OKF tutorial: build your first knowledge bundle

What you will build

The final output is a website content bundle with one root index and three focused concept files. The same shape can later be adapted for APIs, documentation, data catalogs, and support playbooks.

bundle/
  index.md
  pages/
    home.md
    pricing.md
  support/
    contact.md

Before you start

• Choose one narrow scope, such as five public pages or three API endpoints.
• Collect canonical source URLs before writing summaries.
• Do not include secrets, private keys, personal data, or internal-only records.
• Use plain Markdown so the bundle remains readable in Git and static hosting.

Step 1: choose one knowledge scope

Start with a narrow scope. A good first bundle might cover five important pages, three API endpoints, or one support workflow.

Do not start by converting everything. OKF works best when each file has a clear concept boundary.

Step 2: create a folder

Create a bundle folder with an `index.md` file at the root. The index file helps humans and agents see what is available before opening individual concept files.

• `bundle/index.md` lists sections and links.
• `bundle/pages/home.md` describes one page.
• `bundle/support/contact.md` describes one support route.

Step 3: write frontmatter first

Frontmatter is the structured block at the top of a Markdown file. In this guide, the validator checks `type`, `title`, `description`, and `tags` so your documents are useful for search and preview workflows.

Step 4: write a focused Markdown body

• Use headings for audience, facts, rules, examples, and citations.
• Use lists and tables where structure matters.
• Use ordinary Markdown links to connect related concept files.

Step 5: validate and review

Paste each file into the OKF validator on this site. The validator runs in your browser, which means pasted content is not uploaded to this website.

After validation, review the file for accuracy. A passing file can still contain outdated or unsupported claims.

Step 6: consume OKF context in AI agent workflows

Once validated, an AI agent framework (such as LangChain or custom Python scripts) can read the YAML metadata to filter active concepts, check freshness timestamps (`stale_after`), and pass verified Markdown bodies into prompt context (`verified`).

Final output

A useful first bundle has a root index, one concept per file, canonical resource links, and enough body text for a reviewer to understand the source without guessing.

---
type: Website Page
title: Contact page
description: Explains how visitors can report corrections, outdated examples, and validator issues.
resource: https://openknowledgeformat.online/contact/
tags: [website, contact, corrections]
timestamp: 2026-06-21T00:00:00Z
---

# Purpose

This page gives readers a correction path for Open Knowledge Format Guide.

# Related concepts

- [Privacy policy](../pages/privacy.md)
- [Validator](../products/okf-validator.md)

Validation checklist

• The frontmatter parses cleanly and starts at the top of the file.
• `type` is present and describes the concept category.
• `resource` points to a real URL, URN, or stable asset identifier.
• The Markdown body has one clear H1 and focused sections.
• Links point to real related files or public sources.
• Remove private keys, personal data, and internal-only material before public hosting.
• Add a timestamp or review date so stale files are easier to spot.

Review v0.2 trust fields

• Use `sources` and `provenance` notes to point reviewers back to the source material.
• Use `status` and `stale_after` only when an owner can review whether the file still reflects the source.
• A `generated` or `verified` label is not proof that every claim in the Markdown body is correct.

FAQ