agentskills
A collection of AI agent skills and Claude Code plugins. Skills work across any agent that supports the agentskills.io spec. Claude Code plugins layer on top of skills to add hooks, slash commands, and parallel agent teams.
Skills
Skills are portable — they work with Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Warp, and any agent that supports the agentskills.io spec.
Install
# Install all skills globally
npx skills add lamtuanvu/agentskills -g -y
# Install a single skill
npx skills add lamtuanvu/agentskills@gemini-image-gen -g -y
Available Skills
| Skill | Description |
|---|
| agent-manager | Manage local CLI agents via tmux sessions (start/stop/monitor/assign) |
| brand-guidelines | Anthropic brand colors and typography for any artifact |
| find-docs | Fetch up-to-date docs for any library, framework, or SDK via Context7 |
| gemini-image-gen | Generate SVG vector graphics and PNG images via Google Gemini API |
| mcp-builder | Guide and templates for building MCP servers (Python/Node) |
| plugin-creator | Scaffold and develop Claude Code plugins incrementally |
| skill-creator | Guide for creating agentskills-compatible SKILL.md packages |
| speckit-brainstorm | Brainstorm a feature, define requirements, produce idea.md |
| speckit-orchestrator | Run the SpecKit pipeline: specify → clarify → plan → tasks → analyze → implement |
| vectcut-api | CapCut/JianYing video editing API reference and Python client |
| video-proposal | Analyze a clip collection and propose short-form video concepts |
Usage
Once installed, invoke a skill by describing the task. Your agent will automatically select the relevant skill. You can also invoke explicitly:
# Claude Code
/find-docs how do I use React Server Components?
/gemini-image-gen a minimalist settings gear icon
# Other agents (varies by agent)
@find-docs how do I configure Prisma with PostgreSQL?
Claude Code Plugins
Plugins are Claude Code-specific. They extend skills with:
- Stop hooks — auto-continue pipelines without user intervention
- Slash commands —
/speckit-orchestrator:execute, /speckit-orchestrator:status, etc.
- Agent teams — parallel specialist agents for plan review and implementation
Install
# Step 1 — add this repo as a marketplace source (one time)
/plugin marketplace add https://github.com/lamtuanvu/agentskills
# Step 2 — install a plugin
/plugin install speckit-orchestrator@lamtuanvu-marketplace
/plugin install github-ci@lamtuanvu-marketplace
Available Plugins
| Plugin | Enhances | What it adds |
|---|
| speckit-orchestrator | speckit-brainstorm + speckit-orchestrator skills | Stop hook auto-continuation, /speckit-orchestrator:* commands, parallel agent teams for plan review and implementation |
| github-ci | — | Webhook listener for GitHub CI failures; pushes failures into your Claude Code session for automated diagnosis and fix |
| capcut-api | vectcut-api skill | CapCut/JianYing MCP tools for direct video editing |
| ai-video-editor | video-proposal skill | Scene detection, audio transcription, vision analysis for video metadata |
SpecKit Workflow
SpecKit is a structured feature development pipeline built on the speckit-brainstorm and speckit-orchestrator skills.
speckit-brainstorm → idea.md → speckit-orchestrator → spec → clarify → plan → tasks → analyze → implement
With any agent (skill only)
# 1. Install the skills
npx skills add lamtuanvu/agentskills@speckit-brainstorm -g -y
npx skills add lamtuanvu/agentskills@speckit-orchestrator -g -y
# 2. Brainstorm a feature
# Invoke speckit-brainstorm with your feature idea.
# It explores the codebase, asks clarifying questions,
# and produces docs/features/<name>/idea.md on approval.
# 3. Run the pipeline
# Invoke speckit-orchestrator.
# It reads idea.md, runs each step, and pauses at clarify for your review.
# After clarify, re-invoke to continue through to implementation.
With Claude Code (skill + plugin)