Foundation · About 25 minutes with practice

Choose a harness. Keep your agency.

Understand what Codex, Claude Code, Hermes and OpenClaw surround a model with—and how to start deliberately.

Make this yours

A host choice, a verified first session and a portable practice brief.

Start with your own answer ↓
Speculative Academy artwork: the memory of living matter.
Imagined world · The memory of living matter

A moment before the explanation

Which responsibilities belong to a harness rather than to the model itself?

Give yourself a first answer, even if it is incomplete. Keep it in mind as you read—or write it in your field note.

01

A model is one part of the environment

The harness supplies the working environment around a model: context loading, tool access, permissions, execution, session state and feedback. Changing the harness can change behavior even when the model stays the same. Choose it by the work you need to perform and the boundaries you need to inspect.

Codex and Claude Code offer repository-oriented workflows. Hermes documents persistent memory and reusable skills. OpenClaw documents a gateway and connected assistant workflows. These are starting points for investigation, not a ranking. Read the current documentation for your operating system and account.

02

Install once, then prove a small round trip

Use the provider's official installation page below. Read platform prerequisites and what an installer changes before running it. After setup, verify the executable with its help or version command, then open a scratch project containing only public or synthetic material.

Ask the agent to explain one file before permitting changes. Next, request a small reversible edit, inspect the diff and run a relevant check. Record the installed version and whether the requested permission boundary was honored. Keep credentials in the host's credential mechanism; never put them inside a practice brief.

03

Port the method, then test the adapter

A skill packages a repeatable procedure and supporting resources. MCP defines a way for clients and servers to expose tools and context. A portable instruction file, an MCP connection and a running model are separate pieces. Support for one does not establish support for every feature of another.

Use the Academy's existing installation guide to bring a selected practice into your own host. When trying a second host, compare input handling, tool permissions, output shape and failure behavior. A copied folder is not proof of equivalent execution. Keep the learning artifact portable even when host configuration differs.

A worked example · Illustrative

Your first-session record

Host and version:
Official installation page:
Scratch project:
Read-only task and observed result:
Small edit and inspected diff:
Check run and result:
Permission boundary tested:
Unresolved setup issue:

Put the idea to work

Your practice

  1. Choose a host using your task, operating system and desired execution boundary.
  2. Follow its official setup and save the version in your record.
  3. Complete a read → edit → inspect → check round trip in a scratch project.
  4. Open the Academy toolkit installer and select the same practice.

Make the idea available when you need it

Your field note

Keep your first answer, make an artifact and try the idea somewhere new. Save only what you want stored on this device.

Check one distinction

A skill folder loads successfully in a second host. What does that establish?

Take this practice into your own agent.

Copy a coaching brief, then choose your host in the Academy installer.

Choose a host ↗

Your text stays in this page until you save or export. Local saving can be unavailable in restricted browsers. Curriculum 2026-09-15.1. Import, export and browse your notebook ↗

Follow the evidence further

Primary sources & deeper study

Selected 15 September 2026. Provider docs govern current setup; older courses and engineering articles provide foundations.

OpenAI · Official documentation

OpenAI Codex

CLI installation and repository workflow. Check the current platform-specific instructions.

Anthropic · Official documentation

Claude Code quickstart

Host setup and first repository session.

Nous Research · Official documentation

Hermes Agent documentation

Setup, tools, memory and reusable skills; distinguish these from changing model weights.

OpenClaw · Official documentation

OpenClaw getting started

Gateway onboarding and supported host setup. Start locally before connecting messaging accounts.

Agent Skills · Open format

Agent Skills overview

Portable folders of instructions and resources. Host behavior still needs testing.

MCP · Open standard

Model Context Protocol introduction

The client–server boundary for tools and resources. Protocol support does not grant permission.

Take the method into your work

Choose your host and install the toolkit
Another question to follow

Build a team that can disagree usefully