nanoclip

NanoClip

A Claude Code plugin that orchestrates AI organizations. Define agents with rich profiles, assign tasks, and run them as coordinated agent teams — all from within Claude Code.

Inspired by PaperClip.

NanoClip models an organization as a directory on disk. Agents are JSON files. Tasks are JSON files. The directory structure IS the database. No server, no UI, no external dependencies beyond Claude Code and jq.

Installation

Prerequisites

• Claude Code (1.0.33+)
• jq

Install the plugin

claude plugin marketplace add https://github.com/bryonjacob/nanoclip
claude plugin install nanoclip@nanoclip

That's it — all /nanoclip:* commands are now available in any Claude Code session.

Getting Started

1. Set up your organization

Create a new directory and initialize NanoClip inside it:

mkdir my-org && cd my-org && git init

Then from Claude Code:

/nanoclip:init --name "My Team" --goal "Build a web app"

2. Add some agents

/nanoclip:agent add --name "Lead" --title "Tech Lead" \
  --capabilities "Architecture, planning" --management

/nanoclip:agent add --name "Alice" --title "Backend Engineer" \
  --capabilities "Python, APIs" --reports-to lead

/nanoclip:agent add --name "Bob" --title "Frontend Engineer" \
  --capabilities "React, CSS" --reports-to lead

3. Create a task and run

/nanoclip:task create --title "Build the landing page" \
  --assignee lead --priority high

/nanoclip:run

That's it. The lead agent will break down the task, delegate to Alice and Bob, and they'll coordinate to get it done.

Use /nanoclip:dashboard at any time to see progress.

Commands

Command	Description
/nanoclip:init	Initialize a NanoClip organization in the current directory
/nanoclip:agent	Manage agents — add, list, show, edit
/nanoclip:task	Manage tasks — create, list, show, update, comment
/nanoclip:orgchart	Display the reporting hierarchy as a tree
/nanoclip:dashboard	Full dashboard: org chart, tasks, progress, recent runs
/nanoclip:status	Quick status overview
/nanoclip:run	Launch agents as a coordinated team to work through tasks

Organization Structure

my-org/
├── .nanoclip/
│   ├── org.json              # Organization metadata (name, goal)
│   ├── agents/
│   │   ├── director.json     # Agent definitions with rich profiles
│   │   ├── alice.json
│   │   └── bob.json
│   ├── tasks/
│   │   ├── 001.json          # Tasks with full history & comments
│   │   └── 002.json
│   ├── runs/                 # Run logs
│   └── counter.txt           # Auto-incrementing task ID
└── workspace/                # Where agents do actual work
    └── .memory/              # Agent memory persists across cycles

Agent Schema

Agents have rich profiles that shape how they work:

{
  "id": "alice",
  "displayName": "Alice",
  "title": "Backend Engineer",
  "model": "claude-sonnet-4-6",
  "capabilities": ["Python", "APIs", "databases"],
  "expertise": "5 years backend, distributed systems",
  "jobDescription": "Owns backend implementation...",
  "reportsTo": "director",
  "management": false,
  "status": "idle",
  "createdAt": "2024-01-15T10:30:00Z"
}

• title — free-form, any role you want (not limited to CEO/engineer)
• capabilities — array of skills, used for task assignment
• expertise — background and experience
• jobDescription — detailed responsibilities
• management — enables delegation and hiring abilities
• reportsTo — builds the org hierarchy

How It Works

Management agents

Agents flagged as management: true (or any agent with reports) get the delegation skill. They break down large tasks into subtasks, assign work to their reports, review progress, unblock stuck agents, and can hire new agents when capabilities are missing.

Individual contributors

Non-management agents do the actual implementation work: writing code, creating files, running tests in the workspace/ directory.

Agent teams

/nanoclip:run uses Claude Code's agent teams feature for parallel execution. The top-level manager becomes the team lead, other agents become teammates. They coordinate through the shared task list and messaging, while persisting all state changes to .nanoclip/.

If agent teams aren't available, it falls back to running agents sequentially.

Memory

Agents save working context to workspace/.memory/<agent-id>/ before finishing a cycle. This is automatically loaded into their next cycle, providing continuity across runs.

Acknowledgments

NanoClip is inspired by PaperClip, reimagined as a Claude Code plugin.

License

MIT