Toolshed: one repo for every AI tool you use

How it works

You change it once. Every tool picks it up.

A new tool, skill or rule goes through a pull request into the Stack Map and your profiles. The Chief of Staff syncs it to every AI tool, writes your folder files, checks health and opens fixes for you to approve.

A change goes through pull request review into the Stack Map and profiles; the Chief of Staff syncs skills, writes level folders, checks health and opens fix pull requests; every AI tool stays current.

The pieces

Five parts, one repo.

Inventory

Stack Map

Every agent, skill, connector, MCP server, plugin and API, each marked Installed or Not-Installed, with how it connects and what it is for.

Admin settings

Profiles

Global, organization, department, individual. Each level lists its tools and inherits the level above.

Folders

Level files

README, INDEX, SKILLS, CLAUDE, AGENTS and GEMINI in every level's folder, each saying only what that level adds.

Agent

Chief of Staff

One health report: broken keys, stale tools, drifted files, old pull requests, a stopped sync. Fixes arrive as pull requests.

Sync

Every AI tool

Skills and MCP servers install into Claude Code, Codex, Gemini CLI, Cursor, Copilot and Hermes every morning.

Suggested folders

Lean sessions start with lean folders.

Mirror your profiles in your folders. AI tools read instruction files from the folder you work in and the ones above it, so nothing is ever repeated.

~/                         Global  home-file rules + global SKILLS.md
└─ Work/
   └─ Acme Builders/       Organization
      ├─ README.md         what this folder is
      ├─ INDEX.md          level above, levels below, files here
      ├─ SKILLS.md         tools, Installed / Not-Installed
      ├─ CLAUDE.md         this level's rules, then @SKILLS.md
      ├─ AGENTS.md         same rules for Codex, Cursor, Copilot
      ├─ GEMINI.md         same rules for Gemini
      └─ sales/            Department  (same six files)
         └─ people/jane/   Individual  + writing style, outcome goal
~/code/<repo>/             repos keep their own CLAUDE.md

Nothing repeatedA department file never copies the company rules; tools load the folders above.
Only your branch loadsWorking in sales loads Global, company and sales. Not every department.
Small files, better follow-throughEach file says only what its level adds.
Your edits are safeHand-edited generated files are kept and reported, never overwritten.
Tools know what existsSKILLS.md tells the AI what is connected and what is still planned.

Where it works

Terminal, IDE, desktop and web.

Checked against each vendor's documentation on 27 September 2026.

Surface	Type	Reads	Chief of Staff
Claude Code	Terminal, VS Code, JetBrains	CLAUDE.md here and above, skills	Full
Codex CLI and IDE	Terminal, IDE	AGENTS.md from the repo root down, skills	Full
Gemini CLI	Terminal	GEMINI.md here and above, skills	Full
Cursor	IDE	AGENTS.md, shared skill folders	Full
VS Code with Copilot	IDE	AGENTS.md and CLAUDE.md (settings on), skills	Full
Claude desktop, Cowork	Desktop	Folder instructions, account skills	Full with a folder
Hermes Agent	Harness: terminal, chat apps, cron	AGENTS.md here and up to 5 folders above; SOUL.md global; shared skills folder	Full
Gemini Code Assist	IDE	Markdown context files	Partial
claude.ai, ChatGPT, Gemini app	Web, mobile	Account skills; no local folders	Report only

Get started

About ten minutes.

macOS or Linux, Python 3 and git. The starter kit is shared by invitation; ask for read access, then make your own private copy.

Make your copy

Use this template on GitHub, keep it private, and clone it to ~/code/toolshed.

Run setup

bash scripts/setup.sh asks a few questions and asks before writing anything.

Save it

Commit and push. The check runs on every pull request.

Ask your AI

"Run the Chief of Staff." You get one report and one next action.