ClawFlow

ClawFlow is a Python-first orchestration layer designed to sit in front of OpenClaw. It accepts a high-level prompt, decomposes it into a task graph, allocates agents, executes the graph, and synthesizes a final result.

This repository is scaffolded from the provided SRS with two priorities:

1. Keep the orchestration core independent from OpenClaw specifics.
2. Make the eventual OpenClaw integration a single adapter swap instead of a rewrite.

Current scaffold

• src/clawflow/services: orchestration pipeline components
• src/clawflow/adapters/openclaw: integration boundary and a mock adapter
• src/clawflow/api: FastAPI app for status, pipelines, and execution
• src/clawflow/ui: lightweight dashboard served by the API
• tests: basic orchestration and CLI coverage

Quick start

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
activate-clawflow
clawflow run "launch a social media marketing campaign for a startup"
uvicorn clawflow.api.app:app --reload

CLI

• activate-clawflow: initialize the engine and verify the configured adapter
• clawflow run "<prompt>": run a prompt through the orchestration pipeline
• clawflow status: view engine status
• clawflow pipelines: inspect recent pipeline runs

OpenClaw integration plan

The OpenClawGateway protocol in src/clawflow/adapters/openclaw/base.py is the integration contract. To wire in OpenClaw later, replace the mock gateway with a concrete implementation that:

• discovers agents from the OpenClaw registry
• dispatches task payloads to the OpenClaw execution interface
• collects results, failures, and agent health metrics

The orchestrator and API should not require structural changes once that adapter is implemented.