AI knowledge file pack builder

What it does

Open Knowledge Format (OKF) is a folder of Markdown files. Each file holds one idea and starts with a type line. Files link to each other, so people and AI tools can read, reuse and check what you know. You give your AI your notes, facts and sources. It writes one file per idea, picks stable folder names and adds the type and source to each. The optional Python checker then tests the types, reserved file names and links. The skill's own examples are about data tables. For business knowledge, such as a process or a policy, treat each process or policy as one file.

What goes in and out

You give it

• The knowledge, sources, document types, stable folder names, and intended readers.
• Source text (optional)
• Document (optional)

You get

• A linked Markdown knowledge folder, optional index and change log, and format review notes.

How the task works

• Plan stable knowledge files. Choose one concept per file and use stable paths so links keep their meaning.
• Write clear metadata. Add the required document type and useful optional fields. Keep sources and examples with the content.
• Check the format. Review document types, reserved filenames, and links. The helper checks a limited set of rules; it is not a full YAML parser.

Full instructions

• Open Claude, ChatGPT or Gemini. Paste the skill, or upload its folder if your AI can read files.
• Paste your notes, facts and sources. Say who will read the folder and what each topic should be called.
• Ask your AI to write the folder. It must be able to create files. If it cannot, ask it to return the file list and the text of each file, then save each file by hand. Check that each file has a type and that links point to real files.
• The bash steps in the skill and the Python checker are for developers. A business user can skip them and ask the AI to write the files. The checker is optional. To run the checker, you need Python 3.8 or newer with PyYAML. Without Python, your AI can still write the folder, and you check the links by hand. Create each folder before you add files to it.
• Next, save the folder in a repo or shared drive. Point your AI tool's file access at its index.md. Choose the notes wiki builder if your notes grow over time, the company wiki planner for team processes with owners, or the text to knowledge graph builder to pull links out of one long document.

Full instruction file

Open Knowledge Format (OKF) — v0.1

Google Cloud published the Open Knowledge Format v0.1 on June 12, 2026 — an open specification that formalizes the LLM-wiki pattern into a portable, vendor-neutral format for AI agent knowledge.

The design is intentionally minimal: a directory of markdown files with YAML frontmatter. No schema registry, no central authority, no required tooling.

"If you can cat a file, you can read OKF; if you can git clone a repo, you can ship it."

Before you start

Use the material below to do this job. If facts or tool access are missing, explain what you can finish and what must wait.

What you need (use your own material or the linked skill):

• Your question (yours): the decision or problem, and what a good outcome looks like.

Makes it better (optional):

• Source text (yours): a long source in full: an article, a video or talk transcript, or notes from a book, with its title, author and link.
• A document (yours): a finished document, such as a Word file, a PDF or a page of notes.

How to start:

1. Look first. Use what they've already attached, pasted or put in their about-me note. In a coding assistant with file access, look in the project too for these by name.
2. Check what's missing. Look for your question, source text, a document. They can attach files, paste the details, or share a file path your assistant can read. Ask only for missing facts that change the result. If they ask you to start with what they have, complete the parts you can and mark the gaps.
3. Do the job. Follow the skill below with the facts and tools you have. Mark each gap with MISSING: where it belongs and never invent a fact to fill it.

Tools that do this best:

• File access; Python for optional checks: The bundled helper reads a simple subset of YAML. A full YAML parser is needed to check complete YAML syntax.

What you need

The knowledge to put in the bundle: pasted text, a source such as an article or transcript, or a finished document such as meeting minutes.

Core Concepts

Reference Files

Scripts

Assets

Bundle Structure

path/to/bundle/
├── index.md                      # Optional. Directory listing for progressive disclosure.
├── log.md                        # Optional. Chronological update history.
├── <concept>.md                  # A concept at the bundle root.
└── <subdirectory>/               # Subdirectories organize concepts into groups.
    ├── index.md
    ├── <concept>.md
    └── <subdirectory>/
        └── …

Files named index.md and log.md are reserved — they have defined semantics and must not be used for concept documents.

Frontmatter Fields

---
type: <Type name>                  # REQUIRED
title: <Optional display name>
description: <Optional one-line summary>
resource: <Optional canonical URI for the underlying asset>
tags: [<tag>, <tag>, …]            # Optional
timestamp: <ISO 8601 datetime>     # Optional last-modified time
# … other producer-defined key/value pairs
---

Required: type — a short string identifying the kind of concept. Types are NOT registered centrally. Producers pick descriptive values; consumers tolerate unknown types gracefully.

Conventional body sections:

Cross-linking

Concepts link to each other via standard markdown links:

• Absolute (bundle-relative): Begin with /, relative to bundle root. Recommended — stable when documents move.
• Relative: Standard markdown relative paths.

A link from concept A to concept B asserts a relationship. The specific kind (references, depends-on, joins-with) is conveyed by the surrounding prose, not by the link syntax.

Consumers MUST tolerate broken links — a link whose target doesn't exist is not malformed; it may represent not-yet-written knowledge.

Conformance

A bundle is conformant with OKF v0.1 if:

1. Every non-reserved .md file contains parseable YAML frontmatter
2. Every frontmatter block contains a non-empty type field
3. Reserved filenames (index.md, log.md) follow their defined structure

Consumers MUST NOT reject a bundle for: missing optional frontmatter fields, unknown type values, unknown frontmatter keys, broken cross-links, or missing index.md files.

Relationship to Other Formats

Quick Start

# 1. Create a bundle directory
mkdir my-knowledge && cd my-knowledge
mkdir -p datasets

# 2. Create a concept file
cat > datasets/sales.md << 'EOF'
---
type: BigQuery Dataset
title: Sales
description: All sales-related tables for the retail business.
tags: [sales]
timestamp: 2026-06-18T00:00:00Z
---
The sales dataset contains [orders](/tables/orders.md) and [customers](/tables/customers.md).
EOF

# 3. Add an index for navigation
cat > index.md << 'EOF'
# Knowledge Bundle

* Sales Dataset (`datasets/sales.md`) - Retail sales data in an example OKF bundle
EOF

# 4. Validate your bundle
python3 scripts/okf-bundle-validate.py .

Gotchas

• Type values are not registered centrally. Pick descriptive types (BigQuery Table, Playbook, Metric). Consumers should handle unknown types gracefully.
• File path is the identity. Renaming a file changes its concept ID. Use stable paths or add redirects in documentation.
• index.md uses no frontmatter (except optionally at the bundle root for okf_version). The body is a markdown list with links.
• log.md date headings must be ISO 8601 (YYYY-MM-DD).
• Cross-links are untyped. The relationship semantics live in the prose, not in the link syntax. Graph builders treat all links as directed edges.
• OKF does not specify a tag-browsing format. Producers that want tag aggregation should synthesize it at consumption time by scanning frontmatter.
• OKF v0.1 is a draft. The spec will evolve. Minor bumps add backward-compatible features; major bumps may break required fields. Consumers should do best-effort consumption on unknown versions.

After this

When you've handed it back, suggest what they could do next in one short question. It's a suggestion: if they're done, that's fine. Offer these, and say they can type their own:

• The Notes wiki builder. It takes the document.
• The Company wiki planner.
• Decide what to do with the Text-to-knowledge-graph builder. If they pick one and that skill is added here, carry straight on with what you just made. If not, give the link and say what to bring.