by fengshao1227
多模型协作开发系统 - Claude 编排 + Codex 后端 + Gemini 前端,28 个命令覆盖开发全流程,一键安装零配置
# Add to your Claude Code skills
git clone https://github.com/fengshao1227/ccg-workflowLast scanned: 4/19/2026
{
"issues": [
{
"type": "clone-failed",
"message": "Could not clone repository",
"severity": "medium"
}
],
"status": "WARNING",
"scannedAt": "2026-04-19T06:01:14.617Z",
"semgrepRan": false,
"npmAuditRan": false,
"pipAuditRan": false
}简体中文 | English
302.AI is a pay-as-you-go enterprise AI resource hub that offers the latest and most comprehensive AI models and APIs on the market, along with a variety of ready-to-use online AI applications.
n1n.ai — Global LLM API Gateway. One API Key to access 500+ top AI models (GPT-5, Claude 4.5, Gemini 3 Pro, and more).
A multi-model collaboration development system where Claude Code orchestrates Codex + Gemini. Frontend tasks route to Gemini, backend tasks route to Codex, and Claude handles orchestration and code review.
/ccg:*.Claude Code (Orchestrator)
│
┌───┴───┐
↓ ↓
Codex Gemini
(Backend) (Frontend)
│ │
└───┬───┘
↓
Unified Patch
External models have no write access — they only return patches, which Claude reviews before applying.
🎬 See CCG in action → — Real multi-model collaboration demo on X
| Dependency | Required | Notes |
|------------|----------|-------|
| Node.js 20+ | Yes | ora@9.x requires Node >= 20. Node 18 causes SyntaxError |
| Claude Code CLI | Yes | Install guide |
| jq | Yes | Used for auto-authorization hook (install) |
| Codex CLI | No | Enables backend routing |
| Gemini CLI | No | Enables frontend routing |
npx ccg-workflow
On first run, CCG prompts you to select a language (English / Chinese). This preference is saved for all future sessions.
# macOS
brew install jq
# Linux (Debian/Ubuntu)
sudo apt install jq
# Linux (RHEL/CentOS)
sudo yum install jq
# Windows
choco install jq # or: scoop install jq
npx ccg-workflow menu # Select "Install Claude Code"
Supports: npm, homebrew, curl, powershell, cmd.
| Command | Description | Model |
|---------|-------------|-------|
| /ccg:workflow | Full 6-phase development workflow | Codex + Gemini |
| /ccg:plan | Multi-model collaborative planning (Phase 1-2) | Codex + Gemini |
| /ccg:execute | Multi-model collaborative execution (Phase 3-5) | Codex + Gemini + Claude |
| /ccg:codex-exec | Codex full execution (plan → code → review) | Codex + multi-model review |
| /ccg:feat | Smart feature development | Auto-routed |
| /ccg:frontend | Frontend tasks (fast mode) | Gemini |
| /ccg:backend | Backend tasks (fast mode) | Codex |
| Command | Description | Model |
|---------|-------------|-------|
| /ccg:analyze | Technical analysis | Codex + Gemini |
| /ccg:debug | Problem diagnosis + fix | Codex + Gemini |
| /ccg:optimize | Performance optimization | Codex + Gemini |
| /ccg:test | Test generation | Auto-routed |
| /ccg:review | Code review (auto git diff) | Codex + Gemini |
| /ccg:enhance | Prompt enhancement | Built-in |
| Command | Description |
|---------|-------------|
| /ccg:spec-init | Initialize OPSX environment |
| /ccg:spec-research | Requirements → Constraints |
| /ccg:spec-plan | Constraints → Zero-decision plan |
| /ccg:spec-impl | Execute plan + archive |
| /ccg:spec-review | Dual-model cross-review |
| Command | Description |
|---------|-------------|
| /ccg:team-research | Requirements → constraints (parallel exploration) |
| /ccg:team-plan | Constraints → parallel implementation plan |
| /ccg:team-exec | Spawn Builder teammates for parallel coding |
| /ccg:team-review | Dual-model cross-review |
Prerequisite: Enable Agent Teams in
settings.json:CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
| Command | Description |
|---------|-------------|
| /ccg:commit | Smart commit (conventional commit format) |
| /ccg:rollback | Interactive rollback |
| /ccg:clean-branches | Clean merged branches |
| /ccg:worktree | Worktree management |
| Command | Description |
|---------|-------------|
| /ccg:init | Initialize project CLAUDE.md |
| /ccg:context | Project context management (.context/ init, log, compress, history) |
# 1. Generate implementation plan
/ccg:plan implement user authentication
# 2. Review the plan (editable)
# Plan saved to .claude/plan/user-auth.md
# 3a. Execute (Claude refactors) — fine-grained control
/ccg:execute .claude/plan/user-auth.md
# 3b. Execute (Codex does everything) — efficient, low Claude token usage
/ccg:codex-exec .claude/plan/user-auth.md
Integrates OPSX architecture to turn requirements into constraints, eliminating AI improvisation:
/ccg:spec-init # Initialize OPSX environment
/ccg:spec-research implement user auth # Research → constraints
/ccg:spec-plan # Parallel analysis → zero-decision plan
/ccg:spec-impl # Execute the plan
/ccg:spec-review # Independent review (anytime)
Tip:
/ccg:spec-*commands internally call/opsx:*. You can/clearbetween phases — state is persisted in theopenspec/directory.
Leverage Claude Code Agent Teams to spawn multiple Builder teammates for parallel coding:
/ccg:team-research implement kanban API # 1. Requirements → constraints
# /clear
/ccg:team-plan kanban-api # 2. Plan → parallel tasks
# /clear
/ccg:team-exec # 3. Builders code in parallel
# /clear
/ccg:team-review # 4. Dual-model cross-review
vs Traditional Workflow: Team series uses
/clearbetween steps to isolate context, passing state through files. Ideal for tasks decomposable into 3+ independent modules.
~/.claude/
├── commands/ccg/ # 29+ slash commands
├── agents/ccg/ # Sub-agents
├── skills/ccg/ # Quality gates + multi-agent orchestration
├── bin/codeagent-wrapper
└── .ccg/
├── config.toml # CCG configuration
└── prompts/
├── codex/ # 6 Codex expert prompts
└── gemini/ # 7 Gemini expert prompts
Configure in ~/.claude/settings.json under "env":
| Variable | Description | Default | When to change |
|----------|-------------|---------|----------------|
| CODEAGENT_POST_MESSAGE_DELAY | Wait after Codex completion (sec) | 5 | Set to 1 if Codex process hangs |
| CODEX_TIMEOUT | Wrapper execution timeout (sec) | 7200 | Increase for very long tasks |
| BASH_DEFAULT_TIMEOUT_MS | Claude Code Bash timeout (ms) | 120000 | Increase if commands time out |
| BASH_MAX_TIMEOUT_MS | Claude Code Bash max timeout (ms) | 600000 | Increase for long builds |
{
"env": {
"CODEAGENT_POST_MESSAGE_DELAY": "1",
"CODEX_TIMEOUT": "7200",
"BASH_DEFAULT_TIMEOUT_MS": "600000",
"BASH_MAX_TIMEOUT_MS": "3600000"
}
}
npx ccg-workflow menu # Select "Configure MCP"
Code retrieval (choose one):
search_context. Official | Third-party proxyOptional tools:
CCG automatically installs a Hook to auto-authorize codeagent-wrapper commands (requires jq).
Add to ~/.claude/settings.json:
{
"hooks": {
"PreToolUse": [
{
"matcher": "Bash",
No comments yet. Be the first to share your thoughts!