ai-skills

A CLI tool to install production-grade coding skills for AI terminal agents. One command gives your AI agent expert-level capabilities in full-stack engineering, persistent project memory, and self-expanding skill creation.

Works with Claude Code, GPT CLI, Gemini CLI, Qwen CLI, Codex CLI, and any other AI terminal agent that reads markdown skill files.

Quick Start

# Install the CLI
npm install -g @sirbenj/ai-skills

# Install all skills globally (default: Claude Code)
ai-skills install

# Install for a different agent
ai-skills install --agent=gpt
ai-skills install --agent=gemini

You can also install directly from GitHub:

npm install -g kupzsensei/ai-skills

Bundled Skills

| Skill | Description | |-------|-------------| | fullstack-engineer-skill | A full engineering team in one agent — architect, frontend, backend, DevOps, QA, DBA, security, and tech lead. Handles building, debugging, refactoring, deploying, and maintaining any software project. Includes reference guides for API design, databases, DevOps, security, testing, and tech stack selection. | | project-memory-skill | Persistent memory across sessions and across agents. Start in Claude, continue in Gemini, pick up in Qwen — zero context loss. Memory is stored in a shared .ai/memory/ directory that all agents read and write. | | skill-factory | A meta-skill that creates new skills, detects when existing skills are outdated, and expands skills with new capabilities. Keeps your AI agent's knowledge current and production-ready. |

Usage

ai-skills <command> [options]

Commands

install

Install skills to your agent's skill directory.

# Install all skills globally
ai-skills install

# Install a specific skill
ai-skills install fullstack-engineer-skill

# Install to project directory instead of global
ai-skills install --project

# Skip overwrite confirmation
ai-skills install --force

# Install for a specific agent
ai-skills install --agent=gemini

uninstall

Remove an installed skill.

ai-skills uninstall project-memory-skill
ai-skills uninstall skill-factory --project
ai-skills uninstall fullstack-engineer-skill --agent=gpt

list

Show available skills and their install status.

ai-skills list
ai-skills list --project
ai-skills list --agent=codex

Options

| Flag | Description | |------|-------------| | --project | Target the project-level directory (./<agent>/skills/) instead of global (~/.<agent>/skills/) | | --force | Skip confirmation prompts when overwriting or removing | | --agent=<name> | Target a specific agent. Default: claude |

Supported Agents

| Agent | Global Path | Project Path | |-------|-------------|--------------| | claude | ~/.claude/skills/ | .claude/skills/ | | gpt | ~/.gpt/skills/ | .gpt/skills/ | | gemini | ~/.gemini/skills/ | .gemini/skills/ | | qwen | ~/.qwen/skills/ | .qwen/skills/ | | codex | ~/.codex/skills/ | .codex/skills/ |

Install Locations

Global (default) — skills are available across all your projects:

~/.claude/skills/fullstack-engineer-skill/SKILL.md

Project (--project) — skills scoped to the current project:

./your-project/.claude/skills/fullstack-engineer-skill/SKILL.md

What Happens During Install

The installer does three things:

1. Copies skill files to the agent's skill directory
2. Injects auto-activation into the agent's instruction file (e.g., CLAUDE.md) — skills that include an AUTORUN.md get their activation instructions added automatically, so they trigger without manual /slash-command invocation
3. Offers recommended settings — permission rules that let skills work without constant approval prompts

Auto-Activation

Skills with an AUTORUN.md file automatically inject instructions into the agent's instruction file:

| Agent | Global Instruction File | Project Instruction File | Supported | |-------|------------------------|--------------------------|-----------| | claude | ~/.claude/CLAUDE.md | ./CLAUDE.md | Yes | | gemini | ~/.gemini/GEMINI.md | ./GEMINI.md | Yes | | qwen | ~/.qwen/QWEN.md | ./QWEN.md | Yes | | codex | ~/.codex/AGENTS.md | ./AGENTS.md | Yes | | gpt | — | — | No (no instruction file support) |

These sections are wrapped in HTML comment markers (<!-- ai-skills:skill-name:start/end -->) and are cleanly removed on uninstall.

Shared Memory Across Agents

The project-memory-skill stores all memory in .ai/memory/ — an agent-agnostic directory shared by all agents. This means you can:

• Start a project in Claude Code, build the foundation
• Run out of context, switch to Gemini CLI — it loads the same memory and picks up where Claude left off
• Switch to Qwen CLI later — same memory, zero context loss

your-project/
└── .ai/
    └── memory/
        ├── index.md          # Project map (shared)
        ├── sessions.md       # Session log from ALL agents
        ├── todo.md           # Task tracker (shared)
        ├── decisions.md      # Technical decisions (shared)
        ├── bugs.md           # Bug tracker (shared)
        └── file-index.json   # File index (shared)

Recommended Settings (Claude Code)

After installing, the CLI will prompt you to apply recommended permission rules to settings.json. These let skills read/write memory files and scan your project without asking for approval on every action:

{
  "permissions": {
    "allow": [
      "Read(.ai/memory/**)",
      "Edit(.ai/memory/**)",
      "Write(.ai/memory/**)",
      "Bash(cat .ai/memory/*)",
      "Bash(ls .ai/memory/*)",
      "Bash(mkdir -p .ai/memory)",
      "Bash(find . -type f *)",
      "Bash(wc -l *)",
      "Bash(head *)",
      "Bash(grep -rn *)",
      "Skill(project-memory-skill)",
      "Skill(fullstack-engineer-skill)",
      "Skill(skill-factory)"
    ]
  }
}

Use --force to apply settings automatically, or decline the prompt to configure manually.

Requirements

• Node.js >= 16
• Zero runtime dependencies

License

MIT