Getting Started with OMNISKILL
Install OMNISKILL, set up your environment, and create your first skill in minutes.
📋 Prerequisites
Before installing OMNISKILL, make sure you have:
- Git installed on your system
- Python 3.8+ installed
-
At least one supported AI coding assistant:
- Claude Code
- GitHub Copilot CLI
- Cursor
- Windsurf
- Antigravity
⚡ Installation
Full Install (All Bundles, All Detected Platforms)
This is the recommended way to get started. OMNISKILL will auto-detect your installed AI coding assistants and install everything you need.
$ git clone https://github.com/tahaa/omniskill.git $ cd omniskill $ python scripts/install.py
Install a Specific Bundle
If you only need certain domain kits, you can install specific bundles:
$ python scripts/install.py --bundle web-dev-kit $ python scripts/install.py --bundle godot-kit
Install for a Specific Platform
Target specific AI coding assistant platforms:
$ python scripts/install.py --platform claude-code $ python scripts/install.py --platform cursor
Verify Installation
Run the doctor script to verify everything is set up correctly:
$ python scripts/doctor.py
If the doctor script reports no issues, you're ready to start using OMNISKILL!
🚀 Your First Skill
There are two ways to create your first skill:
Option A: Use the Skill Factory (Recommended)
The easiest way to create a new skill is to tell your AI assistant:
This triggers the skill-factory pipeline which guides you through the entire process interactively.
Option B: Manual Creation
For more control, you can create a skill manually:
-
Copy the TemplateStart by copying the skill template to your new skill directory.
$ cp -r skills/_template skills/my-skill -
Edit manifest.yamlFill in the skill metadata including name, description, and triggers.YAML
name: my-skill version: 1.0.0 description: "What your skill does" triggers: keywords: ["trigger phrase"] -
Write SKILL.mdDefine the skill's identity, workflow, rules, and output format.
-
Validate Your SkillEnsure your skill follows all conventions and has no conflicts.
$ python scripts/validate.py skills/my-skill -
Install the SkillDeploy your skill to your AI coding assistant platforms.
$ python scripts/install.py --skill my-skill
📦 Using Bundles
Bundles are domain kits that install multiple related skills at once:
$ # Install the web development kit $ python scripts/install.py --bundle web-dev-kit # This installs: frontend-design, react-best-practices, # vercel-react-best-practices, web-design-guidelines, backend-development # Plus the web-fullstack-expert meta-skill that composes them
Bundles include a meta-skill that routes between constituent skills and resolves conflicts. It acts as an intelligent orchestrator for the domain.
🔄 Using Pipelines
Pipelines are multi-agent workflows. Trigger them with natural language:
| Prompt | Pipeline | What It Does |
|---|---|---|
"Build feature X from scratch" |
SDD Pipeline | Spec → Implement → Review |
"Design feature X" |
UX Pipeline | Research → Wireframe → Visual → Review |
"Fix bug X" |
Debug Pipeline | Investigate → Fix → Test → Review |
"Create a new skill for X" |
Skill Factory | Interactive skill creation workflow |
Behind the scenes, every request passes through the Complexity Router which classifies your task (trivial → simple → moderate → complex → expert) and routes it to the optimal model tier and skill/agent/pipeline.
📚 Knowledge Sources
OMNISKILL can tap into external knowledge repositories:
- GitHub repos — Point to any public or private repository
- Local directories — Reference your project docs, specs, or notes
- URLs — Web documentation or API references
- APIs — Dynamic data sources
Knowledge sources use file-based search (grep/find/cat) — no vector
databases or embeddings required. Configure sources in
templates/source-config.yaml and sync them with:
$ python scripts/admin.py --sync
🔧 Self-Customization
OMNISKILL includes AI-guided skills for extending itself:
-
add-skill— Tell your AI: "Follow the add-skill skill to create a skill for [domain]" -
add-bundle— AI-guided bundle creation with checklists -
add-agent— AI-guided agent creation -
add-adapter— Create adapters for new platforms -
rename-project— Fork OMNISKILL and customize it for your organization
These skills provide step-by-step guidance and validation checks.
💻 Using the SDK
For programmatic access, use the Python SDK:
from sdk.omniskill import OmniSkill
os = OmniSkill()
# List available skills
skills = os.list_skills()
# Route a task to the right skill/agent
route = os.route("Create a React component")
# Sync knowledge sources
os.sync_sources()
# Health check
report = os.health_check()
v3 Runtime Contracts
from src.omniskill.core.session_manager import Session
from src.omniskill.core.policy_engine import PolicyEngine, PermissionRule
from src.omniskill.core.telemetry import TelemetryCollector
# Session lifecycle
session = Session.create("my-pipeline", {"input": "Build auth"})
session.activate()
# Policy-gated tool execution
engine = PolicyEngine()
engine.add_rule(PermissionRule(
id="r1", scope="file_*",
trust_tier="verified", action="allow"
))
decision = engine.evaluate("file_write", {"path": "/src/main.py"}, trust_tier="verified")
# Structured telemetry
collector = TelemetryCollector()
collector.emit_from_session_event(session, "session_start")
🎯 Next Steps
Now that you have OMNISKILL installed, explore these guides to go deeper:
Creating Skills
Deep dive into skill authoring
📦Creating Bundles
Group skills into installable kits
🤖Creating Agents
Define formal agent definitions
🔄Creating Pipelines
Build multi-agent workflows
🏗️Architecture
Understand the 6-layer architecture and runtime contracts
🔄Migration v3
Upgrade guide for v3.0 runtime contracts
🔧Platform Guide
Platform-specific setup details
❓FAQ
Common questions about new features