Knowledge, Context and Agentic Engineering for Knowledge Work
← Knowledge, Context and Agentic Engineering

Neutral prerequisite module

Module 00 · Preparation and AI Coding Literacy

Module 00 establishes the observable technical and methodological readiness required before work is delegated to an AI harness. This tutorial connects an editor, a terminal, Python, a local website and one bounded harness task in a disposable practice folder.

Learning question: What must I be able to observe, describe, execute and verify before I delegate a task?

Last verified 25 August 2026

Work on the laptop you will use later.Keep confidential and irreplaceable files outside the practice folder.

Each path includes the checks from the preceding path.

Orientation

Choose the result you need

The tutorial has three cumulative paths. Begin with browser access and stop after the last path your work requires.

  1. Use a language model in a browser and exchange a non-sensitive test file.
  2. Connect an editor, a terminal, Python and a local website in one folder.
  3. Give one AI harness controlled, observable access to that folder.
  4. Diagnose failures from the complete command and output.

Continue from the shared foundation

The core path uses one browser, one folder, one editor and Python.

Before you begin

Prepare the minimum setup

Use the laptop you will work on later. Keep confidential and irreplaceable files outside ai-workspace.

Core path

A current browser, Visual Studio Code, a supported Python 3 and an empty ai-workspace folder.

Model access

A service such as Claude, ChatGPT, Gemini or Le Chat.

Only when selected

Node.js LTS for Pi, Git for version history and Live Server for automatic reload.

  1. Open one model service in a new conversation and ask it to reply only with browser ready.
  2. Upload a non-sensitive text file and ask the model to name the file without changing its content.

Setup assistant

The prompt requests one observable step at a time and never asks for credentials.

English promptGerman prompt

Check: a new model conversation returns browser ready and accepts a non-sensitive text file.

Schematic representation. The Explorer and terminal both refer to ai-workspace.

Step 1

Create the working folder

Visual Studio Code is an editor that keeps files, an integrated terminal and later the AI harness in one workspace.

  1. Install Visual Studio Code from its official page.
  2. Create and open an empty folder named ai-workspace.
  3. Create readme.md, write one sentence and save it.

Check: the Explorer shows ai-workspace and the saved file.

Schematic representation. The prompt states the current folder; the command output confirms it.

Step 2

Orient yourself in the terminal

A terminal runs typed commands. The shell interprets them. The working directory is the folder in which a command operates.

  1. Choose Terminal, then New Terminal.
  2. Print the current folder and list its files.
    Windows PowerShell
    Get-Location
Get-ChildItem
    macOS or Linux
    pwd
ls
  3. Use cd folder-name to enter a folder and cd .. to return.

Check: the path ends in ai-workspace, and the list contains readme.md.

Schematic representation. A saved Python file produces an observable terminal result.

Step 3

Run one Python file

Python is the runtime that executes the practice scripts. The editor extension helps Visual Studio Code find and explain the interpreter.

  1. Install a supported Python 3 from python.org.
  2. Install the Microsoft Python extension in Visual Studio Code.
  3. Check the interpreter with py --version on Windows or python3 --version on macOS and Linux.
  4. Create check.py with print("python ready"). Run py check.py or python3 check.py.

Check: the terminal prints a Python version and python ready.

Schematic representation. The generated page is served from the practice folder through localhost.

Step 4

Generate and serve a static website

HTML carries content, CSS controls presentation and JavaScript adds browser behaviour. A local server gives these files an HTTP address on your computer.

  1. Download create_website.py into ai-workspace.
  2. Run py create_website.py on Windows or python3 create_website.py on macOS and Linux.
  3. Start the server.
    Windows
    py -m http.server 8000 --directory site
    macOS or Linux
    python3 -m http.server 8000 --directory site
  4. Open http://localhost:8000/, click the counter button and stop the server with Ctrl+C.

The Python server supports local development. Live Server offers automatic reload as an optional extension.

Check: http://localhost:8000/ loads the generated page, one click changes the displayed count, and Ctrl+C returns the terminal to its prompt.

The local chain now works without an AI harness.

Milestone

The local setup works

The same folder now connects the files you edit, the commands you run and the page you inspect in the browser.

  • Explorer and terminal point to ai-workspace.
  • check.py prints python ready.
  • The script creates site/index.html, site/style.css and site/app.js.
  • The browser loads the site through localhost, and the button works.
The harness controls the model's context, tools and permission boundary.

Step 5

Choose one AI harness

A chatbot receives submitted messages and files. An AI harness also connects a model to a working folder, tools and commands, with defined approval points.

Anthropic · proprietary

Claude Code

VS Code extension and terminal interface.

Claude account, API credit or supported organisational provider.

Full introduction · Setup step

OpenAI · open CLI

Codex

VS Code extension and open-source terminal harness.

Supported ChatGPT plan or OpenAI API key.

Full introduction · Setup step

Earendil Works · MIT

Pi with a local model

Inspectable multi-provider terminal harness with an optional local-model endpoint.

Node.js, Pi and a separately configured provider or local model service.

Full introduction · Setup step

Anomaly · MIT

OpenCode

Multi-provider terminal, desktop and editor interfaces.

Access depends on the connected provider.

Documentation

Older material may name Gemini CLI. Google moved the general individual-account route to Antigravity CLI in June 2026. Consult the current Google notice before planning that setup.

Schematic representation. The first request inspects the practice folder without changing it.

Option A

Start Claude Code in Visual Studio Code

  1. Open Extensions, search for Claude Code, verify Anthropic as publisher and install it.
  2. Open its Spark icon, choose Sign in and complete authorization in the browser.
  3. Select Manual or Plan as the permission mode for the first conversation.
  4. Ask: List the files in this folder. Do not create, edit, move or delete anything, and do not run commands.
  5. Read how CLAUDE.md files provide persistent project instructions.
  6. For the terminal interface, install the standalone CLI, reopen the terminal and run claude --version and claude doctor.

Check: Claude Code names the practice files without changing them.

Continue to final check · Full Claude Code introduction · Video: Installing Claude Code · Claude Academy · Terminal quickstart · VS Code tutorial

Schematic representation. The Codex sidebar is attached to the folder and reports a read-only result.

Option B

Start Codex in Visual Studio Code

  1. Install Codex from OpenAI in Extensions.
  2. Open its icon or run Codex: Open Codex Sidebar from the Command Palette.
  3. Choose Sign in with ChatGPT and complete browser sign-in.
  4. Start a local chat and send the same read-only file-list request.
  5. For terminal automation, follow the operating-system tab on the official CLI page.

Check: the sidebar reports the files in ai-workspace without a change.

Continue to final check · Official IDE tutorial · OpenAI Academy · Codex Bootcamp 101

Schematic representation. Pi runs in the terminal and exposes its provider and working folder.

Option C

Start Pi in the integrated terminal

  1. Install Node.js LTS and restart Visual Studio Code.
  2. Run node --version and npm --version.
  3. Install Pi.
    npm install -g --ignore-scripts @earendil-works/pi-coding-agent
  4. Run pi, enter /login and select a supported provider.
  5. Send the read-only file-list request. Stay inside the disposable folder while file and shell tools are available.

Check: Pi shows its provider and folder and lists the practice files without changing them.

Continue to final check · Official Pi documentation · Video: What I Use Instead · Further videos in the teaching playlist

Git records local versions; GitHub is needed only for online publication or collaboration.

Optional

Add Git and GitHub

Git records versions inside the local folder. GitHub stores a repository online when you choose to publish or collaborate.

  1. Install Git and restart Visual Studio Code.
  2. Run git --version, git init and git status in ai-workspace.
  3. Create a GitHub account when you want to publish or collaborate. Follow the official setup guide.

Check: git status reports the current branch and practice files without a not a git repository error.

Every completed path ends with visible evidence on the same laptop.

Final check

Verify the complete setup

  1. The browser service answers and accepts a non-sensitive file.
  2. Editor, terminal and files point to ai-workspace.
  3. Python runs both scripts and generates the site.
  4. localhost:8000 loads the styled, interactive page.
  5. The selected harness passes the read-only file-list test.
A useful diagnosis preserves the complete command, context and output.

If a step fails

Diagnose one condition at a time

  1. Copy the complete command and error as text.
  2. Record the operating system, shell and working directory.
  3. Return to the preceding successful check and consult the official tool documentation.
  4. Ask a model for one reversible diagnostic command, change one condition and repeat the check.