Magentic-One SOP

Appearance

Test ​

Status: PASS (package import confirmed on local + Hetzner server) ​

The core package imports successfully on both local dev and the Hetzner production server. Full multi-agent web task execution requires additional setup.

What was tested ​

bash
pip install autogen-ext[magentic-one]
python -c "from autogen_ext.agents.magentic_one import MagenticOneCoderAgent; print('Magentic-One imported OK')"

Actual output:

Magentic-One imported OK

Version confirmed: autogen-ext 0.7.5

What this test confirms ​

The package installs and the import chain resolves. Magentic-One ships inside autogen-ext, not as its own standalone package.

What is required for full browser task execution ​

  • A Playwright-compatible browser backend (playwright install --with-deps chromium)
  • An LLM API key (OpenAI, Azure, or compatible)
  • A browser runtime accessible to the Playwright install

Smoke test: import chain verification ​

bash
python -c "
from autogen_ext.agents.magentic_one import MagenticOneCoderAgent
print('Magentic-One imported OK')
"

Expected output:

Magentic-One imported OK

Smoke test: Playwright check (requires separate install) ​

bash
python -c "
from playwright.sync_api import sync_playwright
with sync_playwright() as p:
    browser = p.chromium.launch(headless=True)
    browser.close()
print('Playwright + Chromium OK')
"

Expected output:

Playwright + Chromium OK

If Playwright cannot find Chromium you will see:

playwright._impl._errors.Error: Executable doesn't exist at ...

Fix: run playwright install --with-deps chromium.

Full end-to-end test (requires LLM API key) ​

bash
OPENAI_API_KEY=sk-... python -c "
import asyncio
from autogen_ext.models.openai import OpenAIChatCompletionClient
from autogen_ext.teams.magentic_one import MagenticOne
from autogen_agentchat.ui import Console

async def main():
    client = OpenAIChatCompletionClient(model='gpt-4o')
    m1 = MagenticOne(client=client)
    result = await Console(m1.run_stream(task='What is 2 + 2? Reply with just the number.'))
    print('RESULT:', result)

asyncio.run(main())
"

Expected output: Streamed agent turns ending with the answer 4 before TERMINATE.

Hetzner (server) verification ​

Date: 2026-07-01

Server: agent-runtime-hetzner-1 (merlino-fleet-hel1)

Verify command:

bash
ssh -o BatchMode=yes -o ConnectTimeout=15 agent-runtime-hetzner-1 \
  'source ~/browser-tools/venv-m1/bin/activate && \
   python -c "import autogen_ext;print(\"autogen_ext ok\")"'

Real output:

autogen_ext ok

Result: PASS. The autogen_ext package imports cleanly in the dedicated venv ~/browser-tools/venv-m1 on the Hetzner server. No reinstall was needed; the fix script applied prior to this verification resolved the environment.

Safety checklist before running real tasks ​

  • [ ] Run inside a Docker container or dedicated virtual machine
  • [ ] No sensitive credentials in environment variables the agent can read
  • [ ] Internet access scoped or firewalled if the task is file-only
  • [ ] Human oversight or approval function set for code execution
  • [ ] Logs reviewed after each run for unexpected shell commands