Claude Code gets a multi-model strategy engine
CCG turns Claude Code into a multi-model command center that analyzes your intent, picks a strategy, and dispatches work across Claude, Codex, and Gemini.

What it does
CCG is a workflow engine that plugs into Claude Code via JavaScript hooks and slash commands. You type /ccg:go followed by a plain-language goal, and the engine classifies the task, selects one of ten strategies ranging from direct-fix to full-collaborate, then executes it using the appropriate models. For complex jobs, it persists state in a local task directory, enforces phase-gate checkpoints, and can spawn parallel Agent Teams with isolated file ownership.
The interesting bit
The engine fights context loss by injecting a <ccg-state> breadcrumb into every turn and re-hydrating the full project context whenever a session starts or gets compacted. It also auto-injects domain knowledge—security rules, caching patterns, RAG notes—when your prompt mentions relevant keywords, so sub-agents follow your conventions without being reminded.
Key highlights
- Strategy router — Ten predefined strategies match task complexity to model count; simple bugs stay local, complex features spin up dual-model parallel analysis and Agent Teams.
- Hook-based state tracking — Four zero-dependency JavaScript hooks install into Claude Code’s settings to silently inject task state, session context, and spec files at the right moments.
- Persistent task artifacts — Medium-and-up tasks generate a
.ccg/tasks/folder withtask.json,plan.md, andreview.md, making the workflow inspectable and resumable. - Project spec injection — Coding standards stored in
.ccg/spec/are automatically fed into every codeagent-wrapper call and team spawn, keeping sub-agents aligned with your codebase conventions. - Optional Codex-led mode — Flip the orchestrator so Codex writes code directly while delegating analysis and review to Gemini and Claude.
Caveats
- Requires Claude Code CLI; it is not a standalone tool but a heavy configuration layer on top of it.
- Multi-model features depend on optional external CLIs (Codex and Gemini), and Agent Teams need an experimental Claude Code environment flag enabled.
- Installation modifies
~/.claude/settings.json, which may not appeal if you prefer to keep your Claude Code setup pristine.
Verdict
Reach for CCG if you already live in Claude Code and want systematic multi-model collaboration for large or risky tasks. If you use a different editor or just need quick one-off answers, this is overkill.
Frequently asked
- What is fengshao1227/ccg-workflow?
- CCG turns Claude Code into a multi-model command center that analyzes your intent, picks a strategy, and dispatches work across Claude, Codex, and Gemini.
- Is ccg-workflow open source?
- Yes — fengshao1227/ccg-workflow is open source, released under the MIT license.
- What language is ccg-workflow written in?
- fengshao1227/ccg-workflow is primarily written in Go.
- How popular is ccg-workflow?
- fengshao1227/ccg-workflow has 5.8k stars on GitHub and is currently holding steady.
- Where can I find ccg-workflow?
- fengshao1227/ccg-workflow is on GitHub at https://github.com/fengshao1227/ccg-workflow.