Local-first · Portable Markdown · Multi-agent

Four files. One Agent Pack.

Every modern coding agent reads the same handful of files on startup. AGENTS.md tells them what the project is. CLAUDE.md and GEMINI.md tell each tool how it behaves here. DESIGN.md tells them what the UI should look like. Prompt Organizer keeps the four together as one versioned Agent Pack, edit them side-by-side, swap tools without juggling files, ship the bundle as a folder.

Already have an AGENTS.md in a repo? Drop the four files into the import dialog and you have an Agent Pack in five seconds.

  • Plain Markdown
  • Four standard files
  • ZIP-ready
AgentPack: prompt-organizer 4 files · 12 kB · plain text
// AGENTS.md - the canonical file
# Project Overview
Prompt Organizer is a browser-based prompt
management app. Vanilla HTML/CSS/JS. Offline-first.

# Development Commands
python3 -m http.server 8000
npm run check:polish

# Operating Rules
- No bundlers in the runtime
- Edit existing files; don't create new ones
- Update CHANGELOG.md for user-visible changes

// CLAUDE.md
@AGENTS.md

## Claude Code
- Prefer the Edit tool over Write for changes
- Run npm run check:cache-bust-diff before commits

// GEMINI.md
@./AGENTS.md

## Gemini CLI
- Use the read_file tool before any edit
- Confirm dependency changes
Plain Markdown Every Agent Pack is four .md files. Open them in any editor. No proprietary container. No runtime dependency on Prompt Organizer.
Versioned & diffable Drop the folder into Git. Diff every edit to instructions, persona, and design tokens like you diff source code. History stays in the app too, undo what you broke.
Tool-agnostic AGENTS.md is the cross-tool canonical file. CLAUDE.md and GEMINI.md add tool-specific behavior on top. Add or remove a tool by editing one file.

What's inside an Agent Pack

Four files. Each one with a job.

An Agent Pack is not a longer prompt and it isn't a SKILL. It's the canonical context every agent in your project shares, the file every other prompt and SKILL inherits from. The four files below each have a specific role. Each lives in its own tab in the Agent Pack workspace, so each can be edited, diffed, and reviewed independently.

Why an Agent Pack?
  • Version your agent instructions like prompts, undo, search, history
  • Swap Claude / Gemini context without juggling files in different repos
  • Share or import as a folder with one click
  • Keep DESIGN.md tokens next to the prompts that consume them
my-project/ the four files that matter
// 1. AGENTS.md   (canonical)
Project overview, dev commands, operating rules.
Read by every AGENTS.md-aware agent (Claude
Code, Codex CLI, and others).

// 2. CLAUDE.md  (Claude Code memory)
@AGENTS.md
Then Claude-specific overrides: which tools to
prefer, which gates to run, project memory.

// 3. GEMINI.md  (Gemini CLI memory)
@./AGENTS.md
Then Gemini-specific overrides, same idea,
different tool.

// 4. DESIGN.md  (visual system)
---
colors: { primary: "#0d1524", ... }
type:   { scale: 1.2, ... }
---
Components, do's-and-don'ts, tokens.
Any agent rendering UI uses the right styles.
  1. AGENTS.md, the canonical project file.Project overview, development commands, coding rules, things every agent on the project needs to know. This is what gets imported by the other three files. Edit it once; every tool sees the change.
  2. CLAUDE.md, Claude Code's project memory.Starts with @AGENTS.md to inherit the canonical instructions, then adds Claude-specific behavior: which tools to prefer, which quality gates to run, which patterns Claude has historically gotten wrong here.
  3. GEMINI.md, Gemini CLI's project memory.The same role for the Gemini CLI. Same import pattern, different tool-specific overrides. Adding a new agent (Codex, Aider, whatever ships next quarter) is one more file in the same shape.
  4. DESIGN.md, the visual system.YAML frontmatter for tokens (colors, type scale, spacing), Markdown for components and do's-and-don'ts. Any agent rendering UI in this project uses the same buttons, the same spacing, the same colors. Designers can edit it; the AI follows.

Two starter packs. Two common jobs.

Agent Packs earn their keep the day a new agent joins the project, or the day you re-open a repo six months later and forget how it was supposed to behave.

  1. Claude Code minimal default (Solo dev).One AGENTS.md with the project overview and the three commands you run most. CLAUDE.md is a five-line file that imports AGENTS.md and tells Claude to prefer Edit over Write. Minimal GEMINI.md and DESIGN.md so you can swap tools later without scrambling.
  2. Design-system handoff (Design + Eng).AGENTS.md is light, just the project shape. DESIGN.md is heavy: every token, every component, every "don't use the old button" warning. Designers edit DESIGN.md; engineers and their agents read it.
~/my-project/ Folder export
// Exported Agent Pack - portable Markdown
my-project/
├── AGENTS.md              // canonical
├── CLAUDE.md              // @AGENTS.md + Claude rules
├── GEMINI.md              // @./AGENTS.md + Gemini rules
└── DESIGN.md              // tokens + components

// 4 files · ~12 kB · plain text
// Drop into: the root of any repo.
// Every coding agent reads its own file.
// Edit one, the others stay in sync via @-imports.

4 files · ~12 kB · belongs to you · no runtime dependency

An Agent Pack is the canonical context. Build SKILL packages on top for repeatable workflows. Manage day-to-day prompts in the Prompt Manager. Render or diff the four files with the Markdown Renderer. The whole library lives in one local-first browser app, one purchase unlocks every Pro export format.

Edit them in one place. Ship them as a folder.

The Agent Pack workspace shows all four files side-by-side with live preview. Import the four files from any existing repo to get started in seconds. Export the bundle as a folder you can commit straight into Git. Plain Markdown survives the next vendor pivot and the one after that.