OKF vs Claude Skills:
What's the Difference?

Quick Overview

If you're building AI agent knowledge systems in 2026, you've likely encountered two competing approaches: OKF (Open Knowledge Format) and Claude Skills (Anthropic's proprietary skill system). Both aim to give AI agents specialized knowledge, but they take fundamentally different approaches.

Below, we break down both systems, compare them across key dimensions, and help you decide which approach — or combination — is right for your project.

What Are Claude Skills?

Claude Skills is a feature of Anthropic's Claude platform that allows users to define custom behaviors, knowledge, and capabilities for their Claude AI assistant. Think of it as a way to "train" Claude with your own domain expertise — without needing to fine-tune a model.

How It Works

Claude Skills are defined through the Claude interface or API. A Skill consists of:

• Custom instructions — a system prompt that tells Claude how to behave, what tone to use, and what rules to follow
• Knowledge documents — files (markdown, PDF, text) that Claude can reference when answering questions in your domain
• Skill configuration — metadata like name, description, and visibility settings

// Claude Skill structure (conceptual)
{
  "name": "my-domain-expert",
  "description": "A domain expert skill for my project",
  "instructions": "You are a domain expert...",
  "knowledge": [
    "file:///docs/architecture.md",
    "file:///docs/api-reference.md"
  ],
  "visibility": "private"
}

Key Characteristics

• Proprietary platform — Skills only work within the Anthropic/Claude ecosystem
• No open format — the Skill definition is controlled by Anthropic and subject to change
• Knowledge files — you can upload documents, but they're stored inside Anthropic's infrastructure
• Skill Store — Anthropic maintains a marketplace for publishing and discovering Skills
• No version control — Skills are managed through the Claude UI, not through git or standard tooling

Claude Skills are convenient if you're already an Anthropic customer and only need to customize Claude. But the knowledge you put into a Skill is effectively locked into Anthropic's platform.

What Is OKF?

OKF (Open Knowledge Format) is an open standard for packaging structured knowledge as portable, version-controlled markdown bundles. Launched by Google Cloud Platform in mid-2026, it's designed for AI agents to consume directly — regardless of which platform or model they're running on.

How It Works

my-knowledge-bundle/
├── index.md         # Bundle manifest with YAML frontmatter
├── log.md           # Changelog
├── concepts/
│   ├── setup.md     # Concept document with metadata
│   └── architecture.md
└── guides/
    ├── getting-started.md
    └── configuration.md

Key Characteristics

• Open standard — the specification is public and community-governed
• Portable — OKF bundles work with any AI agent, any platform, any model
• Directory structure — knowledge is organized in a hierarchy of markdown files
• YAML frontmatter — every file has structured metadata (title, description, tags, version)
• Version-controlled — bundles live in git, with full history and collaboration via PRs
• Discoverable — bundles can be published on BundleDex, the open directory of OKF bundles

OKF is the format used by BundleDex, the definitive directory of OKF knowledge bundles. It's ideal for portable, reusable knowledge that works across any AI ecosystem — not just one platform.

Key Differences at a Glance

Open vs. Proprietary

This is the most fundamental difference. OKF is fully open — its specification is public, its tooling is open source, and its bundles are portable across any platform. Claude Skills is proprietary — the format, the runtime, and the distribution (Skill Store) are all controlled by Anthropic. If Anthropic changes the Skills format or discontinues the feature, your knowledge investments may be stranded.

Format and Structure

OKF uses a directory of markdown files with YAML frontmatter for metadata. This is a standard filesystem format that any tool can read. Claude Skills uses a proprietary JSON-based definition with knowledge files uploaded to Anthropic's servers. There's no standard way to inspect, validate, or version-control a Claude Skill outside of the Anthropic platform.

Portability

An OKF bundle can be used with any AI agent, any model, any platform. Download it from GitHub, load it into an agent's context, or serve it via an API — it works everywhere. A Claude Skill is locked to the Claude platform. You cannot use a Claude Skill with ChatGPT, Gemini, or any other AI assistant. This is a critical consideration for organizations that value platform independence.

Version Control and Collaboration

OKF bundles are git-native. You can fork, branch, PR, diff, and merge knowledge changes just like code. Claude Skills are managed through the Claude UI — there's no built-in version control, no diffing, no collaborative review workflow. Teams that treat knowledge as code will find OKF far more aligned with their development practices.

Ecosystem and Distribution

OKF has BundleDex — an open, searchable directory of knowledge bundles. Anyone can publish a bundle, and anyone can discover and use bundles through the BundleDex API, MCP endpoint, or web interface. Claude Skills have the Skill Store — a marketplace controlled by Anthropic. Publishing on the Skill Store requires Anthropic's approval, and distribution is limited to Claude users.

Offline and Air-Gapped Use

An OKF bundle is just files on disk. You can use it completely offline — load it into an agent, reference it in a prompt, or process it with custom tooling. Claude Skills require connectivity to Anthropic's servers. For air-gapped environments, defense, on-premise deployments, or any scenario where data cannot leave the network, OKF is the only viable option.

Side-by-Side Comparison

When to Use Each

Use OKF When...

• You want your knowledge to work across multiple AI platforms (Claude, ChatGPT, Gemini, open-source models)
• You value platform independence and don't want vendor lock-in
• Your knowledge changes over time and needs version control and collaboration workflows
• You need to use knowledge in offline or air-gapped environments
• You want your knowledge discoverable on BundleDex by a wide audience of AI agents
• You're building SDKs, libraries, or developer tools that agents need to reference
• You want to share knowledge with a community outside the Anthropic ecosystem

Use Claude Skills When...

• Your entire AI workflow is already within the Anthropic/Claude ecosystem
• You need deep integration with Claude's native features (Artifacts, Projects, etc.)
• You want to publish a Skill in the Anthropic Skill Store for Claude users
• You don't need portability — you're optimizing for a single platform
• You prefer the simplicity of configuring skills through a UI rather than managing files
• You're building a custom Claude experience for your team and don't need to share it outside

The Hybrid Approach: Best of Both Worlds

Many teams use OKF as their source of truth and then generate Claude Skills as one of several output formats. This gives you the portability and version control of OKF while still being able to deploy on Claude. Here's how it works:

Source of Truth: OKF Bundle (git repository)
├── concepts/
├── guides/
├── references/
└── index.md
        │
        ▼
Generated Outputs:
├── Claude Skill (via conversion script)
├── ChatGPT GPT configuration
├── Gemini knowledge config
├── Static HTML docs
└── llms.txt reference

This approach ensures your knowledge is always portable and open, while still being accessible on the platform of your choice.

Can You Use OKF Bundles with Claude?

Indirectly, yes. While Claude does not natively support the OKF format, you can use OKF bundles with Claude in several ways:

Method 1: Upload Files Directly

Claude supports uploading markdown files, PDFs, and text files as context. You can download an OKF bundle from BundleDex and upload individual files from the bundle into a Claude conversation. Claude will read the content and use it to inform its responses. This is the simplest approach for one-off use cases.

Method 2: Create a Claude Skill from an OKF Bundle

You can convert an OKF bundle into a Claude Skill by combining its files into a single instruction document or by uploading the bundle's markdown files as knowledge documents. While there's no automated conversion tool yet, the process is straightforward:

# Step 1: Clone the OKF bundle
git clone https://github.com/example/my-okf-bundle.git
cd my-okf-bundle

# Step 2: Concatenate files into a single knowledge document
cat concepts/*.md guides/*.md > claude-skill-knowledge.md

# Step 3: Create a Skill in the Claude UI with:
#   - Instructions: "You are an expert in [domain]..."
#   - Knowledge: Upload claude-skill-knowledge.md
#   - Visibility: Private or Published

Method 3: Copy-Paste Context

For quick reference, you can copy the contents of an OKF bundle's key files directly into a Claude conversation. This is useful for ad-hoc questions where you don't need a persistent Skill.

Method 4: BundleDex MCP Endpoint

If you're using Claude with MCP (Model Context Protocol) support, you can connect to the BundleDex MCP server. This allows Claude to search and retrieve OKF bundles on demand during a conversation, without manual file uploads.

Ecosystem & Community

The OKF Ecosystem

OKF has a rapidly growing ecosystem centered around BundleDex — the open directory of knowledge bundles. The ecosystem includes:

• BundleDex — the definitive directory, with search, browse, and API access
• Open source tooling — validators, generators, and converters for OKF bundles
• Cross-platform support — OKF bundles work with any AI agent, including Claude, ChatGPT, Gemini, and open-source models
• Community governance — the OKF specification is community-driven and open to contributions
• MCP integration — BundleDex provides an MCP endpoint for agent-native discovery

The Claude Skills Ecosystem

Claude Skills are part of Anthropic's closed ecosystem:

• Skill Store — a marketplace controlled by Anthropic, with approval required for publishing
• Claude Pro / Team / Enterprise — Skills are available on paid Claude plans
• No open tooling — there are no official open-source tools for creating, validating, or managing Skills outside the Claude platform
• Platform lock-in — Skills cannot be exported or used on other platforms

Which Ecosystem Is Right for You?

If you value openness, portability, and community governance, OKF's ecosystem is the clear choice. If you're all-in on Anthropic and don't need platform independence, Claude Skills may be more convenient — but be aware of the vendor lock-in tradeoff.

Many organizations are adopting a dual strategy: maintain knowledge as OKF bundles for maximum portability, and generate Claude Skills as a convenience for team members who work primarily in Claude. This gives you the best of both worlds without sacrificing platform independence.

FAQ