Okf Skill

OKF (Open Knowledge Format) Skill

Use the Open Knowledge Format (OKF) when storing, retrieving, or documenting knowledge. OKF makes knowledge portable across agents, tools, and teams.

What to do

When documenting data systems (databases, APIs, pipelines, metrics) or building a knowledge base, produce OKF-conformant bundles. When reading existing knowledge, prefer OKF bundles.

File structure

knowledge/
├── index.md                  # Root listing of all concepts
├── datasets/
│   ├── index.md
│   └── sales.md
└── tables/
    ├── index.md
    └── orders.md

Concept format

Every .md file (except index.md) is a concept. Required and recommended frontmatter:

---
type: Table                    # REQUIRED (e.g. Table, Dataset, Metric, API, Playbook)
title: Orders                  # Recommended, display name
description: Customer orders   # Recommended, one-line summary
resource: https://...          # Optional, canonical URI
tags: [sales, orders]          # Optional, cross-cutting tags
timestamp: 2026-06-17T00:00:00Z # Optional, ISO 8601
---

# Schema
| Column     | Type   | Description                |
|------------|--------|----------------------------|
| `order_id` | STRING | Unique identifier           |

# Related
See [customers](/tables/customers.md) for the join key.

Rules

1. type is the only required field. All else is optional.
2. Reserved filenames: index.md (directory listing) and log.md (changelog). Never use as concept names.
3. Cross-link with markdown: [text](/path/to/concept.md) (absolute) or [text](./relative.md).
4. Always add description, it's used in indexes and search.
5. Create index.md in each directory listing its contents.
6. After writing, verify the bundle is self-consistent: links resolve, types are set, index exists.

Types (suggested)

| Type | Use for | |------|---------| | Dataset | Collection of tables or views | | Table | A database table or view | | API | REST/gRPC endpoint | | Metric | Business or technical metric | | Playbook | Operational procedure | | Pipeline | Data processing pipeline | | Reference | External docs or glossary |

Define your own types as needed. Consumers will handle unknown types gracefully.

Workflow

1. Document (when asked to document a system, create an OKF bundle with index.md at root, organized by type.)
2. Read (before answering questions about a system, check if an OKF bundle exists. Read index.md first, then drill into relevant concepts.)
3. Update (when schemas or systems change, update the relevant concept files and their timestamps.)
4. Cross-reference (link related concepts. The value of OKF is in the graph, not the individual files.)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: catancs
• Source: catancs/okf-skill
• License: MIT

Install and usage instructions live in the source repository linked above.