Get the kit

Finish projects with Claude.

Agentcase is a folder of plain text files for people who use Claude as a chat and want to run real projects with it. Each project keeps its memory there, so a new session can start where the last one stopped.

29 EUR, paid once. Built for Claude Code, and other agents that read AGENTS.md can use the files too.

00_START_HERE.md
  1. Unzip this folder.
  2. Open it in the Code tab.
  3. Say: install
agentcase/
_CONTEXT/
_INPUTS/
_OUTPUTS/
_PROJECTS/
_SYSTEMS/
_ARCHIVE/
FIRST_PROJECT.md

From an idea to one finished file.

The agent holds his briefcase open and lifts out a sheet titled STATUS.md

Point at him to see his pixels.Touch him to see his pixels.

When a project lives in a chat.

What people say, and what the kit does about it.

The agent points to the side, his briefcase in the other hand
In a chatWith Agentcase, in the folder agentcase/
“I change chat and the history of my project is gone.”Each project keeps its state in files, so a new session can start where the last one stopped.
“My drafts and files are spread over chats and downloads.”One folder on your computer, with a place for inputs, outputs and each project.
“It starts building before it knows what I want.”Scope comes first. You agree what done looks like before the work starts.
“It says done, and I find out later what is missing.”Prove comes last. The agent checks the result and says what is still open.
“I hit the usage limit in the middle of the afternoon.”Subagents on smaller models take the long reading, so a session uses less of your limit.
“Every session behaves a bit differently.”Every session starts in the same folder under the same rules.

The protocol shapes the work.

Without rules of its own, an agent has two things to go on: its training and what you type. So every session comes out different. With Agentcase, every session starts in the same folder and reads the same rules, and you stop repeating them in each prompt.

Four agents of different shapes, each formed by a different prompt be brief!! you are an expert make it pop just do it
Prompt only. Each session is shaped by what you happened to write.
Four identical agents shaped like puzzle pieces that fit together, under one set of rules AGENTS.md · S.H.I.P.
One protocol. Every session starts from the same rules and fits the same work.
The agent raises an open hand to ask first, his briefcase in the other hand

Every project, four steps.

The protocol is called S.H.I.P., one order of work for any project, client and professional work included. A quick question stays a quick question.

  1. Scope

    Agree what done looks like before the work starts.

  2. Hook up

    Check that files, tools and accounts work before relying on them.

  3. Implement

    Do the work and keep the decisions on record.

  4. Prove

    Check the result and say what is still open.

What's in the case.

44 files of plain text and a few small scripts. These nine do most of the work.

  1. 00_START_HERE.md

    Where to begin, written for someone who has never done this.

  2. INSTALL.md

    The steps your agent follows to set up the folder. They tell it to wait for your yes before it writes.

  3. The protocol skill

    The heart of the kit. A skill is a method your agent loads when a task calls for it. This one sets up each project and runs S.H.I.P.

  4. Project memory

    Each project keeps its state in short files. A new session reads where things stand and opens the full history only when it needs it.

  5. AGENTS.md

    The rulebook of your workspace.

  6. FIRST_PROJECT.md

    A guided first project, from an idea to one finished file.

  7. BEST_PRACTICES.md

    How to brief an agent and check its work, and the session habits that make your limit last.

  8. Model routing

    Subagents on smaller models do the long reading and hand back a short report.

  9. WRITE_YOUR_OWN_SKILL.md

    How to turn a method you keep explaining into a skill of your own.

Pay once, keep the files.

The core kit works on its own. Each add-on is a separate set of files, installed the same way, now or later.

order

Core kit and add-ons
  • is the core kit with Intel and Remote, 49 EUR instead of 58.
  • is all four, 65 EUR instead of 77.

Total before VAT29 EUR

Get the kit Payment through Stripe.

  • No subscription. The files stay on your computer.
  • Download link by email right after payment.
  • You can withdraw within 14 days, until you start the download.

Make it yours and keep getting updates.

No two people work the same way. Your agent adapts the kit to you, and you keep changing it from there. Updates are built for your copy.

Your changes come first
The agent shows you what an update changes and touches only the files you have not edited. In the others, you choose what to bring in.
Fixes are free
If something in your version does not work as described, the fix comes as a free patch of that version.
New versions, when they matter
A new version comes out when there is something significant to add. Owners of an earlier version can upgrade at a reduced price, 7 EUR today. The release notes say what is new, so you can judge whether it is worth it for the way you work.

I build Agentcase and I use it. My own projects, client work included, run on this setup every day. About me

The add-ons, in detail.

add-on intel/

Intel

12 EUR

A folder for what you know.

Projects end. What you learn from them should not. Intel is a second folder next to your projects, for sources, people, decisions and ideas. Your agent reads it when a project needs it, and files new knowledge there when you ask.

Two folders side by side: agentcase, where you do the work, and intel, where you keep what you know. The workspace reads intel, and files to it when you ask. agentcase/what you do intel/what you know reads fileswhen you ask

What changes

  • “I explain the same background to every new project.”

    The agent looks up what you already know in Intel before it starts.

  • “My notes only grow, and I stopped trusting them.”

    The rules tell the agent to update, merge or remove a note before it adds a new one, and to bring contradictions to you.

  • “Links, clippings and voice memos pile up everywhere.”

    One inbox folder. When you ask, each item becomes a note, a kept source, or goes.

What you get

  • The Intel folder, set up next to your workspace, with a map of your notes that the agent keeps current.
  • Three skills: one reads Intel back, one files new knowledge, one audits and tidies the whole folder and changes only what you approve.
  • A guide to read it in Obsidian, free, on your computer and your phone.

Good to know

  • Needs the core kit. Obsidian is optional. Intel is plain text, and any editor opens it.
  • Start small. A few notes you reread are worth more than an import you never open.

add-on remote/

Remote

17 EUR

Your projects on your phone.

Your workspace stays on your computer, and you keep working on it from the Claude app on your phone. On the sofa, on a train, with the computer shut in a bag. Same projects, same rules, same memory as at your desk.

From your phone you reach your workspace on the computer when it is on, and a private copy on GitHub when it is off. The work done in the copy comes home. your phone computer on computer off agentcase/ GitHub copy comes home

What changes

  • “My projects are stuck on the computer at home.”

    From the Code tab of the Claude app on your phone you open sessions in your own workspace.

  • “When the computer is off, nothing moves.”

    Cloud sessions work on a private copy on GitHub, and what they do comes home when you are back.

  • “A chat on my phone knows nothing about my projects.”

    Every session, on the phone or at the desk, reads the same rulebook and the same project memory.

What you get

  • The setup, installed by your agent step by step. The steps tell it to wait for your yes before anything touches GitHub.
  • Two ways in: through your computer when it is on, through cloud sessions when it is off.
  • github-sync, a skill for the two habits. Say “commit and push” before you leave and “pull the latest” when you are back. Its steps include a check for keys, passwords and heavy files before each commit.
  • git-housekeeping, a skill that tells you when the copy has grown heavy and is written to clean its history only after your yes.
  • With Intel, cloud sessions can read what you know too.

Good to know

  • Images, videos, PDFs and your inputs stay on your computer. Cloud sessions work on text.
  • Nothing extra to pay: private GitHub repositories are free, and sessions count toward your usual Claude limit.
  • Needs the core kit, a paid Claude plan, the Claude app on your phone and a free GitHub account.

add-on team/

Team

19 EUR

One folder for a small team.

Bring one or more people into your workspace. Each works on the same projects with their own Claude, and nothing they change reaches the main copy until you approve it.

Your work goes straight into the main copy. A teammate's work arrives as a proposal and joins the main copy only after you approve it. teammate you proposal you approve your work main copy

What changes

  • “We send files back and forth and lose the latest one.”

    One shared workspace. Everyone works on the same projects and the same project memory.

  • “I can't let someone else's agent loose on my files.”

    A teammate's work waits as a proposal. Your agent shows you what it changes, and only what you approve goes in.

  • “Bringing someone in means explaining everything again.”

    Teammates start in the same folder, under the same rules, with a one-page guide you send them.

What you get

  • The setup of a shared workspace in a free GitHub organization. Your own way of working does not change.
  • A review step: your agent lists each teammate's proposal, what it changes and what it could not finish, and merges what you approve.
  • A one-page guide for each teammate, and a short procedure to add one later.

Good to know

  • Every teammate sees the whole workspace. Keep a private project in a separate one.
  • Intel, if you have it, is shared only if you choose, and read-only for them.
  • Needs Remote. Each person needs a paid Claude plan and a free GitHub account. One licence covers one shared workspace and its teammates.

Start with the free setup.

An extract of the kit: two of its pieces, one for safety and one for your usage limit.

  • Workspace rules that tell your agent to ask before it deletes, moves or pushes, and settings that make Claude Code ask before those commands.
  • Subagents on smaller models for the long reading.
  • A short guide to the habits that make your limit last: sessions, compacting, connectors, models and effort.

The paid kit includes this setup as it is, so you never need both. It adds the protocol, the project memory and the first project.

You confirm by email. One message per update, unsubscribe anytime. Privacy policy.

The agent holds out a sheet of paper, his briefcase in the other hand

Before you buy.

The agent walks away with his briefcase

The kit

Is Agentcase an app?

No. It is a folder of files that tells your agent how to work. You still need Claude Code, or another agent, to run it.

Is this the same as Claude Duet or Cowork?

No. Duet and Cowork are Claude apps that do work for you, Duet mostly in your mail, calendar and connected apps. Agentcase is a folder on your computer with your projects, their memory and the rules your agent follows, in files you own. You use it from the Code tab, and it works fine next to them.

Does it work without Claude Code?

Mostly. I build and test the kit with Claude Code, in my experience the strongest agent for this work today. The folder, the rulebook, the project memory and the S.H.I.P. protocol are plain files that any agent reading AGENTS.md can follow, such as Codex, Cursor or OpenCode. The subagents of the model routing and the Remote and Team add-ons are made for Claude Code.

I already have a workflow. Is it for me?

Starting from zero, you gain the most. If you already have a workflow of your own, the kit is plain text, so take the pieces that help and leave the rest.

Can the agent still make mistakes?

Yes. Every AI agent can make mistakes of any kind, and how it behaves depends on the model and on the app that runs it. The kit's rules tell it to show what will happen and to wait for a clear yes before anything that cannot be taken back. That lowers the risk. It does not remove it, so keep backups and read what the agent proposes before you approve it.

The install

Do I need to know how to code?

No. You unzip a folder, open it in the Code tab of the Claude desktop app and say “install”. The install steps tell Claude to explain each step and to wait for your yes.

What lands on my computer?

One folder of text files and a few small scripts, where you choose. Nothing runs in the background. To remove it, delete the folder.

Which Claude plan do I need?

A paid plan, such as Pro, Max or Team. The whole kit, core and add-ons, runs in Claude Code, and the free plan does not include it.

Do I need GitHub?

Only for Remote and Team. A free account is enough.

Is it only in English?

The files are in English, and your agent still talks to you in your language. I wrote the protocol in English and work with my agent in Italian.

Buying

Is it a subscription?

No. You pay once for the version you buy. Fixes for that version are free, and a new version is an optional upgrade at a reduced price.

Can I add an add-on later?

Yes. The link is in your order email, and the agent adds it to the workspace you already have.

Can I use it for client work?

Yes. The licence covers personal and professional use, work for clients included. It is granted to one person. To share a workspace with other people, add Team.

What if it is not for me?

The free setup lets you try a piece on your computer first. After paying you can withdraw for a full refund within 14 days, as long as you have not started the download.