agentleFS
memi-designmemi-design/design-skills/skills/multi-agent/SKILL.md

multi-agent

Skill7 starsChanged 4 months agoMITView on GitHub

Orchestrate multiple Claude instances on Figma canvas with box widgets, coordinated handoffs, error recovery

9 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every agent creates a status box on the Figma canvas for human visibility.

    line 37
  2. Distinctive:

    When an agent encounters an error:

    line 127
  3. Distinctive:

    Not collapsing boxes on completion (visual clutter)

    line 156

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/multi-agent — Parallel Agent Workflows in Figma

Orchestrate multiple Claude instances on the Figma canvas with full transparency via box widgets, coordinated handoffs, and error recovery. Requires /figma-use.

Freedom Level: High

Each agent operates autonomously within its scope. The orchestrator coordinates handoffs and resolves conflicts.

Architecture

Port Allocation

Port 9223 → Primary agent (orchestrator)
Port 9224 → Token engineer
Port 9225 → Component architect
Port 9226 → Layout designer
Port 9227 → Code generator
Port 9228-9232 → Additional specialists

The Mémoire plugin auto-discovers all instances via port scanning (9223-9232) every 5 seconds.

Instance Identification

Each agent instance should use a distinct bridge session name:

memi connect --name "Token Agent"
memi connect --name "Component Agent"
memi connect --name "Layout Agent"

Box Widget Protocol

Every agent creates a status box on the Figma canvas for human visibility.

Creating a Box

let agentSection = figma.currentPage.findOne(
  n => n.type === 'SECTION' && n.name === 'Active Agents'
);
if (!agentSection) {
  agentSection = figma.createSection();
  agentSection.name = 'Active Agents';
  agentSection.x = -400;
  agentSection.y = 0;
}

const box = figma.createFrame();
box.name = `[${role}] ${task}`;
box.layoutMode = 'VERTICAL';
box.primaryAxisSizingMode = 'AUTO';
box.counterAxisSizingMode = 'FIXED';
box.resize(300, 1);
box.paddingLeft = box.paddingRight = 12;
box.paddingTop = box.paddingBottom = 10;
box.itemSpacing = 4;
box.cornerRadius = 8;
box.fills = [{ type: 'SOLID', color: { r: 0.06, g: 0.06, b: 0.12 }, opacity: 0.95 }];
box.strokes = [{ type: 'SOLID', color: statusBorderColor }];
box.strokeWeight = 1.5;
agentSection.appendChild(box);

Status Colors

const statusColors = {
  idle:  { r: 0.3, g: 0.3, b: 0.4 },   // Gray-blue
  busy:  { r: 0.96, g: 0.62, b: 0.04 }, // Amber
  error: { r: 0.94, g: 0.27, b: 0.27 }, // Red
  done:  { r: 0.06, g: 0.73, b: 0.51 }, // Green
};

Updating & Collapsing

// Update status
myBox.name = `[${role}] ${newTask}`;
myBox.strokes = [{ type: 'SOLID', color: statusColors[status] }];

// Collapse on completion
myBox.resize(300, 28);
myBox.name = `✓ [${role}] Complete`;
myBox.fills = [{ type: 'SOLID', color: { r: 0.04, g: 0.45, b: 0.34 }, opacity: 0.3 }];

Coordination Patterns

Pattern 1: Pipeline (Sequential Handoff)

Token Engineer → Component Architect → Layout Designer → Code Generator
     ↓                   ↓                    ↓                ↓
  Variables         Components            Pages           React code

Each agent waits for the previous to broadcast completion.

Pattern 2: Parallel Atoms (Fan-Out)

Orchestrator
├── Agent A: builds Button, Badge, Avatar, Icon
├── Agent B: builds Input, Label, Select, Checkbox
├── Agent C: builds Card, Separator, Tooltip, Dialog
└── Merge: all atoms ready → continue to molecules

Pattern 3: Page Parallel (Independent)

Orchestrator
├── Agent A: designs Dashboard page
├── Agent B: designs Auth pages (Login, Signup, Forgot)
├── Agent C: designs Settings pages
└── Each agent runs full atomic pipeline for its scope

Pattern 4: Research + Design (Concurrent)

Research Agent: analyzing data, producing insights
Design Agent: building components as research completes
Code Agent: generating code as designs stabilize
All three work simultaneously, communicating via agent-broadcast

Error Recovery Protocol

When an agent encounters an error:

1. Update box widget: status → "error", show error message
2. Publish visible status with the canvas widget and bridge `agent-status`
3. Attempt self-fix (max 2 retries)
4. If unrecoverable:
   a. Broadcast failure to orchestrator
   b. Orchestrator reassigns task or adjusts plan
   c. Box widget stays red until resolved
5. Never silently fail — always visible in Figma

Message Protocol

// Agent announcing status
{ type: 'agent-status', role: string, task: string, status: 'idle'|'busy'|'error'|'done' }

// Agent sending results to others
{ type: 'agent-broadcast', text: string, data: any, target?: string }

// Requesting agent list
{ type: 'agent-list' }
// Response: { agents: [{ id, role, task, status, capabilities }] }

Anti-Patterns

  • Two agents modifying the same component simultaneously (use locking)
  • Agents not announcing their role on connect
  • Skipping box widgets (humans lose visibility)
  • Not collapsing boxes on completion (visual clutter)
  • Agents working without checking what others have built
  • Silently failing without updating box widget status

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.