Plain-English Tour

This page is for a first-time visitor who wants to know what this repo means without installing anything.

digital-twin is a public blueprint and demo workspace. It is not a hosted bot, not a private clone of Steven, and not a claim that a full product runtime already exists.

The core ideas

Personal Agent OS

A Personal Agent OS is a folder structure plus rules that tell an agent how to reuse your notes, writing habits, and work patterns.

In this repo, that means files like:

• playground/AGENTS.md, which tells the agent how the demo workspace is organized.
• playground/wiki/_index.md, which points to the context the agent should read first.
• capabilities/content-creation.md, which describes the steps for turning rough notes into a blog draft.

The important part is not the label. The important part is that the working habits are written down where an agent can inspect them.

Digital twin

Here, a digital twin is not a clone of a person and not a chatbot running on a server.

It is a file-first workspace that an agent can read and update. The "twin" part comes from the workspace carrying your working memory: notes, prior outputs, folder rules, style rules, and lessons from past runs.

The demo version lives under playground/. A visitor can inspect the raw input in playground/raw/thoughts/, the prior writing in playground/Blog/Published/, and the reusable lessons in playground/wiki/outputs/agent-learnings/.

Capability routing

Capability routing means choosing the right workflow file for the task instead of putting every instruction into one giant prompt.

For example, a request to turn a rough thought into a blog post should use capabilities/content-creation.md. That file says what to read, how to avoid duplicates, where to write the draft, and when to save a lesson for next time.

If the task were wiki cleanup, resume work, or codebase research, it should route to a different capability file.

Write-back

Write-back means the useful result is saved to files, not only returned in chat.

In the Steven workflow demo, a successful run should leave:

• a draft under playground/Blog/Published/
• a reusable lesson under playground/wiki/outputs/agent-learnings/

If the agent only replies with text in the chat window, the demo did not prove the operating loop.

Learning loop

A learning loop means reviewing what changed during the task and turning the useful part into a rule the next run can reuse.

The rule might be a writing preference, a folder convention, a duplicate-check rule, or a mistake to skip next time. In this repo, those lessons are written as files under playground/wiki/outputs/agent-learnings/.

60-second tour

Open these files in order. You do not need to install anything to understand the public proof.

1. THESIS.md explains why the repo exists: making a person's working method readable by an agent.
2. WORKFLOW.md shows the loop: understand intent, read context, choose a capability, write output, save a lesson.
3. capabilities/content-creation.md is a concrete workflow file for turning raw notes into a draft.
4. playground/FIRST_PROMPT.md is the test prompt for the demo workspace.
5. docs/demo/steven-workflow.md walks through the demo and its success checklist.
6. docs/demo/proof-chain.md maps each public claim to a file a reviewer can inspect.

What to check

When you inspect the repo, look for this chain:

Raw input
  -> existing notes and prior outputs
  -> chosen capability file
  -> saved draft or report
  -> saved lesson for the next run

That is the whole idea in plain English: keep the working system in files, make the agent read the files before acting, and save the useful result back into the same workspace.