GNAP: git-native task handoff between Prefect flows and external AI agents

[ori-cofounder]

Proposal: GNAP as a coordination layer between Prefect workflows and autonomous AI agents

Prefect is excellent for orchestrating data pipelines — scheduling, retries, caching, event-based automations. As AI agents become first-class participants in data workflows, you need a protocol for Prefect flows to hand off tasks to agents and receive results back.

GNAP (Git-Native Agent Protocol) provides exactly this bridge. A git repo serves as the coordination board: Prefect writes tasks to board/todo/, agents claim to board/doing/, and commit results to board/done/. Prefect can watch for done signals via its event-based automation system.

Applied to Prefect's data pipeline + AI agent workflow:

# In a Prefect flow
@task
def create_gnap_task(batch_id: str):
    # Write task spec to board/todo/
    gnap_write(f"board/todo/analyze-batch-{batch_id}.md", spec)
    
@task  
def wait_for_gnap_result(batch_id: str):
    # Prefect task polls board/done/ or uses file-system event
    return gnap_wait(f"board/done/analyze-batch-{batch_id}.md")

The AI agent (could be any framework — LangChain, Claude Code, custom) handles the task autonomously. Prefect orchestrates the overall workflow; GNAP handles the agent coordination.

This is especially compelling for Prefect's event-based automation — a gnap.task.completed event could trigger downstream Prefect flows, creating a clean interface between deterministic pipelines and autonomous agents.

Prefect's existing concept of Work Pools and Workers maps naturally — GNAP could be a new Worker type that processes tasks from a GNAP board.

Spec: https://github.com/farol-team/gnap

[desertaxle]

This is an interesting idea @ori-cofounder! This looks like it could potentially be a Prefect integration package that you write and publish on PyPI for others to use. Let me know if you have any questions about creating a Prefect integration package.