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 are portable — they work with Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Warp, and any agent that supports the agentskills.io spec.
# 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| 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 |
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?
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
# 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
| 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 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
# 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.# Install skills first
npx skills add lamtuanvu/agentskills@speckit-brainstorm -g -y
npx skills add lamtuanvu/agentskills@speckit-orchestrator -g -y
# Then install the plugin for full automation
/plugin install speckit-orchestrator@lamtuanvu-marketplace
The plugin adds:
- Auto-continuation: stop hook feeds
/speckit-orchestrator:executeafter each step — pipeline runs to completion unattended (pauses only at the clarify checkpoint) - Parallel brainstorm: agent team (architect, UX analyst, feasibility analyst, devil's advocate) analyzes your feature from all angles before writing
idea.md - Parallel plan review: security, performance, conventions, and UI specialists review the plan in parallel
- Parallel implementation: tasks partitioned across multiple implementers + dedicated test-writer and QA reviewer
Commands:
/speckit-orchestrator:brainstorm <feature description>
/speckit-orchestrator:execute
/speckit-orchestrator:status
/speckit-orchestrator:rollback <step>
/speckit-orchestrator:cancel-pipeline
agentskills/
├── skills/ # Portable agentskills.io skills
│ ├── speckit-orchestrator/
│ │ ├── SKILL.md # Pipeline instructions + loop logic
│ │ ├── scripts/ # Shared Python scripts (canonical)
│ │ │ ├── orchestrator.py # State machine
│ │ │ ├── verify_state.py # State verification
│ │ │ ├── partition_tasks.py # Task partitioning
│ │ │ └── init_feature.py # Feature initialization
│ │ └── references/ # Step-by-step guides
│ │ ├── step-specify.md
│ │ ├── step-clarify.md
│ │ ├── step-plan.md
│ │ ├── step-tasks.md
│ │ ├── step-analyze.md
│ │ └── step-implement.md
│ └── ... # Other skills
├── plugins/ # Claude Code plugins
│ ├── speckit-orchestrator/
│ │ ├── hooks/ # Stop hook (auto-continuation)
│ │ ├── commands/ # Slash commands
│ │ └── agents/ # Team agent definitions
│ │ # Scripts inherited from skills/speckit-orchestrator/scripts/
│ └── github-ci/ # CI failure MCP channel server
└── .claude-plugin/
└── marketplace.json # Claude Code plugin registry
Inheritance: The speckit-orchestrator plugin has no scripts of its own — it references skills/speckit-orchestrator/scripts/ directly. The skill is the base; the plugin is the Claude Code enhancement layer.
See CONTRIBUTING.md for how to add a skill or plugin.
Apache 2.0 — see LICENSE