Ccg Workflow
多模型协作工作流引擎 — /ccg:go 一个命令，AI 自动分析意图、选择策略、编排 Codex + Gemini + Claude 协作执行
Last verified:
What is Ccg Workflow?
CCG Workflow is a multi-model AI collaboration development system implemented as a Node.js CLI that orchestrates Claude, Codex, and Gemini to work in parallel rather than in isolation. The tool routes frontend development tasks to Gemini CLI (which excels at UI/styling), backend tasks to Codex CLI (stronger at logic/algorithms), and uses Claude Code as the orchestrator and code reviewer. This creates a true
Ccg Workflow pricing
Pricing model: Freemium
CCG Workflow is completely free and open-source under MIT license. It is installed via npx ccg-workflow with no subscription fees. The tool itself is free, but users must pay for their own API usage with Claude, Codex, and Gemini services separately. The 302.AI sponsor mentioned on the repository is a pay-as-you-go enterprise AI resource hub offering various AI models and APIs, but it is optional and not required to use CCG Workflow.
Ccg Workflow pros
- Orchestrates Claude + Codex + Gemini working in parallel
- Intelligent model routing: frontend to Gemini, backend to Codex
- 29 slash commands covering entire development workflow
- Hook-based state tracking prevents context loss after compaction
- Task persistence with .ccg/tasks/ directory for medium+ complexity
- Agent Teams enables parallel Builder execution for large tasks
- Security by design: external models only generate patches
- Claude reviews and approves all changes before application
- One-command zero-config setup with npx ccg-workflow
- Automatic language selection (Chinese/English) with saved preference
- OPSX specification-driven development converts requirements to constraints
- Quality gates: verify-security, verify-quality, verify-change
- Domain knowledge hooks auto-inject relevant files for security, caching, RAG
- Codex-Led Mode available with Codex CLI as lead orchestrator
- 10 strategies automatically selected based on task type and complexity
- Git tools: smart conventional commits, interactive rollback, branch cleanup
- MIT license, completely open-source
- 5.4k+ GitHub stars with active community on Linux.do
Ccg Workflow cons
- Requires Node.js 20+ and Claude Code CLI installed
- Codex CLI and Gemini CLI required for multi-model features
- v3.0 replaced 29 commands with fewer commands, may lose granularity
- Agent Teams requires experimental flag CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
- Hook engine modifies ~/.claude/settings.json automatically
- Learning curve for understanding different strategies and commands
- External API costs for Claude, Codex, and Gemini not included
- Configuration complexity with config.toml, MCP tools, and environment variables
Frequently asked questions about Ccg Workflow
What is CCG Workflow?
CCG Workflow is a multi-model AI collaboration development system that orchestrates Claude, Codex, and Gemini to work together in parallel. It routes frontend tasks to Gemini, backend tasks to Codex, and uses Claude for orchestration and code review, enabling true team collaboration instead of using one AI in isolation.
How do I install CCG Workflow?
Simply run 'npx ccg-workflow' in your terminal. This requires Node.js version 20 or higher. The installer walks through 4 steps: API configuration, model routing, MCP tools, and performance mode. New users get a streamlined 2-step flow with sensible defaults.
What slash commands are available?
v3.0 has 13 core commands including /ccg:go (smart entry), /ccg:commit, /ccg:rollback, /ccg:clean-branches, /ccg:worktree, /ccg:init, /ccg:context, /ccg:spec-init, /ccg:spec-research, /ccg:spec-plan, /ccg:spec-impl, /ccg:spec-review, and legacy mode adds 18 more commands including /ccg:workflow, /ccg:frontend, /ccg:backend, /ccg:plan, /ccg:execute, /ccg:debug, /ccg:optimize, /ccg:test, /ccg:review.
How does the intelligent model routing work?
The engine reads project context including git status, tech stack, and file structure, then classifies the task as feature/bug, complexity level, frontend/backend, and risk level. Based on this classification, it automatically routes frontend tasks (UI/styling) to Gemini which excels at those, and backend tasks (logic/algorithms) to Codex which is stronger there, while Claude handles overall orchestration.
What strategies does the engine select?
The engine picks from 10 strategies: direct-fix (simple bug), quick-implement (small feature), guided-develop (medium feature), full-collaborate (complex feature with Agent Teams), debug-investigate (complex bug), refactor-safely (code restructuring), deep-research (technical research), optimize-measure (performance), review-audit (code review), and git-action (commit/rollback). Simple tasks run fast with no overhead, complex tasks get the full engine.
How does the hook engine prevent context loss?
CCG installs 4 hooks into ~/.claude/settings.json: workflow-state.js injects task state every turn, session-start.js injects full project context on session start/compact, subagent-context.js injects spec + task context into sub-agent calls, and skill-router.js auto-injects domain knowledge when keywords are detected. This ensures no state is lost even after context compaction.
What is Agent Teams?
Agent Teams is a feature for large tasks that spawns parallel Builder teammates via TeamCreate. Each Builder gets isolated file ownership and works in parallel. This requires setting CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 environment variable. It's particularly useful for full-collaborate strategy on complex, multi-module features.
How does the security design protect my code?
External models (Codex and Gemini) are restricted to patch file generation only. They cannot directly modify your code. All patches must be reviewed and approved by Claude (the orchestrator) before being applied. This prevents AI from making unwanted or dangerous changes to your codebase.
What is OPSX specification-driven development?
OPSX is integrated into CCG to convert vague requirements into verifiable constraints. Commands like /ccg:spec-init initialize the OPSX environment, /ccg:spec-research converts requirements to constraint sets, /ccg:spec-plan creates zero-decision plans from constraints, and /ccg:spec-impl executes the plan. This makes AI programming into a standardized SOP and prevents AI from freely interpreting requirements.
Is CCG Workflow free?
Yes, CCG Workflow is completely free and open-source under the MIT license. There are no subscription fees for the tool itself. However, users must pay for their own API usage with Claude, Codex, and Gemini services separately. The tool is installed via 'npx ccg-workflow' with zero cost.