feature-swarm

Feature Swarm

A Claude Code plugin that spawns a swarm of parallel AI agents to discover, research, and document features for your next app or project.

Inspired by Ruflo's swarm feature, adapted for Claude Code's native Agent infrastructure.

What It Does

You describe your app idea. Feature Swarm handles the rest:

flowchart TD
    Input["Your app description"]

    Input --> D

    subgraph D["Phase 1: Discovery (parallel)"]
        direction LR
        A1["Competitor\nAnalyst"]
        A2["User Journey\nExplorer"]
        A3["Technical\nCapability Scout"]
    end

    D --> Merge["Merge & Deduplicate\nAssign tiers + complexity"]

    Merge --> Select["Phase 2: Select\nTiered checklist\nMVP / Nice-to-Have / Future"]

    Select --> R

    subgraph R["Phase 3: Deep Research (parallel)"]
        direction LR
        R1["Agent:\nFeature 1"]
        R2["Agent:\nFeature 2"]
        R3["Agent:\nFeature 3"]
        R4["Agent:\nFeature N"]
    end

    R --> Output["Phase 4: Output"]

    Output --> README["docs/research/README.md\nMaster index"]
    Output --> Brief["docs/research/project-brief.md\nStructured brief for next AI"]
    Output --> Features["docs/research/features/\nPer-feature research files"]

Phase 1: Discover -- 3 agents research features in parallel from different angles:

• Competitor Analyst -- searches the web for similar products and extracts their feature sets
• User Journey Explorer -- maps user workflows and identifies implied features
• Technical Capability Scout -- researches what modern APIs/libraries make possible

Phase 2: Select -- presents a tiered checklist (MVP / Nice-to-Have / Future) with complexity ratings and agent consensus scores. Pick specific features, select all, or just the MVP tier.

Phase 3: Research -- spawns 1 agent per selected feature for deep competitive analysis + technical feasibility, all running in parallel.

Phase 4: Output -- saves structured research as markdown files, ready for the next AI session to build from.

Example

/feature-swarm I want to build a habit tracking app where users create daily habits, mark them complete, and see their streaks

Output:

docs/research/
├── README.md              # Master index with feature table
├── project-brief.md       # Structured brief for AI-assisted planning
├── features/
│   ├── 01-user-auth.md    # Competitive + technical research
│   ├── 02-streak-tracking.md
│   ├── 03-notifications.md
│   └── ...

To continue building after research:

Read docs/research/project-brief.md and create an implementation plan

Installation

Option 1: As a Claude Code Plugin (Recommended)

Add the GitHub repo as a marketplace, then install:

/plugin marketplace add lucatescari/claude-feature-swarm
/plugin install feature-swarm@lucatescari/claude-feature-swarm

The skill becomes available as /feature-swarm in all your Claude Code sessions.

Option 2: Manual Installation

Copy the skill files directly to your Claude Code skills directory:

git clone https://github.com/lucatescari/claude-feature-swarm.git
mkdir -p ~/.claude/skills/feature-swarm
cp -r claude-feature-swarm/skills/feature-swarm/* ~/.claude/skills/feature-swarm/

The skill will be auto-discovered by Claude Code -- no restart needed.

How It Works

Feature Swarm is a Claude Code skill (a markdown guidance document, not application code). When invoked, it instructs Claude to:

• Read prompt templates from the prompts/ directory
• Dispatch parallel agents using Claude Code's built-in Agent tool
• Merge and deduplicate results using Claude's reasoning
• Present an interactive selection UI via AskUserQuestion
• Write structured output files to your project

No external dependencies. No API keys beyond your existing Claude Code setup. No servers to run.

File Structure

skills/feature-swarm/
├── SKILL.md                           # Main orchestration logic
├── prompts/
│   ├── competitor-analyst.md          # Discovery agent: market research
│   ├── user-journey-explorer.md       # Discovery agent: user perspective
│   ├── technical-capability-scout.md  # Discovery agent: tech possibilities
│   └── feature-researcher.md         # Deep research agent (1 per feature)

Output Format

Feature Selection Checklist

After discovery, you see a tiered list like:

-- MVP / Must-Have --
[1] User Authentication          * complexity | 2/3 agents
    Registration, login, password reset
[2] Streak Tracking              * complexity | 3/3 agents
    Consecutive day counts with visual display

-- Nice-to-Have --
[3] Offline Support              ** complexity | 3/3 agents
    Works without internet, syncs when connected

-- Future / Stretch --
[4] AI Habit Coaching            *** complexity | 1/3 agents
    Personalized insights and streak predictions