AI Agent Skills

Cross-IDE AI agent skills with interactive CLI, dual-source SKILL.md discovery, GitHub skill fetching, and installers that render the right layout for each IDE.

Quick Install

npx @votruongdanh/ai-agent-skills init

The CLI launches an interactive setup — it always asks you to choose an IDE from the full supported list, then picks project vs global scope:

 AI Agent Skills  v3.2.0

🔧 Choose your IDE:
  1) Cursor
  2) Kiro
  3) Claude Code
  4) Windsurf
  5) Antigravity
  6) Codex
  7) VS Code
  8) GitHub Copilot
  9) All supported IDEs

📂 Install scope:
  1) This project only    ← recommended
  2) Global (all projects)

[1] Installing skills...
✔ 16 skills installed for Cursor

Done! Next steps:
  1. Reopen Cursor
  2. Open agent chat and type / to list skills
  3. Try: /create, /debug, /explain, or /plan

Non-interactive mode

For CI/CD or scripting, pass --ide to skip prompts:

npx @votruongdanh/ai-agent-skills init --ide=cursor
npx @votruongdanh/ai-agent-skills init --ide=all
npx @votruongdanh/ai-agent-skills init --ide=kiro
npx @votruongdanh/ai-agent-skills init --ide=claude-code
npx @votruongdanh/ai-agent-skills init --ide=windsurf
npx @votruongdanh/ai-agent-skills init --ide=antigravity
npx @votruongdanh/ai-agent-skills init --ide=codex
npx @votruongdanh/ai-agent-skills init --ide=vscode
npx @votruongdanh/ai-agent-skills init --ide=copilot
npx @votruongdanh/ai-agent-skills init --no-interactive

The CLI still supports marker-based context detection for status/list flows when applicable, including parent-folder lookup.

For non-interactive mode (--no-interactive), --ide is required. You can pass --ide=all to install for every supported IDE in one run.

Add Skills from GitHub

Fetch skills directly from any GitHub repository:

# Browse and pick skills interactively
npx @votruongdanh/ai-agent-skills add owner/repo

# Install a specific skill
npx @votruongdanh/ai-agent-skills add anthropics/skills --skill=pdf-processing

# Specify branch and target IDE
npx @votruongdanh/ai-agent-skills add anthropics/skills --ide=cursor --branch=main

The add command:

• Fetches the GitHub repo tree via API
• Lists all SKILL.md files found
• Lets you pick one skill or install all
• Downloads companion scripts/, references/, assets/ directories
• Installs to your chosen IDE format and scope

Commands

| Command | Description | |---------|-------------| | npx @votruongdanh/ai-agent-skills init | Interactive setup (choose IDE + scope) | | npx @votruongdanh/ai-agent-skills init --ide=cursor | Non-interactive install for specific IDE | | npx @votruongdanh/ai-agent-skills global | Install globally for all projects | | npx @votruongdanh/ai-agent-skills add owner/repo | Add skills from a GitHub repository | | npx @votruongdanh/ai-agent-skills list | List all bundled skills | | npx @votruongdanh/ai-agent-skills list --json | JSON output for scripts | | npx @votruongdanh/ai-agent-skills status | Show install status & versions | | npx @votruongdanh/ai-agent-skills help | Show help |

Supported IDEs

| IDE | Installed layout | Notes | | --- | --- | --- | | Kiro | .kiro/skills/<skill>/SKILL.md | Native skill format | | Cursor | .cursor/skills/<skill>/SKILL.md | Native skill format | | Cursor legacy | .cursor/rules/<skill>.mdc | Generated compatibility layer | | Claude Code | .claude/skills/<skill>/SKILL.md | Native skill format (Agent Skills standard) | | Windsurf | .windsurf/skills/<skill>/SKILL.md | Native skill format; global: ~/.codeium/windsurf/skills/<skill>/SKILL.md | | Antigravity | .agent/workflows/<skill>.md | Generated workflow bridge | | Antigravity legacy | agent/workflows/<skill>.md | Compatibility alias for older setups | | Codex | .agents/skills/<skill>/SKILL.md | Detected from .agents/skills, AGENTS.md, and/or memories/ | | VS Code | .github/skills/<skill>/SKILL.md | Native SKILL.md format for Copilot Chat | | GitHub Copilot | .github/skills/<skill>/SKILL.md | Alias of VS Code target |

Native skill targets keep the full skill directory. Generated compatibility targets (Cursor .mdc rules, Antigravity workflows) only render SKILL.md content — companion scripts/, references/, and assets/ remain native-only.

For AI Agents & Developers: How to Edit Skills

All skills are now stored in the skills/ directory at the root of the project. This is the canonical source of truth.

When you need to edit, update, or add a new skill:

1. Edit the Markdown files directly in the skills/ directory.
2. Run npm run sync:all to automatically synchronize your changes to all supported IDE formats (Cursor, Kiro, Windsurf, Antigravity, VS Code, Codex, etc.).

Note: DO NOT manually edit the generated files in .agent/workflows/, .cursor/rules/, or similar IDE-specific folders, as your changes will be overwritten upon sync.

Available Skills (16)

| Skill | Description | |-------|-------------| | /agents | Agent routing — auto-selects specialist agent per request | | /brainstorm | Ideation, option generation, feature exploration | | /clean | Clean junk files, caches, AI artifacts | | /create | Build new features, files, components, endpoints | | /debug | Root-cause analysis for bugs and errors | | /deploy | Deployment, CI/CD, release preparation | | /enhance | Refactor, optimize, tighten security, improve UX | | /explain | Explain code, walk through logic and architecture | | /integrate | Safely merge sample/snippet code into the existing system with minimal changes | | /motion-ui | Motion UI with Framer Motion micro-interactions and smooth animations | | /orchestrate | Coordinate multi-step plans across domains | | /plan | Implementation plans with milestones and breakdown | | /preview | Preview output, UX flows, mockups before building | | /status | Project health, dependency, and progress reports | | /test | Generate and run tests, coverage analysis | | /ui-ux-pro-max | UI/UX design, accessibility, responsive layouts |

All skills support both English and Vietnamese trigger keywords.

Project Memory & .ai-memory.md

The skills share a project-wide memory file (.ai-memory.md) that captures tech stack, architecture, decisions, known issues, and recent work.
Memory writes are optimized to stay short and de-duplicated:

• Central Memory Compaction Rules limit bullets per section and enforce concise, file-focused notes.
• A lightweight dedupe protocol prevents duplicate bullets by normalizing/merging similar entries instead of appending.
• Every skill’s Memory Protocol uses these rules, so memory stays useful over long-lived projects without growing out of control.

For Codex-style workspaces, memory can also be mirrored into:

• memories/ (persistent user notes)
• memories/session/ (conversation-scoped notes)
• memories/repo/ (repo-scoped facts)

.ai-memory.md remains the cross-IDE canonical memory file.

Agent Routing System

The /agents skill includes 11 specialist agent personas that are automatically selected based on your request:

• Architect — system design, scalability
• Backend — APIs, databases, server logic
• Frontend — UI, components, styling
• DevOps — CI/CD, Docker, infrastructure
• QA — testing, quality assurance
• Security — vulnerability analysis, hardening
• Data — databases, queries, migrations
• Performance — optimization, profiling
• Documentation — docs, README, guides
• Reviewer — code review, best practices
• Fullstack — cross-domain coordination

The router handles keyword matching, Vietnamese triggers, compound keywords, and vague-request fallback.

How the Package is Structured

skills/                # Canonical skill source (16 skills)
.agents/skills/        # Cross-client interop root
lib/skill-bundle.js    # Discovery, YAML parsing, catalog, render, install
bin/cli.js             # Interactive CLI with colors, prompts, GitHub fetcher
bin/install-skills.ps1 # Optional PowerShell wrapper
scripts/render-targets.js  # Sample output renderer
scripts/sync-all.js    # Script to sync 'skills/' to all IDE folders
verify.js              # 44-check test suite

• Canonical source prefers skills/, then falls back to .skills
• lib/skill-bundle.js contains all discovery, parsing, rendering, and install logic
• bin/cli.js provides interactive install, GitHub add, list, status, and help commands
• Each skill exposes: slug, name, description, sourceRoot, hasScripts, hasReferences, hasAssets, and per-target compatibility metadata

Development

# Run test suite (44 checks)
node verify.js

# Render sample targets
npm run build:targets

See docs/skills-benchmark.md for the benchmark summary behind this design.

Updating

See Update.md for the full release workflow:

node verify.js                # Test first
git add . && git commit -m "description"
npm run release:patch         # Bug fixes: x.x.0 → x.x.1
npm run release:minor         # New features: x.0.x → x.1.0
npm run release:major         # Breaking: 0.x.x → 1.x.x

License

MIT