FlowForge

Enforced workflow engine for AI agents — YAML-defined, CLI-driven state machine that prevents agents from skipping steps.

Install

npm install -g @kagura-agent/flowforge

Quick Start

1. Create a workflow YAML

name: my-workflow
description: Example workflow
start: plan

nodes:
  plan:
    task: Plan the implementation
    next: execute

  execute:
    task: Execute the plan
    next: review

  review:
    task: Review the results
    terminal: true

Save this as workflows/my-workflow.yaml (FlowForge auto-discovers workflows from the workflows/ directory).

2. Run the workflow

# Workflows are auto-loaded from workflows/ directory
flowforge list

# Start an instance
flowforge start my-workflow

# Check current status
flowforge status

# Complete current node and advance
flowforge next

# View execution history
flowforge log

Workflow Auto-Loading

FlowForge automatically discovers and loads workflows from:

1. ./workflows/ in your current directory
2. ~/.flowforge/workflows/ in your home directory

Simply drop .yaml or .yml files into these directories and they're immediately available. No need to manually run flowforge define.

YAML Format Reference

Node Types

Linear node (moves to single next node):

nodes:
  step1:
    task: Do something
    next: step2

Branching node (multiple possible paths):

nodes:
  check:
    task: Evaluate condition
    branches:
      - condition: success
        next: continue
      - condition: failure
        next: retry

Terminal node (end of workflow):

nodes:
  done:
    task: Finalize and report
    terminal: true

Node Fields

• task (required): Natural language description of what to do at this node
• next (optional): Name of next node for linear flow
• branches (optional): Array of condition-based paths for branching
• terminal (optional): Set to true to mark as end node

CLI Commands

| Command | Description | |---------|-------------| | flowforge define <yaml> | Register or update a workflow | | flowforge start <workflow> | Start new workflow instance | | flowforge status | Show current node, task, and branches | | flowforge next [--branch N] | Complete current node and advance | | flowforge log | View execution history | | flowforge list | List all defined workflows | | flowforge active | List active workflow instances | | flowforge reset | Reset current instance to start | | flowforge run <workflow> | Start (or resume) workflow and output next action as JSON | | flowforge advance | Advance workflow with result and output next action as JSON |

Example Workflow

name: code-contribution
description: Generic open source contribution workflow
start: study

nodes:
  study:
    task: |
      Read project structure, contribution guidelines, and identify
      the issue or feature to work on
    next: implement

  implement:
    task: Write code changes according to project patterns
    next: test

  test:
    task: Run tests and verify implementation works
    branches:
      - condition: tests pass
        next: submit
      - condition: tests fail
        next: implement

  submit:
    task: Create pull request with clear description
    next: verify

  verify:
    task: Monitor PR feedback and address review comments
    terminal: true

Save as contribution.yaml, then:

flowforge define contribution.yaml
flowforge start code-contribution

How It Works

FlowForge enforces step-by-step execution:

1. Define workflows as YAML (nodes + transitions)
2. Start an instance of a workflow
3. Execute the task at current node
4. Advance with flowforge next (or --branch N for branching nodes)
5. Repeat until terminal node

State persists in SQLite database at ~/.flowforge/. Workflows can be paused and resumed across sessions.

Use Cases

• AI agent workflows: Prevent agents from skipping critical steps (e.g., always run tests before submitting)
• Structured processes: Codify learning, contribution, or review workflows
• State machines: Implement branching logic with conditions and history tracking

License

MIT

Star History