Practical notes: OKF Is More Than Markdown: How I Think About a Portable

This walkthrough rebuilds the path from raw materials to a working system for: OKF Is More Than Markdown: How I Think About a Portable Knowledge Layer for AI Agents. The focus is operable steps, explicit checks, and code that you can drop into a repo without guessing intent. For the Overview stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

knowledge/
├── index.md
├── orders/
│   ├── index.md
│   └── order-lifecycle.md
├── payments/
│   ├── index.md
│   └── payment-failures.md
└── policies/
    └── refund-eligibility.md

---
type: Business Rule
title: Refund Eligibility
description: Rules for determining whether an order is eligible for a refund.
tags: [orders, refunds]
---
# Refund Eligibility
An order is eligible for a refund when...
See also [Order Lifecycle](../orders/order-lifecycle.md).

Agents are powerful, but they still don’t know how your product works

When working through the Agents are powerful but stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Wiki
PDF
Product docs
Shared Drive
API documentation
Source code
JIRA Tickets
Slack threads
Google Drive
People's heads

The important part of OKF is not Markdown

When working through the The important part of stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Bundles, concepts, and the part you find most interesting: index.md

When working through the Bundles concepts and the stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the Bundles concepts and the stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

store-knowledge/
├── index.md
├── orders/
│   ├── index.md
│   ├── order-lifecycle.md
│   └── cancellation.md
├── payments/
│   ├── index.md
│   └── payment-status.md
└── policies/
    ├── index.md
    └── refund-eligibility.md

Thousands of documents
          ↓
Load everything into context

index.md

Orders
Payments
Promotions
Refund Policies
Customer Support

policies/index.md

Refund Eligibility
Partial Refunds
Manual Review

policies/refund-eligibility.md

[Order Lifecycle](../orders/order-lifecycle.md)

Bundle
  ↓
index.md
  ↓
policies/
  ↓
index.md
  ↓
refund-eligibility.md
  ↓
order-lifecycle.md

Isn’t this just RAG?

The Isn t this just stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Separate chunking policy from retrieval policy. Changing one should not force a rewrite of the other when quality metrics move.

Question
   ↓
Retrieve relevant knowledge
   ↓
Put knowledge into context
   ↓
Generate an answer

User Question
      ↓
Vector / Hybrid Search
      ↓
Find an OKF concept
      ↓
Read the concept
      ↓
Follow indexes or links if needed
      ↓
Build the final context

{
  "title": "...",
  "source": "...",
  "type": "...",
  "updated_at": "...",
  "owner": "..."
}

What about Agent Skills?

The What about Agent Skills stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

1. Identify the customer issue
2. Inspect the order status
3. Check the applicable refund policy
4. Determine whether escalation is required
5. Draft a response

customer-support-skill/
├── SKILL.md
└── references/
    └── commerce-knowledge/
        ├── index.md
        ├── orders/
        ├── payments/
        └── policies/

The OKF use case you find most interesting is actually local

The The OKF use case stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts. The The OKF use case stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

storefront/
├── src/
├── tests/
├── AGENTS.md
└── knowledge/
    ├── index.md
    ├── checkout/
    ├── orders/
    ├── payments/
    ├── refunds/
    └── promotions/

read code
→ understand code
→ modify code

Coding Task
     ↓
Read code
     +
Read refund policy
     +
Read payment constraints
     +
Read order lifecycle
     ↓
Understand actual product behavior
     ↓
Modify code

Modify code
     ↓
Product behavior changed
     ↓
Update knowledge

Local and central knowledge don’t have to compete

For the Local and central knowledge stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Company terminology
Shared authentication rules
Common API contracts
Billing definitions
Customer policies
Security guidelines

                   Agent
                  /     \
                 /       \
                ↓         ↓
          Local OKF    Central OKF
          project       shared
          knowledge     knowledge

OKF
= Knowledge Artifact

MCP
= Serving / Tool ProtocolSearch Index
= Retrieval Implementation

v0.2 is where OKF started to feel much more complete

For the v0 2 is where stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

sources:
  ...
generated:
  by: ...
  at: ...verified:
  ...status: stablestale_after: 2026-12-31

Where did this come from?

Who generated it?Has anyone verified it?Is it still current?Is this a draft or a stable concept?

Attested Computation is a different kind of knowledge.

For the Attested Computation is a stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the Attested Computation is a stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Here is something we know.

Here is the approved way
to establish that something is true.

So what problem does OKF actually solve?

When working through the So what problem does stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Operational checklist

The Operational checklist stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope.

Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Add a smoke test that exercises the critical path in CI with fixtures, not live paid APIs, whenever budgets allow.

Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Before promoting the stack, freeze versions, capture a golden transcript for the critical path, and confirm rollback steps. Shared environments need rate limits, tenancy checks, and a clear owner for secret rotation. Prefer boring reliability over clever one-off demos.

Batch note for 758c51495df5: keep provider keys out of the repo, set a per-session token ceiling, and store transcripts next to the eval fixtures so later model swaps stay comparable.

The hardening note 0 stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Hardening detail 0/782: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

For the hardening note 1 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish.

Hardening detail 1/782: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

When working through the hardening note 2 stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion.

Hardening detail 2/782: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

The hardening note 3 stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Hardening detail 3/782: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.

For the hardening note 4 stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline.

Hardening detail 4/782: measure wall time, error class, and token spend for this note, then decide whether to keep the change based on a fixed question set rather than anecdote.