fengshao1227/ccg-workflow

What it solves

CCG is a workflow engine designed to enhance Claude Code by transforming it into a multi-model orchestrator. It solves the problem of relying on a single LLM for complex software engineering tasks by allowing Claude to dispatch specialized work to other models like Codex (OpenAI), Gemini (Google), and Grok (xAI) for parallel analysis, review, and implementation.

How it works

CCG uses a Go-based binary bridge (codeagent-wrapper) to connect Claude Code to external models. It employs a "Hook Engine" that injects state and project context into every turn of the conversation, ensuring context is preserved even after session compaction.

When a user provides a command (e.g., /ccg:go), the engine analyzes the task complexity and selects one of 10 built-in strategies (such as full-collaborate or debug-investigate). For complex tasks, it creates a persistent task directory containing requirements, plans, and review results, and can spawn parallel agent teams for implementation.

Who it’s for

Developers using the Claude Code CLI who want to automate complex coding workflows, implement multi-model cross-reviews, and maintain persistent state across long-running software development tasks.

Highlights

• Multi-Model Orchestration: Dispatches tasks to Codex, Gemini, and Grok while Claude remains the lead orchestrator.
• Hook Engine: Prevents context loss via four specialized JavaScript hooks that inject state, project context, and domain knowledge.
• Strategy-Based Execution: Automatically selects from 10 strategies based on task complexity, ranging from simple direct fixes to full collaboration.
• Persistent Task System: Maintains a .ccg/tasks directory for tracking status, requirements, and plans for medium-to-high complexity work.
• Domain Knowledge Injection: Auto-injects relevant knowledge from over 100 domain files (e.g., Security, DevOps, AI/MLOps) based on keywords.
• Quality Gates: Built-in verification commands for security, quality, and module structure checks.