AI Workspace

This is my current setup for any project workspace.

The biggest shift is that I moved projects from a “project” perspective to an “organisation” perspective.

I spent a lot of time finding the right structure, and it’s been months since anything changed, so I guess it’s a stable version for me.

File System

This is the high-level view, and we go deeper into each folder and why they are there.

I use the word “workspace” to refer to an organisation.

<root_organisation>/
├── .git/
├── apps/
├── config/
│   └── os.yaml
├── context/
│   ├── system/
│   │   └── AGENTS.md
│   └── user/
│       └── AGENTS.md
├── general/
│   └── wiki/
├── managed_organsations/
├── prompts/
├── skills/
│   ├── custom/
│   ├── private/
│   ├── public/
│   └── skill_index.py
├── AGENTS.md
└── CLAUDE.md

Apps

All your apps live there; each subfolder is an app (a Git repo, not a Git submodule).

Config

Everything related to configurations. I’ll get back to it.

Context

The context/system/AGENTS.md is shared across all my workspaces, and context/user/AGENTS.md is specific to this workspace.

General

It mainly includes the wiki with a classic LLM-wiki pattern.

It stores all the information regarding the organisation.

Prompts

I still copy-paste prompts sometimes, mainly to run a skill, but still, it’s a copy-paste thing, so it’s good to store them.

Skills

• custom: mine, but can be public.
• private: mine, but specific to my organisation.
• public: skills I downloaded from the outside.
• skill_index.py: a simple Python script that lists all skills in the skills/ folder, run at the start of the session.

AGENTS.md

Shared across all my workspaces

# Workspace context

Before handling the first request in a session, read and follow these files in order:

1. `context/system/AGENTS.md`
2. `context/user/AGENTS.md`

`context/system/AGENTS.md` opens with a Precedence section. It resolves every conflict between the user, the host harness, a subtree `AGENTS.md`, this shared context, and skills.

# Updates

The instructions in `context/` and `skills/` might have changed during your execution.
You should re-read them if you have an outdated version in your current context.

CLAUDE.md

Just read AGENTS.md

@AGENTS.md

Conclusion

I’ll explain more details about each folder, how I use NanoClaw and Telegram to connect everything.