ccg-workflow

Why we included this project

For anyone whose daily work is Claude Code, the limit you hit is usually a single model's context and judgment. CCG gets around that by letting Claude act as lead orchestrator: it reads a plain-language request and picks a strategy from a small set (direct fixes up through full collaboration and deep research), then hands the parallel work to Codex and Gemini through a compiled Go bridge. One /ccg:go command takes the pipeline from reading project context and classifying the request to a hard stop for plan approval before the agent teams start writing code. Task state persists in .ccg directories, and hooks inject context on every turn, so nothing is lost when a conversation gets compacted. Quality gates and a cross-model review run before it reports back, which matters most on medium-to-complex features and bug hunts, where a second model checking the first one's plan catches more than a solo session does.