dotjez

A copy-and-go workspace for working alongside an AI agent. Your knowledge lives as plain markdown the agent reads and keeps up to date, so you never solve the same thing twice. You don't need to be a developer, and it works for any job, not just code.

No app, no database, no cloud. It's just folders. Clone it, open your agent in the folder, and go.

The idea

Keep what you learn as one file per thing: a client, a decision, a project, a gotcha, an idea. Over time it compounds: instead of re-researching, you and the agent read what's already there and add to it. Three moves, the whole loop:

• Add — learned something reusable? Write it down, or update the file that's already there.
• Ask — starting something? Check what's already here first.
• Tidy — now and then, fix what's stale and bin what's dead. Wrong notes are worse than none.

Start

You don't set anything up by hand; your agent does it. Tell it (Claude Code, or whatever you use):

> Clone https://github.com/jezweb/dotjez into a new folder for me, then help me set up the workspace.

That's the whole install. The agent gets the folder, reads the briefs inside, then interviews you, or borrows a profile from an AI you already use (ChatGPT, Claude, Gemini), and writes your about.md.

Prefer to fetch it yourself? git clone https://github.com/jezweb/dotjez ~/workspace, then open your agent in the folder and say "help me set up this workspace."

Name the folder whatever you like; nothing inside cares. Once it holds real client data, keep it private: don't push your filled-in workspace to a public remote (the agent re-points the git remote during setup).

What the first session looks like

You    →  "help me set up this workspace"
Agent  →  interviews you (or borrows a profile from an AI you already use),
          writes about.md.

You    →  "new client signed — Acme Digital, web + email, contact Jane"
Agent  →  copies the clients/ example to clients/acme-digital.md, fills it in.

   ...you work...

Agent  →  at the end, logs a line to journal/: what you did, what's next.

Next time, the agent reads about.md and the last journal entry first, so it already knows who you are and where you left off. Every session starts ahead of the last.

For the agent

Your brief is CLAUDE.md (how to operate this workspace, loads automatically in Claude Code) and .jez/CLAUDE.md (the detail on how it's organised; it only auto-loads when you work inside .jez/, so read it yourself). Follow them; don't reinvent the conventions. In short: read .jez/about.md to learn who you serve (run .jez/playbooks/onboard.md if it's still blank), then run Ask → Add → Tidy and keep the briefs current.

What's in here

dotjez/                  ← clone this; it becomes your workspace
├── CLAUDE.md               the agent's brief (how to operate this workspace)
├── .jez/                   your knowledge hub, shared across every project
│   ├── CLAUDE.md              how it's organised (the detail)
│   ├── about.md              who you are + how you like to work (filled in first)
│   ├── clients/  contacts/  decisions/  projects/
│   ├── knowledge/  playbooks/  research/  ideas/
│   ├── journal/              the agent's running log, per session
│   ├── inbox/                somewhere to drop notes to file later
│   └── secrets/              where credentials go (and don't)
├── skills/                  reusable agent skills (use in place, or install)
└── sample-project/         what a project's own .jez looks like

Most folders ship a worked example; copy it to make a real one. A .jez works at any level: the workspace has one (knowledge across all your work), and each project gets its own (scratch for that project). The question that decides where something goes: will another project ever care about this? The agent makes new folders as your work grows; it doesn't need them up front.

Two notes for browsing it yourself: .jez is a hidden folder, so your agent always sees it but Finder won't until you press Cmd+Shift+. (or just ask the agent to open a file). And the name honours the workspace's first user; call your folder anything, but keep the .jez/ directory name itself: it's what the briefs and sibling tools key on.

The capabilities travel two ways. The workspace briefs load when your agent is opened in (or under) this folder; during setup the agent offers to add a pointer in your agent's global config so sessions started elsewhere still find the workspace. The skills install once and travel everywhere: in Claude Code, /plugin marketplace add jezweb/dotjez, then install dotjez from the /plugin menu.

Back it up

Your workspace is just files on your Mac, so back them up. dotbackup is the sibling tool: a small desktop app that backs up your .jez workspace (and any other folders) to your own Cloudflare R2, encrypted, on a schedule. You own the bucket and the keys, same spirit as dotjez.

Credit

The same idea as Andrej Karpathy's LLM wiki and the DOX pattern, kept small.