Show HN: Swarm – Claude Code에서 일관된 결과 얻기

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. 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. 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. /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 /swarm:code # Code team /swarm:write # Writing team /swarm:general # General team | 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 | | Shape | Members | Facilitator | |---|---|---| | Balanced (default) | Sonnet | Opus | | Ultra | Opus | Opus | /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 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. 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. Run /swarm:create-workflow to scaffold a custom mode for your project: domain-specific phases plus a shortcut command. Use /swarm:workflow to launch against any existing custom mode. Run /swarm:update to check for and install the latest version. claude plugin marketplace update swarms claude plugin update swarm@swarms --scope project | 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 | All changes go through branches and pull requests. Automated version bumps by github-actions[bot] are the only exception. MIT