Alexander's Skills

Personal Agent Skills library for Codex, Claude Code, Kimi Code CLI, and other AI agents.

⚠️ Development Note: This project uses strict git workflow. See CONTRIBUTING.md and /skill:git-workflow before making changes.

Skills

Skill	Description	Status	Platform
algorithmic-art	Creating algorithmic art using p5.js	✅ Ready	🟣 Universal
architecture-diagrams	Creating architecture diagrams with Mermaid	✅ Ready	🟣 Universal
beautiful-mermaid	Render Mermaid diagrams as SVG and PNG	✅ Ready	🟣 Universal
brainstorming	Explore user intent before implementation	✅ Ready	🟣 Universal
brand-guidelines	Anthropic's official brand colors and typography	✅ Ready	🟣 Universal
canvas-design	Create beautiful visual art in PNG/PDF	✅ Ready	🟣 Universal
Claudeception	Multi-layered agent collaboration framework	✅ Ready	🔵 Claude
cli-agent-caller	Multi-agent collaboration via CLI	✅ Ready	🟣 Universal
deep-requirement-analysis	Production-grade task planner with progressive disclosure	✅ Ready	🟢 Kimi
dispatching-parallel-agents	Dispatch 2+ independent tasks in parallel	✅ Ready	🟣 Universal
doc-coauthoring	Structured workflow for co-authoring documentation	✅ Ready	🟣 Universal
docx	Create and manipulate Word documents	✅ Ready	🟣 Universal
executing-plans	Execute written implementation plans	✅ Ready	🟣 Universal
find-skills	Discover and install agent skills	✅ Ready	🟣 Universal
finishing-a-development-branch	Complete development work with structured options	✅ Ready	🟣 Universal
frontend-design	Production-grade frontend interfaces	✅ Ready	🟣 Universal
git-workflow	Git workflow for skill development	✅ Ready	🟣 Universal
internal-comms	Resources for internal communications	✅ Ready	🟣 Universal
kimi-mem	Cross-session memory management system	✅ Ready	🟢 Kimi
mcp-builder	Guide for creating MCP servers	✅ Ready	🟣 Universal
mermaid-diagrams	Comprehensive guide for creating software diagrams with Mermaid	✅ Ready	🟣 Universal
openspec	OpenAPI specification tools and utilities	✅ Ready	🟣 Universal
pdf	Read, extract, create, and manipulate PDFs	✅ Ready	🟣 Universal
planning-with-files	Manus-style file-based planning for complex tasks	✅ Ready	🟣 Universal
pptx	Create and edit PowerPoint presentations	✅ Ready	🟣 Universal
receiving-code-review	Handle code review feedback properly	✅ Ready	🟣 Universal
requesting-code-review	Request comprehensive code reviews	✅ Ready	🟣 Universal
skill-creator	Guide for creating effective skills	✅ Ready	🟣 Universal
skill-validation	Comprehensive validation framework for skills	✅ Ready	🟣 Universal
slack-gif-creator	Create animated GIFs optimized for Slack	✅ Ready	🟣 Universal
subagent-driven-development	Execute plans with independent tasks	✅ Ready	🟣 Universal
systematic-debugging	Debug bugs and test failures	✅ Ready	🟣 Universal
systematic-literature-review	Conduct systematic literature reviews with AI	✅ Ready	🟣 Universal
test-driven-development	TDD workflow for features and bugfixes	✅ Ready	🟣 Universal
theme-factory	Styling artifacts with themes	✅ Ready	🟣 Universal
tooluniverse-literature-deep-research	Comprehensive literature research with evidence grading	✅ Ready	🟣 Universal
using-git-worktrees	Create isolated git worktrees	✅ Ready	🟣 Universal
using-superpowers	How to find and use skills	✅ Ready	🟣 Universal
verification-before-completion	Verify work before claiming completion	✅ Ready	🟣 Universal
vllm-deploy-docker	Deploy vLLM with Docker	✅ Ready	🟣 Universal
vllm-deploy-simple	Quick install and deploy vLLM locally	✅ Ready	🟣 Universal
webapp-testing	Test local web applications with Playwright	✅ Ready	🟣 Universal
web-artifacts-builder	Create elaborate multi-component web artifacts	✅ Ready	🟣 Universal
writing-plans	Create implementation plans before coding	✅ Ready	🟣 Universal
writing-skills	Create and verify skills before deployment	✅ Ready	🟣 Universal
xlsx	Create and manipulate Excel spreadsheets	✅ Ready	🟣 Universal

Platform Legend:

• 🟣 Universal - Documentation-based skills that can be installed into Codex and Claude Code
• 🟢 Kimi - Kimi-specific (requires MCP or Kimi-specific features)
• 🔵 Claude - Claude Code-specific (requires Claude plugin, hook, or command features)

💡 Note: Most documentation-based skills are marked as Universal since the core concepts work across agents. Platform-specific skills typically involve MCP integration or agent-specific tool usage.

Agent-Specific Installation Guide

For Codex Users

Install the default Codex profile:

./install.sh --agent codex

The installer links compatible skills into ~/.codex/skills and prompt entrypoints such as PUA into ~/.codex/prompts.

Preview first:

./install.sh --agent codex --dry-run

For Claude Code Users

Install shared skills plus Claude Code plugins:

./install.sh --agent claude-code

For a skills-only install that skips claude plugin commands:

./install.sh --agent claude-code --skip-plugins

See INSTALL.md, targets/codex, and targets/claude-code for the exact target layout.

For Kimi / Kimi Code CLI Users

Install only these Kimi-specific skills (works for both "Kimi" and "Kimi Code CLI"):

Skill	Why	Do NOT Install
kimi-mem	Cross-session memory management	-

💡 Note: "Kimi" and "Kimi Code CLI" refer to the same AI assistant. Skills marked for 🟢 Kimi work with both names.

Claude Code Plugin Notes

Codex Integration Inside Claude Code

The Claude Code bridge to Codex is a Claude plugin, separate from Codex native skills installed by ./install.sh --agent codex.

For Claude Code, use the official OpenAI plugin maintained in openai/codex-plugin-cc:

/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
/codex:setup

PUA Plugin

For PUA (我们不养闲 Agent), you have two installation options:

Option 1: Via Official Marketplace (Recommended for users)

/plugin marketplace add tanweai/pua
/plugin install pua@pua-skills
/reload-plugins

Option 2: Via Symlink (For developers)

# Link from this repository's third-party directory
ln -s /path/to/alexander-skills/third-party/pua ~/.claude/plugins/pua
/reload-plugins

💡 Which to choose?

• Marketplace: Better for regular users, version-controlled, easier to update
• Symlink: Better for developers who need to modify PUA source code

Other Claude-Specific Features

Skill	Purpose
Claudeception	Multi-layered agent collaboration
using-superpowers	Skill discovery and usage

Third-Party Projects

The following are third-party skill libraries included as git submodules:

Project	Description	Location	Installation
pua	AI Coding Agent skill plugin using PUA rhetoric to force AI to exhaust every solution	third-party/pua/	See Claude Code installation
orchestra-ai-research-skills	Orchestra AI research skills collection	third-party/orchestra-ai-research-skills/	Auto-loaded
anthropics-skills	Official Anthropic skills and examples	third-party/anthropics-skills/	Auto-loaded
vllm-omni-skills	vLLM-Omni multimodal model skills	third-party/vllm-omni-skills/	Auto-loaded

💡 Note: Most third-party projects are auto-loaded as skills. PUA requires explicit plugin installation for Claude Code users (see installation options above).

Plugins

The following are Claude Code plugins (not standalone skills):

Plugin	Description	Location
claude-mem	Cross-session memory for Claude Code	claude-plugins/plugins/

For Codex support inside Claude Code, install the external plugin from openai/codex-plugin-cc. For Codex native skills, use ./install.sh --agent codex.

🧪 Skill Validation

All skills must pass comprehensive validation before release:

# Quick validation (Levels 1-3)
python skills/skill-validation/scripts/validate_skill.py skills/<skill-name>/

# Headless agent testing (Levels 4-5)
python skills/skill-validation/scripts/test_multi_agent.py skills/<skill-name>/

# Or test specific agent
kimi -p "Test the <skill-name> skill"
claude -p "Test the <skill-name> skill"

Validation Levels:

Level	Test	Automated	Required
1	Structure (YAML, files)	✅	PR
2	Unit Tests	✅	If scripts
3	Compatibility	✅	PR
4	Kimi Integration	✅ Headless	Release
5	Claude Integration	✅ Headless	Release

See /skill:skill-validation for complete validation framework.

Quick Start

1. Clone the repository

git clone git@github.com:Alexander-Qiu/alexander-skills.git
cd alexander-skills
git submodule update --init --recursive

2. Install for your agent

# Codex
./install.sh --agent codex

# Claude Code
./install.sh --agent claude-code

3. Use the skills

After cloning, simply tell your AI agent to load the skills you need. For example:

Load skills from ./skills: kimi-mem, git-workflow

Most agents will automatically discover and load skills from the skills/ directory.

⚠️ Important for AI Agents: When loading skills, some files may be symbolic links (e.g., third-party skills linked from third-party/). You MUST follow these links and load the actual file content from the hard path. Do NOT skip or ignore symlinked files - resolve them to their target and load the real content into your context.

Some skills (like kimi-mem) require additional setup - the skill's documentation will guide you through it.

4. Install Claude Code plugins directly (optional)

For Claude Code users, install plugins separately:

# Use the interactive installer
cd claude-plugins
./install.sh

Or see claude-plugins/README.md for manual installation.

Note: Skills and plugins have different installation methods:

• Skills: Agent auto-discovers from skills/ directory
• Plugins: Must be installed to ~/.claude/plugins/ or ./.claude/plugins/

Skill Structure

alexander-skills/
├── README.md
├── INSTALL.md             # Agent install quickstart
├── install.sh             # Wrapper around scripts/init-agent.py
├── CONTRIBUTING.md       # ⭐ Development workflow
├── manifests/
│   └── skills.json        # Codex / Claude Code profiles
├── targets/
│   ├── codex/
│   └── claude-code/
├── skills/               # Standalone skills
│   ├── kimi-mem/         # Memory management skill
│   ├── git-workflow/     # Git workflow skill
│   └── ...
├── claude-plugins/       # Claude Code plugins
│   └── plugins/
│       └── claude-mem/   # Memory plugin for Claude
└── ...

🛠️ Development Workflow

⚠️ IMPORTANT: Never push directly to main!

Quick Workflow

# 1. Start from fresh main
git checkout main
git pull origin main

# 2. Create feature branch
git checkout -b feature/my-skill

# 3. Make changes and test
# ... edit files ...
npm run build  # if applicable

# 4. Commit
git add .
git commit -m "feat: Add my-skill"

# 5. Push and create PR
git push -u origin feature/my-skill
# ... create PR on GitHub ...

# 6. After merge, cleanup
git checkout main
git pull origin main
git branch -d feature/my-skill

See full workflow:

• CONTRIBUTING.md - Detailed contribution guide
• skills/git-workflow/SKILL.md - AI workflow skill

Branch Naming

feature/<skill-name>     # New skill
fix/<skill-name>-<issue> # Bug fix
docs/<what>              # Documentation
update/<skill-name>      # Update existing

Requirements

For kimi-mem

• Node.js 18+
• Kimi Code CLI
• SQLite (bundled)

Usage Examples

Save a memory

User: We fixed that memory leak issue!

AI: Let me record this important finding...

[Uses memory_save tool]
✅ Memory saved (ID: 123)

Search memories

User: How did we handle authentication before?

AI: Let me search our memory bank...

[Uses memory_search tool]
Found 3 related memories:
- [42] decision: Use JWT instead of Session
- [38] architecture: Auth service design
- [35] bugfix: Fixed token expiration issue

CLI Tool

Each skill may provide CLI tools:

# After npm install and build
cd kimi-mem

# Save memory via CLI
node dist/cli/index.js save -t "Important fix" -c "Details..." --type bugfix

# Search memories
node dist/cli/index.js search -q "authentication"

# View recent memories
node dist/cli/index.js recent

Development

Add a new skill

1. Create feature branch: git checkout -b feature/my-skill
2. Create directory: mkdir my-skill
3. Add SKILL.md with proper frontmatter
4. Add supporting code/files
5. Test locally
6. Commit and push: Follow CONTRIBUTING.md
7. Create PR for review

Update existing skill

1. Create fix branch: git checkout -b fix/skill-name
2. Modify files
3. Test locally
4. Commit with clear message
5. Push and create PR

Inspired By

This project is inspired by:

• obra/superpowers - Complete software development workflow for coding agents

Key lessons from superpowers:

• Systematic development workflow
• Composable, auto-triggering skills
• Test-driven skill development
• Clear documentation patterns

License

MIT - See individual skill directories for details.

Contributing

This is a personal skill library, but suggestions are welcome via GitHub Issues!

Before contributing:

1. Read CONTRIBUTING.md
2. Use /skill:git-workflow in Kimi CLI
3. Follow feature branch workflow