0.7390.108",
    "Node": "v24.7.0"
  },
  "Dependencies": {
    "click": "8.3.0",
    "docutils": "0.22.2",
    "itsdangerous": "2.2.0",
    "jedi": "0.19.2",
    "markdown": "3.9",
    "narwhals": "2.8.0",
    "packaging": "25.0",
    "psutil": "7.1.0",
    "pygments": "2.19.2",
    "pymdown-extensions": "10.16.1",
    "pyyaml": "6.0.3",
    "starlette": "0.48.0",
    "tomlkit": "0.13.3",
    "typing-extensions": "4.15.0",
    "uvicorn": "0.38.0",
    "websockets": "15.0.1"
  },
  "Optional Dependencies": {
    "pandas": "2.3.3",
    "loro": "1.8.1"
  },
  "Experimental Flags": {
    "performant_table_charts": true,
    "external_agents": true,
    "chat_modes": true,
    "rtc_v2": true,
    "sql_linter": true
  }
}

Code to reproduce

Steps to Reproduce
Start the Claude Code agent bridge: npx stdio-to-ws "npx @zed-industries/claude-code-acp" --port 3017
Open a Marimo notebook and the Agents sidebar.
Connect to the agent via WebSocket at ws://localhost:3017.
Send a multi-sentence prompt (e.g., “Can you add a graph to this notebook?”).
Observe how the response streams in the chat UI.