🚧 Work in Progress — Open Source & Community Driven
OMNISKILL is built by and for the open-source community. This project is actively evolving — contributions, ideas, and feedback are all welcome. Fork it, remix it, break it, build on it — it's yours. If you find it useful, give it a star and share it. If you want to make it better, open a PR. No gatekeeping, no strings attached. MIT licensed, forever free.
One repo. One format. Every platform. Best-in-class.
Write skills once. Run them on Claude Code, Copilot CLI, Cursor, Windsurf, and Antigravity.
OMNISKILL is a universal framework for AI coding assistant skills, agents, and workflows. It solves the fragmentation problem: instead of maintaining separate skill files for every AI tool, you write once in a universal format and deploy everywhere.
| Feature | Description |
|---|---|
| 🎯 Universal Skills | 83 skills in a single format that works on 5 AI platforms |
| 📦 Bundles | 12 domain kits (Godot, Web Dev, UX, Django, Security, DevOps...) |
| 🤖 Formal Agents | 10 agents with personas, skill bindings, guardrails, and handoff protocols |
| 🔄 Pipelines | 8 multi-agent workflows with failure recovery and context curation |
| 🧠 Synapses | 5 cognitive capabilities that shape HOW agents think |
| 🛡️ Guardrails & Hooks | 5 lifecycle hooks enforcing discipline at runtime |
| 📜 Runtime Contracts | Session state machine, central policy engine, telemetry, replay harness |
| 🧪 500 Tests | Full test suite with zero regressions across v2→v3 |
| 🔌 Cross-Platform | Claude Code, Copilot CLI, Cursor, Windsurf, Antigravity |
git clone https://github.com/SufficientDaikon/omniskill.git
cd omniskillpip install omniskill # From PyPI
omniskill init # Auto-detect your AI platforms
omniskill install --all # Install all 83 skills
omniskill doctor # Verify everything worksOr install manually:
python scripts/install.py # Auto-detects platforms
python scripts/install.py --platform claude-code # Specific platform
python scripts/install.py --bundle web-dev-kit # Specific bundle onlyomniskill pipeline run sdd-pipeline --project ./myappThis triggers the full Spec-Driven Development flow: spec-writer → context-curator → implementer → context-curator → reviewer with automatic failure recovery.
OMNISKILL uses a 6-layer architecture where each layer builds on the one below:
| Layer | What It Does |
|---|---|
| Layer 1 — Skills & Knowledge | 83 skills, 12 bundles, prompt library, knowledge sources |
| Layer 2 — Agents & Personas | 10 agents with formal personas, handoff contracts, and quality gates |
| Layer 3 — Synapses & Cognition | 5 cognitive synapses that shape HOW agents reason |
| Layer 4 — Pipelines & Orchestration | 8 pipelines with real execution, context curation, failure recovery |
| Layer 5 — Guardrails & Hooks | 5 lifecycle hooks enforcing discipline at runtime |
| Layer 6 — Runtime Contracts | Session state machine, policy engine, telemetry, MCP trust routing |
omniskill/
├── skills/ # 83 skills (SKILL.md + manifest.yaml + resources)
├── bundles/ # 12 domain bundles (bundle.yaml + meta-skill)
├── agents/ # 10 agent definitions (AGENT.md + agent-manifest.yaml)
├── pipelines/ # 8 multi-agent workflows
├── synapses/ # 5 cognitive synapses (SYNAPSE.md + manifest.yaml)
├── hooks/ # 5 lifecycle hooks
├── src/ # Core engine (v3.0 — session, policy, telemetry, replay)
├── prompts/ # Prompt library (router, system, personas)
├── sdk/ # Python SDK
├── adapters/ # 5 platform adapters
├── schemas/ # 15 validation schemas (9 v2 + 6 v3)
├── scripts/ # CLI tools (install, doctor, validate, admin)
├── tests/ # 500 automated tests
└── docs/ # Documentation site + markdown guides
83 skills in a universal format. Every skill has:
skills/my-skill/
├── SKILL.md # Instructions (Identity, Workflow, Rules, Output, Handoff)
├── manifest.yaml # Metadata (triggers, platforms, tags, dependencies)
├── resources/ # Reference materials (cheat sheets, style guides)
├── examples/ # Sample interactions
├── templates/ # Output templates
├── tests/ # Validation test cases
└── overrides/ # Platform-specific overrides
Key skills include: complexity-router (smart task routing), context-curator (pipeline context management), guard-chain (auth middleware patterns), structured-logging (observability), and 79 more.
See skills/_template/ for the template.
| Bundle | Skills | Description |
|---|---|---|
| godot-kit | 5 | Godot 4 / GDScript development |
| web-dev-kit | 10 | Frontend, React, RSC, i18n, error handling, backend |
| ux-design-kit | 7 | Full UX pipeline: research → wireframe → visual → test |
| django-kit | 4 | Django framework, ORM, REST APIs |
| sdd-kit | 6 | Spec-Driven Development: spec → implement → review |
| testing-kit | 5 | Unit tests, E2E testing, QA planning, debugging |
| mobile-kit | 2 | Mobile design, Capacitor best practices |
| meta-kit | 5 | Skill creation, discovery, packaging, prompts |
| prompts-chat-kit | 17 | Plugin system, builders, quality gates, webhooks, SDK |
| security-kit | 4 | Guard chain, webhooks, error handling, logging |
| data-layer-kit | 4 | Prisma ORM, singletons, deduplication, backend |
| devops-kit | 2 | Docker production builds, structured logging |
| Agent | Role | Primary Skills |
|---|---|---|
| spec-writer-agent | Specification Architect | spec-writer, prompt-architect |
| implementer-agent | Implementation Engineer | implementer |
| reviewer-agent | Compliance Reviewer | reviewer |
| debugger-agent | Debug Investigator | systematic-debugging |
| context-curator-agent | Context Architect | context-curator |
| dissector-agent | Codebase Analyst | knowledge-sources |
| ux-research-agent | UX Researcher | ux-research |
| ui-design-agent | Visual Designer | ui-visual-design, frontend-design |
| qa-master-agent | QA Engineer | e2e-testing-patterns, qa-test-planner |
| security-reviewer-agent | Security Auditor | guard-chain, error-handling-architecture |
Every agent has guardrails (guardrail-enforcement: strict), skill bindings, and handoff protocols. Run omniskill cards for machine-readable agent cards.
| Pipeline | Flow | Trigger |
|---|---|---|
| sdd-pipeline | spec-writer → context-curator → implementer → context-curator → reviewer | "build feature X from scratch" |
| ux-pipeline | research → context-curator → wireframe → visual → review → handoff | "design feature X" |
| debug-pipeline | debug → context-curator → implement → test → review | "fix bug X" |
| skill-factory | prompt → spec → context-curator → implement → validate → review | "create a new skill for X" |
| full-product | ux-pipeline → context-curator → sdd-pipeline → testing | "build product X end-to-end" |
| godot-pipeline | design → implement → test → review | "build Godot feature X" |
| dissect-to-skill | dissect → diff → specify → implement → validate → register | "dissect codebase X into skills" |
| skill-upgrade | assess → specify-upgrade → rewrite → verify | "upgrade stub skill to gold" |
Pipelines are resumable — if interrupted, they save state and continue from the last completed step.
Synapses are cognitive capabilities that shape HOW agents think, not WHAT they do.
| Synapse | Type | Phases | Description |
|---|---|---|---|
| metacognition | core | PLAN → MONITOR → REFLECT | Structured self-awareness: plan before acting, tag confidence |
| anti-rationalization | core | DETECT → CHALLENGE → ENFORCE | Prevents excuse-making via 10 Iron Laws |
| sequential-thinking | core | DECOMPOSE → REASON → VALIDATE → SYNTHESIZE | Step-by-step task decomposition |
| security-awareness | cross-cutting | SCAN → FLAG | Injects OWASP security checks during code tasks |
| pattern-recognition | optional | DETECT → SUGGEST → APPLY | Scans code for patterns, surfaces matching skills |
Core synapses fire automatically for every agent. Cross-cutting fire for specific task types. Optional require explicit binding.
| Platform | Adapter | Target Location |
|---|---|---|
| Claude Code | adapters/claude-code/ |
~/.claude/skills/ |
| GitHub Copilot CLI | adapters/copilot-cli/ |
~/.copilot/skills/ |
| Cursor | adapters/cursor/ |
.cursor/rules/ |
| Windsurf | adapters/windsurf/ |
.windsurfrules |
| Antigravity | adapters/antigravity/ |
.antigravity/skills/ |
📖 Browse the Full Documentation Site →
| Guide | Description |
|---|---|
| Getting Started | Installation, setup, your first skill |
| Creating Skills | Skill anatomy, manifest, SKILL.md authoring |
| Creating Bundles | Domain kits with meta-skill routing |
| Creating Agents | Agent personas, skill bindings, handoffs |
| Creating Pipelines | Multi-agent workflows with branching |
| Creating Synapses | Custom cognitive capabilities |
| Platform Guide | Platform-specific setup details |
| CLI Guide | Full CLI command reference |
| Architecture | 6-Layer design, data flow, validation |
| Guardrails | Guardrails engine, Iron Laws, deviation protocol |
| Sequential Thinking | Chain-of-thought protocol |
| Pipeline Orchestration | Execution engine, state persistence |
| DDE Methodology | Dissection-Driven Enhancement process |
| Agent Cards | Machine-readable agent metadata |
| MCP Integration Catalog | 20 curated MCP servers |
| Migration Guide (v2) | Upgrading from v0.x to v2.0 |
| Migration Guide (v3) | Upgrading from v2.0 to v3.0 |
| FAQ | Common questions answered |
See CONTRIBUTING.md for guidelines on adding skills, bundles, agents, pipelines, synapses, and hooks.
MIT © SufficientDaikon