One repo. One format. Every platform.
The universal AI agent & skills framework. Runtime contracts, policy-gated tool execution, smart complexity routing, pluggable knowledge sources, self-customization, and a Python SDK. Deploy to Claude Code, Copilot CLI, Cursor, Windsurf, and Antigravity.
A complete framework covering the full lifecycle — from skill creation to cross-platform deployment.
Write skills in a single, standardized format that works seamlessly across 5+ AI coding platforms.
Install entire domain kits as one unit — web dev, UX design, game dev, and more, all pre-configured.
Define personas, skill bindings, guardrails, and handoff protocols for sophisticated AI agent behavior.
Multi-agent workflows that chain agents together — from spec writing to implementation to review.
AI-powered pipeline that creates new skills automatically — design, spec, implement, validate, review.
Built-in adapters for Claude Code, Copilot CLI, Cursor, Windsurf, and Antigravity.
Cheat sheets, style guides, decision trees, and references that give your agents deep domain knowledge.
Schema-based validation for everything — skills, agents, bundles, and pipelines all checked for correctness.
OMNISKILL uses its own skill-factory pipeline to create, validate, and improve itself recursively.
Smart routing classifies tasks by complexity and routes to the optimal model — fast models for trivial, premium for expert.
Plug GitHub repos, local dirs, and URLs as knowledge. File-based search with grep/find — no vector DB needed.
AI-assisted skills guide agents through creating new skills, bundles, agents, and platform adapters.
Organized system prompts and switchable personas — Expert, Quick, and Teacher modes for any interaction style.
Full programmatic access — list skills, route queries, install to platforms, sync sources, and run health checks.
Comprehensive CLI with stats, error logs, source management, sync controls, and full health reports.
Cognitive enhancements that shape how agents think — metacognition, confidence calibration, stuck detection, and self-reflection.
v3 enforced session lifecycle, policy-gated tool execution, and replay determinism — no tool runs without a policy decision.
Default-deny tool gating with trust tiers (builtin → verified → community → untrusted) and full audit trail.
Versioned telemetry envelopes, correlation IDs, and structure-only checksums for deterministic replay across environments.
Curated skill collections that install as a single unit. Each bundle is a ready-to-use toolkit.
Godot 4, GDScript, scene management, physics, and shaders.
Frontend, React, backend APIs, and full-stack patterns.
Full UX pipeline — research, wireframes, design systems, and testing.
Django framework, REST APIs, models, and deployment patterns.
Spec-Driven Development — spec writing, implementation, and review.
E2E testing, debugging, test generation, and quality assurance.
Mobile development patterns and cross-platform strategies.
Skill creation, validation, packaging, and meta-tooling.
Chain agents together into powerful automated workflows that handle entire development lifecycles.
Write once. Adapters handle the rest. Your skills work everywhere your team does.
Anthropic
GitHub
Anysphere
Codeium
Antigravity
Each layer builds on the one below it — from raw platform adapters up to runtime contracts that enforce policy and determinism.
Clone the repo, install the skills to your preferred platform, and run the doctor to verify everything is working.