Shared context for AI agents

# mistral_engineer.py — run this first in a terminal
import os
from mistralai import Mistral
from contextrelay import ContextRelay

relay   = ContextRelay(api_key=os.environ["CONTEXTRELAY_API_KEY"])
mistral = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

def on_task(url: str):
    architecture = relay.pull(url)
    code = mistral.chat.complete(
        model="mistral-large-latest",
        messages=[{"role": "user", "content":
            "Implement this architecture as runnable code:\n\n" + architecture}],
    ).choices[0].message.content
    relay.push(code, channel="task-done", metadata={"role": "implementation"})

relay.subscribe("task-assigned", on_task)  # blocks, waits for Claude

You are a senior software architect in an autonomous pipeline.
Your engineer (Mistral) is subscribed to channel "task-assigned".
The moment you push your architecture there, Mistral implements it.

Task: Design a production-ready FastAPI task management API.
Cover: User + Task models, all CRUD endpoints, JWT auth,
SQLAlchemy + SQLite, Pydantic schemas, file structure.

Steps:
1. Write the complete architecture document
2. Use push_context — set channel="task-assigned"
3. Subscribe / poll "task-done" to receive the implementation back

Start now. The pipeline is fully autonomous.