agent-scaffold
v1.4.0
Published
AI agent collaboration infrastructure initializer. Sets up CLAUDE.md, AGENTS.md, memory, decision logs, and .claude/settings.json for any project. Framework-specific rules, multi-agent safety, monorepo support, auto-detection, health checks.
Maintainers
Readme
Agent Scaffold
AI Agent 协作基础设施一键初始化工具。为任何项目搭建跨会话记忆和决策日志。
One-command AI agent collaboration infrastructure. Sets up cross-session memory, decision logging, and optional signal system for any project.
What It Creates
your-project/
├── CLAUDE.md ← AI Agent instructions (Claude Code)
├── AGENTS.md ← Cross-agent instructions (Codex/Gemini/Copilot)
├── MEMORY.md ← Long-term curated memory (pre-filled with detected facts)
├── memory/ ← Daily session logs
├── DECISIONS.md ← Design decision records (ADR)
├── WIP.md ← Cross-session task handoff
└── .gitignore ← Updated with runtime exclusions
# With --signals flag:
├── signals/
│ ├── active/ ← Currently claimed tasks (YAML)
│ ├── observations/ ← Environment observations
│ └── archive/ ← Completed signalsInstall
Method 1: npx (zero install, any AI agent)
npx agent-scaffold initWorks with any AI agent — Claude, Codex, Gemini, Copilot, Cursor.
Method 2: Global npm install
npm install -g agent-scaffold
# Then in any project:
agent-scaffold initMethod 3: Claude Code skill (recommended for Claude users)
npx agent-scaffold install-skillThis copies the skill to ~/.claude/skills/agent-scaffold/. Then in any project, just tell Claude: "init agent scaffold" — Claude will analyze your project and generate customized files.
Method 4: Git clone (for contributors)
git clone https://github.com/liubo988/agent-scaffold.git ~/agent-scaffold
bash ~/agent-scaffold/install.shUsage
CLI
# Initialize current directory
npx agent-scaffold init
# Initialize a specific directory
npx agent-scaffold init ./my-project
# Include signal-driven task coordination system
npx agent-scaffold init --signals
# Minimal output: only CLAUDE.md + AGENTS.md + .gitignore
npx agent-scaffold init --slim
# Check scaffold health
npx agent-scaffold doctor
# Install as Claude Code skill
npx agent-scaffold install-skillAuto-Detection
The CLI auto-detects your project and generates customized files:
- Languages: Node.js/TypeScript, Python, Rust, Go, Java/Kotlin
- Frameworks: Next.js, React, Vue, Express, NestJS, Django, FastAPI, Flask, etc.
- Monorepos: pnpm workspaces, npm workspaces, Lerna, multi-package
- Tools: ESLint, Prettier, Biome, Ruff, Clippy, etc.
- Infrastructure: Docker, CI/CD pipelines
Claude Code Skill
After installing as a skill, just tell Claude in any project:
"init agent scaffold"
Claude will:
- Analyze your project (language, framework, build commands)
- Generate customized
CLAUDE.mdandAGENTS.md - Create memory directories and template files
- Update
.gitignore
With Codex / Gemini / Copilot
Run the CLI first, then ask the AI to:
"Read the AGENTS.md in this project, analyze the codebase, and customize it"
Options
| Flag | Description |
|------|-------------|
| --signals | Enable signal-driven task coordination (signals/ directory with YAML task files) |
| --slim | Minimal output: only CLAUDE.md + AGENTS.md + .gitignore update |
Health Check
npx agent-scaffold doctorValidates your scaffold setup and reports any missing or broken components.
Team Rollout
Share one command with your team:
npm install -g agent-scaffoldOr for Claude Code users:
npx agent-scaffold install-skillEveryone gets the same scaffold across all projects. No per-project setup needed.
Update
npm update -g agent-scaffoldCross-Agent Compatibility
| Agent | How to use |
|-------|-----------|
| Claude Code | Install as skill, say "init agent scaffold" |
| Codex | npx agent-scaffold init, then ask to customize AGENTS.md |
| Gemini | npx agent-scaffold init, then ask to customize AGENTS.md |
| Copilot | npx agent-scaffold init, then ask to customize AGENTS.md |
| Cursor | npx agent-scaffold init, then ask to customize AGENTS.md |
Design Principles
- Memory as infrastructure: File system is the persistence layer
- AGENTS.md standard: Compatible with 8+ AI agents via the Linux Foundation Agentic AI Foundation standard
- Monorepo aware: Scans workspace packages and generates per-workspace commands
- Idempotent: Never overwrites existing files; safe to re-run
- Zero dependencies: Pure Node.js, no external packages required
- Signal system opt-in: Task coordination via
--signalsfor advanced multi-agent workflows
License
MIT
