Install standardized AI agent workflow skills and shared rules-file content into any repository with a single command.
Supports 40 AI coding agents including Claude Code, Cursor, Windsurf, Copilot, Cline, Roo Code, and more.
Languages: English | Tiếng Việt
Running the installer:
- Lets you select which agents to configure (auto-detects already-installed agents)
- Lets you choose whether the skills are installed as symlinks or copied files
- Installs workflow skills into each agent's skills directory via the Vercel skills CLI
- Injects shared agent instructions (
AGENTS.md) into each agent's rules file using idempotent markers
npx @buiducnhat/agent-skills@latestWalks you through agent selection and install mode (symlink or copy).
npx @buiducnhat/agent-skills@latest --non-interactiveSkips all prompts and installs skills for all agents.
curl -fsSL https://raw.githubusercontent.com/buiducnhat/agent-skills/main/install.sh | bashChecks for Node.js 18+ and runs the installer automatically.
npx @buiducnhat/agent-skills@latest --globalInstalls skills to your home directory (~/<agent>/skills/) so they are available across all projects.
npx @buiducnhat/agent-skills@latest -a claude-code -a cursorInstalls skills only for the listed agents. Repeat -a or pass multiple agent IDs after one flag.
┌ Agent Skills Installer
│
◇ Select agents to install skills for:
│ ◼ Claude Code ◼ Cursor ◻ Windsurf ...
│
◇ How should the skills be installed?
│ ● Symlink (recommended) ○ Copy
│
◇ Installing skills via skills CLI...
│
◇ Installation complete!
│
│ What was set up:
│ CLAUDE.md - updated
│
│ Agent configurations updated for:
│ - claude-code
│ - cursor
│
└ Done! Your AI agent skills are ready.
| Flag | Description |
|---|---|
-a, --agent |
Target one or more specific agents |
--non-interactive |
Skip prompts; install all skills to all agents |
--copy |
Copy skill files instead of symlinking |
-g, --global |
Install to ~/ instead of the current directory |
-h, --help |
Show help |
-v, --version |
Show version |
This repository currently defines nine first-party workflow skills and also pins additional upstream skills in skills-lock.json:
| Skill | Description |
|---|---|
as-ask |
Ask clarifying questions to gather requirements and context |
as-fix |
Diagnose and resolve bugs with root-cause analysis and verification |
as-review |
Review uncommitted changes with codebase context and severity levels |
brainstorm |
Explore ideas and break down complex problems before planning |
docs |
Create or refresh project documentation based on the current repo |
execute-plan |
Execute written plans systematically with checkpoints |
git-commit |
Create conventional commit messages from staged or unstaged changes |
quick-implement |
Rapid implementation for small, well-scoped changes |
write-plan |
Create detailed implementation plans with phases and tasks |
/docs
brainstorm → write-plan → execute-plan
Use when requirements are unclear or multiple approaches are possible.
# Step 1 — explore and define the design
/brainstorm add dark mode support
# Agent clarifies requirements, proposes approaches, writes:
# docs/brainstorms/260306-1430-dark-mode/SUMMARY.md
# Then prompts: "Proceed to write-plan?"
# Step 2 — create a phased implementation plan
/write-plan
# Agent reads the brainstorm, writes:
# docs/plans/260306-1445-dark-mode/SUMMARY.md
# docs/plans/260306-1445-dark-mode/phase-01-tokens.md
# docs/plans/260306-1445-dark-mode/phase-02-components.md
# Ends with: "Use /clear then /execute-plan docs/plans/... to execute"
# Step 3 — execute the approved plan in a fresh context
/clear
/execute-plan docs/plans/260306-1445-dark-mode/SUMMARY.md
write-plan → execute-plan
Use when the task is clear but too large or risky to implement without a plan.
# Step 1 — plan the work
/write-plan migrate auth to JWT
# Agent writes:
# docs/plans/260306-1020-jwt-auth/SUMMARY.md
# docs/plans/260306-1020-jwt-auth/phase-01-schema.md
# docs/plans/260306-1020-jwt-auth/phase-02-middleware.md
# Ends with: "Use /clear then /execute-plan docs/plans/... to execute"
# Step 2 — execute in a fresh context
/clear
/execute-plan docs/plans/260306-1020-jwt-auth/SUMMARY.md
quick-implement
Use for small, well-scoped changes where a formal plan would be overhead.
# Implement directly — no plan needed
/quick-implement add a tooltip to the submit button
as-fix
Use when you have a concrete error, failing test, or unexpected behavior to diagnose.
/as-fix TypeError: Cannot read properties of undefined at checkout.ts:42
# For simple bugs: agent diagnoses, applies fix, verifies
# For complex bugs: agent stops and routes to write-plan
View all 40 supported agents
AdaL, Amp, Antigravity, Augment, Claude Code, Cline, CodeBuddy, Codex, Command Code, Continue, Cortex Code, Crush, Cursor, Droid, Gemini CLI, GitHub Copilot, Goose, iFlow CLI, Junie, Kilo Code, Kimi Code CLI, Kiro CLI, Kode, MCPJam, Mistral Vibe, Mux, Neovate, OpenClaw, OpenCode, OpenHands, Pi, Pochi, Qoder, Qwen Code, Replit, Roo Code, Trae, Trae CN, Windsurf, Zencoder
- Node.js 18+
gitavailable inPATH- Network access (to clone templates from GitHub)
Re-running is safe. Rules injection is idempotent — the existing content between markers is replaced without duplication. New agents can be added to an existing setup at any time.
Tag pushes matching v* trigger the GitHub Actions workflow .github/workflows/release.yml to publish @buiducnhat/agent-skills to npm automatically.