Skillget

swarms

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Uyari
  • network request — Outbound network request in scripts/check-update.sh
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This tool is a collaborative agent framework designed for Claude Code. It structures AI sessions into repeatable workflows by deploying "swarms" of agents that research, execute, and review tasks to produce high-quality outcomes.

Security Assessment
The overall risk is Low. The tool does not request any dangerous system permissions and contains no hardcoded secrets. As a framework that orchestrates AI agents, it inherently executes commands and interacts with your local project files. The automated scan flagged an outbound network request, but this is limited to a benign script that checks for software updates. Users should be aware that the AI agents themselves might execute shell commands or modify code depending on the specific task they are given.

Quality Assessment
The project is actively maintained, with its most recent push occurring today. It is protected under the standard MIT license and features clear, comprehensive documentation. However, community trust and visibility are currently minimal. The repository has only 5 GitHub stars, indicating that very few external developers have reviewed or tested the code. It is essentially an early-stage personal project recently opened to the public.

Verdict
Use with caution; the code itself appears safe and well-documented, but its early stage and lack of widespread community review warrant a careful watch over what the AI agents do during your sessions.
SUMMARY

Achieve extraordinary results with claude code across a variety of tasks

README.md

Swarm

Good Claude Code sessions are repeatable. Swarm is the structure that makes them that way.

I built this across hundreds of sessions, pruning rules, memories, and skills until the quality stopped varying. When model quality shifted, small targeted changes kept it working, even on smaller models. Once the results were consistent enough to rely on, I started sharing with teammates and friends.

Prerequisites

Agent teams must be enabled in Claude Code. Add to ~/.claude/settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

The launch command checks this automatically and guides you through enablement if it's missing.

Installation

claude plugin marketplace add DheerG/swarms
claude plugin install swarm@swarms --scope project

For development or local testing:

claude --plugin-dir /path/to/swarms

Changes take effect in the next session.

Usage

Launch a team

/swarm:launch

Describe your outcomes, choose defaults or configure each step, review the plan, then launch. Pass outcomes inline to skip the opening question:

/swarm:write Help me write a blog article on healing traumas

Mode shortcuts

/swarm:code           # Code team
/swarm:write          # Writing team
/swarm:general        # General team

Modes

Mode Lead Facilitator Review
Code Writes code Principal Engineer Technical review
Writing Coordinates (can write) Editorial Director Editor-sandwich review
General Produces deliverable Chief of Staff Facilitator-driven review

Team shapes

Shape Members Facilitator
Balanced (default) Sonnet Opus
Ultra Opus Opus

Other skills

/swarm:audit-context      # Check your project for artifacts that may interfere with swarm
/swarm:refine-outcomes    # Reframe ideas into outcome statements
/swarm:suggest-members    # Recommend team composition

How it works

Every team follows the same phase arc: Research, Converge, Approve, Execute, Review, Refine, Deliver. Phases run in order.

Review is recursive: if the team doesn't reach a 9/10 confidence score, it cycles back through Refine before presenting to you. The facilitator drives this, not the lead. Work reaches you only after the team has agreed it clears the threshold. Most of the quality work happens in the cycles you never see.

Watching your team

The most useful signal isn't the final output. It's the discussion. Watching agents reason, push back, and converge tells you whether the team is actually working. AgentChat is a companion tool that surfaces agent conversations in real time. You'll likely find yourself watching the discussion more than Claude Code itself.

Custom workflows

Run /swarm:create-workflow to scaffold a custom mode for your project: domain-specific phases plus a shortcut command. Use /swarm:workflow <mode-name> to launch against any existing custom mode.

Updating

Run /swarm:update to check for and install the latest version.

claude plugin marketplace update swarms
claude plugin update swarm@swarms --scope project

Ubiquitous Language

Term Meaning
swarm The plugin
team A group of agents launched for a task
lead The main session that coordinates work
member A teammate agent (read-only)
facilitator The Socratic facilitator role (Principal Engineer / Editorial Director / Chief of Staff)
outcome What the user wants to achieve
mode The team's domain configuration (Code, Writing, General)
shape Model allocation tier (Balanced, Ultra)
phase arc Research, Converge, Approve, Execute, Review, Deliver
launch Start a team via /swarm:launch or a mode shortcut

Contributing

All changes go through branches and pull requests. Automated version bumps by github-actions[bot] are the only exception.

License

MIT

Yorumlar (0)

Sonuc bulunamadi