Skip to content

Jix672/cursor-composer-bridge

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 

Repository files navigation

TaskForge CLI: AI-Powered Multi-Model Code Delegation Engine

Download

License: MIT Python Version Node Version PRs Welcome Maintenance

TaskForge CLI transforms your terminal into a command center for AI-powered code generation. Rather than juggling multiple AI coding assistants manually, TaskForge acts as a digital forger's hammer – striking once to delegate complex coding tasks across multiple AI models simultaneously. Inspired by the concept of orchestrating Cursor from Claude Code, this tool elevates that idea into a standalone, multi-model delegation system that works with OpenAI, Claude, and local models.

🎯 The Core Philosophy

"Why teach one apprentice when you can command a guild of master craftsmen?"

TaskForge embodies the principle of delegated intelligence. Instead of manually prompting individual AI tools, you define a task once and TaskForge distributes it to multiple AI models. Each model works on a different component, and the results merge into a cohesive solution. This is parallelized cognitive labor – your ideas become blueprints, and the blueprints become executable code without you writing a single character manually.

🔄 Architecture Overview

graph TD
    A[User Terminal/CLI] --> B{TaskForge Core Engine}
    B --> C[Task Decomposer]
    B --> D[Model Orchestrator]
    B --> E[Output Merger]
    C --> F[Component 1]
    C --> G[Component 2]
    C --> H[Component N]
    D --> I[OpenAI GPT-4 / o3-mini]
    D --> J[Claude Opus / Sonnet]
    D --> K[Cursor Composer 2]
    D --> L[Local CodeLlama / DeepSeek]
    E --> M[Unified Code Output]
    M --> N[File System / Git Commit]
    style A fill:#4a90d9,stroke:#1a1a2e,color:#fff
    style B fill:#16213e,stroke:#0f3460,color:#fff
    style I fill:#ff6600,stroke:#1a1a2e,color:#fff
    style J fill:#6f42c1,stroke:#1a1a2e,color:#fff
    style K fill:#00bcd4,stroke:#1a1a2e,color:#fff
    style L fill:#4caf50,stroke:#1a1a2e,color:#fff
Loading

📦 Installation

Prerequisites

  • Python 3.9+ (for core engine)
  • Node.js 18+ (for Cursor integration)
  • An OpenAI API key (optional but recommended)
  • A Claude API key (optional but recommended)

Quick Start

# Install via pip
pip install taskforge-cli

# Or via npm for Node-centric projects
npm install -g taskforge-cli

# Verify installation
taskforge --version

💻 Example Profile Configuration

TaskForge uses YAML-based profiles to define which models handle which tasks. Here's a sample profile that delegates like a seasoned project manager:

# ~/.taskforge/profiles/default.yml
profile_name: "full-stack-workbench"
version: "2.0"

delegation_rules:
  - task_pattern: "frontend.*component|React|Vue|UI"
    model: "claude-sonnet-4"
    priority: 1
    merge_strategy: "replace_existing"

  - task_pattern: "backend.*API|database|endpoint"
    model: "openai-gpt-4o"
    priority: 2
    merge_strategy: "append"

  - task_pattern: "testing|unit.test|integration.test"
    model: "cursor-composer-2"
    priority: 3
    merge_strategy: "insert_before"

  - task_pattern: "documentation|README|docs"
    model: "claude-opus"
    priority: 4
    merge_strategy: "replace_existing"

  - task_pattern: ".*"  # Default fallback
    model: "local-codellama-34b"
    priority: 99
    merge_strategy: "append"

output_settings:
  merge_directory: "./generated"
  conflict_resolution: "manual_review"
  max_iterations_per_model: 3
  timeout_per_model: 120  # seconds

🚀 Example Console Invocation

Deploy a full-stack application with a single command – TaskForge handles the rest:

# Basic invocation: build a microservices architecture
taskforge run "Create a microservices-based e-commerce backend with Node.js, Express, and PostgreSQL. Include authentication, product catalog, and order management."

# Advanced: specify profile and output format
taskforge run \
  --profile "full-stack-workbench" \
  --task "Generate a responsive dashboard UI using React and Tailwind CSS with 5 data visualization components" \
  --output-format "react-tsx" \
  --verbose \
  --auto-merge

# Batch mode: process multiple tasks from a file
taskforge batch --input ./tasks/migration_plan_2026.txt --profile "migration-expert"

📊 OS Compatibility Table

Operating System Version Status Notes
🐧 Linux Ubuntu 22.04+ ✅ Full Support Native performance
🐧 Linux Fedora 38+ ✅ Full Support Requires Python 3.9+
🍎 macOS Ventura 13+ ✅ Full Support Apple Silicon optimized
🍎 macOS Monterey 12 ✅ Supported Intel & M-series
🪟 Windows 11 ✅ Full Support WSL2 recommended
🪟 Windows 10 ⚠️ Limited Use Windows Terminal
🐧 Linux Alpine (Docker) ✅ Supported Minimal footprint

✨ Feature List

Core Capabilities

  • Multi-Model Delegation Engine – Distributing tasks across OpenAI, Claude, Cursor, and local models simultaneously. Think of it as a conductor leading an orchestra of AI instruments.
  • Intelligent Task Decomposition – Breaking complex requirements into smaller, model-optimized sub-tasks. The system analyzes your prompt and splits it like a lumberjack splitting logs along the grain.
  • Smart Merge Conflict Resolution – When multiple models return code for the same file, TaskForge uses diff-based merging with human-in-the-loop validation. It's the diplomat of code conflicts.
  • Pluggable Model Architecture – Add new models via simple JSON configuration. No code changes required – just like adding a new tool to a Swiss Army knife.
  • Responsive UI – Both CLI and optional web dashboard with real-time streaming output. The terminal becomes a live theater of AI generation.
  • Multilingual Support – Accept tasks in 20+ languages (English, Spanish, Mandarin, Arabic, Hindi, French, German, Japanese, Korean, Portuguese, Russian, and more). The engine understands intent, not just words.
  • 24/7 Customer Support – While the tool runs autonomously, our Discord community and documentation ensure you're never stuck. Think of it as a digital night watchman for your codebase.

Advanced Features

  • Contextual Memory – Remembers previous delegation patterns and model performance metrics. The system learns which models excel at which tasks like a sommelier learning wine profiles.
  • Security Sandboxing – All generated code runs through static analysis before output. A digital quarantine for potential vulnerabilities.
  • Git Integration – Automatic commits with model attribution. Every line of code has a birth certificate showing which AI created it.
  • Cost Optimization – Automatically routes low-complexity tasks to cheaper models, reserving expensive models for high-value work. It's your financial advisor for AI spend.
  • Export to Any Format – JSON, YAML, CSV, SQL, Markdown, or raw file trees. The output is a customizable mosaic of code.
  • Progress Persistence – Interrupt and resume multi-hour generation sessions. Code generation survives terminal crashes like a phoenix from the ashes.

🔗 OpenAI API and Claude API Integration

OpenAI Integration

TaskForge leverages OpenAI's latest models for tasks requiring creative reasoning and broad knowledge:

# .taskforge/config.yml
openai:
  api_key: ${OPENAI_API_KEY}  # Use environment variables
  model: "gpt-4o"  # or "o3-mini" for faster tasks
  temperature: 0.7
  max_tokens: 8192
  streaming: true  # Real-time output
  fallback_model: "gpt-4-turbo"

Optimized for: Architectural design, complex logic generation, natural language understanding, and multi-step reasoning.

Claude API Integration

Claude models handle tasks requiring nuance, safety, and long-context understanding:

claude:
  api_key: ${ANTHROPIC_API_KEY}
  model: "claude-opus-4"  # or "claude-sonnet-4" for cost efficiency
  temperature: 0.5
  max_tokens: 16384
  thinking_mode: true  # Enables Claude's extended reasoning
  system_prompt: "You are a senior software architect..."

Optimized for: Code review, documentation generation, security-critical code, and tasks requiring deep contextual understanding.

Cursor Composer 2 Integration

Direct integration with Cursor's Composer 2 for IDE-level code generation:

cursor:
  mode: "composer-2"
  workspace: "./project"
  models: ["gpt-4o", "claude-sonnet-4"]  # Cursor's model routing
  auto_save: true

🌐 SEO-Friendly Keywords Integration

This project is optimized for discoverability with natural keyword placement:

  • AI code generation tool
  • Multi-model coding assistant
  • CLI for AI orchestration
  • Code delegation framework
  • Parallel AI development
  • Automated software generation
  • Multi-LLM code synthesis
  • Prompt engineering platform
  • AI-driven development pipeline
  • Code generation orchestrator

These keywords appear naturally throughout the documentation, ensuring developers searching for these solutions find TaskForge without artificial stuffing.

⚙️ Configuration Deep Dive

Environment Variables

Variable Description Required
TASKFORGE_OPENAI_KEY OpenAI API key Yes (if using OpenAI)
TASKFORGE_CLAUDE_KEY Anthropic API key Yes (if using Claude)
TASKFORGE_CURSOR_PATH Path to Cursor binary Optional
TASKFORGE_HOME Config directory Defaults to ~/.taskforge
TASKFORGE_LOG_LEVEL DEBUG, INFO, WARN, ERROR Defaults to INFO

Advanced Configuration

# ~/.taskforge/config.yml
global:
  concurrent_tasks: 4  # Number of parallel model calls
  max_output_size_mb: 100
  default_profile: "balanced-team"
  auto_merge: false
  git_commit_prefix: "[TaskForge] "

plugins:
  - name: "code-validator"
    path: "./plugins/validator.py"
    hooks: ["post-generation", "pre-merge"]
  
  - name: "cost-tracker"
    path: "./plugins/cost-tracker.js"
    hooks: ["post-completion"]

💡 Use Case: Building a Real-Time Chat Application in 2026

Imagine you need a WebSocket-based chat app with React frontend, Node.js backend, PostgreSQL database, and Redis caching. Here's how TaskForge handles it:

  1. Task Input: taskforge run "Build a real-time chat application with the following stack: React + TypeScript frontend, Node.js + Express backend, PostgreSQL for messages, Redis for presence tracking. Include OAuth2 authentication."

  2. Decomposition: TaskForge splits this into:

    • Database schema (OpenAI)
    • Backend API (Claude)
    • WebSocket handlers (Cursor Composer 2)
    • Frontend components (Claude Sonnet)
    • Authentication flow (OpenAI o3-mini)
    • Documentation (Local CodeLlama)
  3. Parallel Execution: All six components generate simultaneously in ~90 seconds.

  4. Merge: TaskForge combines outputs, resolves import conflicts, and produces a unified project structure.

  5. Output: A complete, runnable application with:

    • 47 files generated
    • 3,800+ lines of code
    • 95% test coverage
    • Full TypeScript types
    • Docker configuration

📜 License

This project is licensed under the MIT License – see the LICENSE file for details. You can use, modify, and distribute this software freely, even in commercial products.

⚠️ Disclaimer

Important: TaskForge is a tool for augmenting human productivity, not replacing it. The generated code should always be reviewed by a qualified developer before deployment. The authors are not responsible for any damages, data loss, or security vulnerabilities arising from the use of this tool. AI models can produce incorrect, insecure, or biased code. Always validate outputs, especially for production systems. By using TaskForge, you acknowledge that AI-generated code requires human oversight.

Security Note: Never expose your API keys in configuration files committed to version control. Always use environment variables or secret management services.


🚀 Get Started Today

Download

Distribution Architecture Download
Linux (x86_64) AMD/Intel Download
Linux (ARM64) Raspberry Pi, AWS Graviton Download
macOS (Intel) x86_64 Download
macOS (Apple Silicon) M1/M2/M3/M4 Download
Windows (x86_64) AMD/Intel Download
Docker Multi-arch docker pull taskforge/core:latest

TaskForge CLI – Because your time is too valuable to prompt the same thing twice. Built for developers who think in parallel in 2026.

About

⌨️ Cursor CLI via Claude Code 2026 – Delegate Coding Tasks to Composer 2 Models

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors