5-minute twin demo

This is the shortest public path for answering: "Can I actually see how this works?"

Yes. This repo is a blueprint and demo workspace, not a hosted digital-twin runtime. The point of this page is to show the loop with files you can inspect:

knowledge files -> capability -> prompt -> durable output -> write-back learning

You only need this repo and an AI IDE or coding agent that can read and write local files.

What you will use

Part	File	What to check
Workspace rules	playground/AGENTS.md	The sample workspace tells the agent to retrieve context first, write files, and save reusable rules.
Knowledge index	playground/wiki/_index.md	The index points to existing articles, summaries, and learning notes.
Raw input	playground/raw/thoughts/2026-04-23-why-most-ai-feels-generic.md	This is the rough source material for the run.
Prior output	playground/Blog/Published/2026-04-10-ai-workflow-is-the-product.md	The agent should compare against this before drafting, so it does not repeat the same point.
Reusable context	playground/wiki/summaries/2026-04-12-agent-products.md	This gives reusable language like "answer machine" and "asset-building machine."
Existing learning	playground/wiki/outputs/agent-learnings/2026-04-15-writing-rules.md	This shows how durable writing preferences are stored.
Capability	capabilities/content-creation.md	This is the workflow contract for turning raw notes into a publishable article.
Starter prompt	playground/FIRST_PROMPT.md	This is the ready-made prompt for the run.

Minute 1: inspect the knowledge

Open playground/wiki/_index.md. Notice that it is not a giant database or hidden memory service. It is a small map of files the agent should read before acting.

Then skim the raw thought file and the existing writing rule:

• playground/raw/thoughts/2026-04-23-why-most-ai-feels-generic.md
• playground/wiki/outputs/agent-learnings/2026-04-15-writing-rules.md

The important idea: the agent should inherit working context from files, not from a one-off chat history.

Minute 2: choose a capability

For this demo, choose capabilities/content-creation.md.

That capability tells the agent to:

• check existing work before creating a new article
• extract the strongest throughline instead of mechanically rewriting a transcript
• keep raw material separate from generated output
• write reusable lessons back into the workspace when the run reveals a stable rule

This is the "skill" part of the twin. It is a plain Markdown workflow file that a builder can inspect, edit, or replace.

Minutes 3-4: run the prompt

Use an AI IDE or coding agent with this repository open. Paste this prompt:

Use only the public sample files in this repository.

Treat playground/ as the sample workspace.
First read:
- playground/AGENTS.md
- playground/wiki/_index.md
- playground/Blog/Published/
- playground/wiki/summaries/
- playground/wiki/outputs/agent-learnings/

Then use capabilities/content-creation.md to turn
playground/raw/thoughts/2026-04-23-why-most-ai-feels-generic.md
into a publishable blog draft.

Requirements:
- do not mechanically rewrite the raw transcript
- compare against the existing published article before drafting
- write the draft under playground/Blog/Published/
- write any reusable writing rule under playground/wiki/outputs/agent-learnings/
- if you only return a chat answer and do not write files, the run failed

If you want the shortest version, open playground/FIRST_PROMPT.md and run that prompt.

Minute 5: check the write-back

After the run, inspect the workspace. A successful run should leave durable files, not only a chat response.

Check	Expected result
Draft output	A new Markdown draft appears under playground/Blog/Published/.
Learning output	A new learning note appears under playground/wiki/outputs/agent-learnings/.
Retrieval evidence	The agent mentions or uses the wiki index, prior article, summary, and writing rules.
Capability evidence	The agent follows the content-creation workflow instead of improvising a generic answer.

The exact title of the generated draft can vary. The important test is whether the agent completed the full loop:

read existing files
choose the content-creation workflow
create a durable article draft
save a reusable rule for the next run

What this proves

This demo does not prove that the repo is a complete hosted product. It proves something smaller and more useful for a builder:

• the knowledge layer can be simple files
• a capability can be a readable workflow contract
• a prompt can route work through that contract
• the result can become durable workspace state
• the next run can start from the saved state instead of starting over

For a stricter reviewer path, continue to the proof chain or the full Steven workflow demo.