A structured state machine engine for AI agents to execute complex, multi-step workflows without skipping critical steps.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agent-flowforge
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agent-flowforge using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is FlowForge Workflow Runner?
FlowForge is a powerful workflow runner designed for AI agents that require strict adherence to multi-step processes. By utilizing a YAML-based state machine engine, it ensures that complex tasks are executed in a specific order, preventing the common AI pitfall of skipping intermediate steps. This Openclaw Skills integration allows agents to manage long-running tasks, handle branching logic, and maintain state across sessions.
The skill distinguishes between tasks requiring a sub-agent (parallel delegation) and prompt-based tasks (direct reasoning). By using Openclaw Skills like FlowForge, developers can build more resilient AI systems where the main agent acts as an orchestrator, delegating heavy implementation to sub-agents while maintaining the high-level workflow logic.
FlowForge Workflow Runner Use Cases
- Executing complex coding refactors that require sequential file modifications.
- Orchestrating multi-agent research tasks where data collection and analysis must happen in stages.
- Managing long-running CI/CD or DevOps pipelines that survive session restarts.
- Enforcing business logic through branching workflows based on previous step results.
How FlowForge Workflow Runner Works
- The agent identifies the need for a structured process and initiates a workflow using flowforge start.
- The agent polls for the next action via flowforge run to receive a JSON object containing the current task and executor type.
- If the executor is a subagent, the main agent spawns a specialized session; if inline, the agent performs the task directly.
- Once the task result is obtained, the agent advances the state machine using the flowforge advance command, including any branch identifiers.
- The loop repeats until a terminal node is reached, at which point results are reported to the user.
FlowForge Workflow Runner Setup
Ensure the FlowForge CLI is available in your environment. You can verify the installation or set it up using the following commands:
# Check if FlowForge is installed
flowforge --version
# If not found, initialize the setup flow
# (Assuming setup.md logic is mapped to your environment)
FlowForge Workflow Runner Data Schema & Taxonomy
FlowForge uses YAML files to define the state machine. The execution state is persisted locally, allowing workflows to be resumed even after a crash or restart.
| Component | Description |
|---|---|
| Workflow YAML | Defines nodes, tasks, executors (subagent/inline), and transition logic. |
| Node Task | String-based instructions passed to the executor. |
| Executor Type | Determines if the task is delegated (spawn) or handled in-session (prompt). |
| State Persistence | Managed by the FlowForge engine to track the current active node and history. |
FlowForge Workflow Runner Advanced Features
- Parallel Sub-agent Spawning: Offload heavy implementation tasks to independent sub-agents to unblock the main orchestrator.
- Branching Logic: Support for complex decision trees where the next step depends on the success or failure of a previous node.
- Persistent State Management: Use flowforge active and flowforge status to recover interrupted workflows across different sessions.
- Manual Override Mode: Provides granular control for developers to step through nodes or reset the state machine manually.
SKILL.md
Loading
Related Openclaw Skills
OpenClaw SaaS Builder
A comprehensive orchestration skill for building SaaS projects end-to-end using OpenClaw as the driver and Gemini CLI as a planning copilot.
PowerShell Sandbox
A restricted execution environment for PowerShell scripts that enforces command whitelisting, resource limits, and timeout protection.
React Orchestrator
A sophisticated reasoning framework that orchestrates AI agents using a dual-system approach for both rapid execution and deep strategic planning.
Unreal Engine Integration for OpenClaw
A comprehensive integration suite for connecting OpenClaw agents to Unreal Engine 5.x to automate editor tasks, scaffold C++ plugins, and drive Blueprint-driven workflows.
Interaction Pipeline
A standardized communication protocol that ensures AI agents provide consistent status updates and concise two-line summaries after every action.
Skill Interview Builder
A structured interview agent that automates the creation of ready-to-use AI skill packages by guiding users through a three-round requirement gathering process.
Featured*
