wtcraft
Git-native, bounded scaffolding for multi-agent coding
为多智能体编码打造的、Git 原生的有界脚手架
Spawn parallel AI agents inside isolated git worktrees with strict, verifiable
task boundaries — instead of reinventing another runtime.
在隔离的 git worktree 中并行运行 AI 智能体，用可验证的任务边界约束它们 —— 而不是再造一个 runtime。
Install / 安装
# Node / npm
npm install -g wtcraft
# Python / pipx (recommended)
pipx install wtcraft
# Homebrew (macOS)
brew tap zywkloo/wtcraft https://github.com/zywkloo/wtcraft && brew install wtcraft
Then run wtcraft init in your repo root to scaffold the harness.
随后在仓库根目录运行 wtcraft init 即可生成脚手架。
Articles / 系列文章 · TokenChef
- 👨🍳 TokenChef
TokenChef (Part 1): Vibe Coding with Git Worktrees — A Playbook Most Devs Are Missing
Most people vibe-code with one model in one branch. Here is how to use multiple models, git worktrees, and a phased workflow to ship refactors without breaking everything.
- 👨🍳 TokenChef
TokenChef (Part 2): Chief Token Orchestrator — Manage Claude, Codex, and Gemini as a Structured Software Team
Spawning parallel agents is easy. Spawning them without hitting token limits, context pollution, or review bottlenecks is hard. Discover why you need a structured, layered team architecture.
- 👨🍳 TokenChef
TokenChef (Part 3): wtcraft — A Lightweight, Git-Native Scaffolding for Bounded Multi-Agent Coding
How do you actually enforce boundaries on autonomous agents? Explore Harness Engineering, the competitive landscape of task contracts, and how the wtcraft CLI implements bounded sandboxes locally.
- 👨🍳 TokenChef
TokenChef (Part 4): Stop Reinventing the Agent — wtcraft Boundaries, Ledgers & Sign-off First
从 Claude Code 源码泄露与 CCB 逆向架构出发，用同一套六层架构对比 Codex 和 Claude Code，得出结论：wtcraft 不该再造 runtime，而该把任务边界、运行账本和验收流程做扎实。
Guides / 营销与优化文档
- 🧭 Overview
SEO & GEO Field Notes
Why / how / when + the GEO acceptance loop — the strategic overview behind the three hands-on guides.
- 📄 Docs
Marketing an Open-Source Package (NPM / PyPI / GitHub) for SEO & GEO
How to turn the domain authority of GitHub, NPM, and PyPI into backlinks and AI citations for your own site — registry metadata, a RAG-friendly README, and the backlink loop.
- 📄 Docs
Marketing a Personal GitHub Pages + Astro Blog (SEO & GEO)
The full playbook for a bilingual Astro blog on GitHub Pages: the Astro-specific technical SEO that's free, the title/metadata gotchas that aren't, and how to get cited by AI engines fast via canonical cross-posting.
- 📄 Docs
Rebuilding Two Legacy Commercial Sites for SEO & GEO (a Forward Plan)
Architecture and migration decision matrices for two common commercial-site archetypes — a client-rendered product landing page and a legacy-WordPress B2B site — covering framework choice, zero-risk migration, hosting TCO, and SEO/GEO.
FAQ
What is wtcraft?
wtcraft is a lightweight, git-native CLI that runs parallel AI coding agents inside isolated git worktrees with strict, verifiable task boundaries — so you can orchestrate a multi-agent team without reinventing a runtime.
How is wtcraft different from plain git worktrees or Claude Code's --worktree?
Raw worktrees give you isolation but leave token budget, file scope, and task handoff up to you. wtcraft adds the missing contract layer: a per-task .worktree-task.md with explicit Scope and Off-limits blocks, plus a run ledger and a sign-off step.
Does wtcraft work with Claude Code, Codex, and Gemini?
Yes. wtcraft is agent-agnostic scaffolding; run `wtcraft init --patch-agent-files` to opt into integration with Claude Code, OpenAI Codex, or the Gemini CLI.
How do I install wtcraft?
Install globally via npm (npm install -g wtcraft), pipx (pipx install wtcraft), or Homebrew, then run `wtcraft init` in your repository root.