swarm-coordination

Coordinate multi-agent swarm workflows with mandatory swarmmail and nine lifecycle steps.

Updated Aug 30, 2024

Search Tags:#multi-agent #swarm #coordination #bd #swarmmail

Ready to run

npx skills add https://github.com/minhtri2710/dotfiles --skill swarm-coordination-minhtri2710

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: swarm-coordination
Source: https://github.com/minhtri2710/dotfiles/tree/main/dot_config/opencode/skill/swarm-coordination
Command: npx skills add https://github.com/minhtri2710/dotfiles --skill swarm-coordination-minhtri2710

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Multi-agent orchestration for OpenCode swarm workflows. The coordinator decomposes tasks, spawns workers, tracks progress, and aggregates results. Swarm Mail is mandatory for all coordination.

Core Features & Use Cases

Mandatory Swarm Mail: All coordination must use swarm mail for messaging and reservations.
Worker Survival Checklist: Nine mandatory steps for workers.
Progress & Checkpoints: Regular updates and checkpoints.

Quick Start

swarmmail_init({ project_path: "/abs/path/to/project", task_description: "bead-id: Task description" })

Frequently Asked Questions about swarm-coordination

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I coordinate multiple agents to work on different parts of a task in parallel?▼

Multi-agent coordination decomposes tasks across workers using swarm patterns. Initialize with swarmmail, decompose the task into subtasks, spawn workers via swarm_spawn_subtask, track progress with swarm_status, and aggregate results after all agents complete. Swarm Mail is mandatory for all messaging and reservations.

What are the required steps for workers in a swarm workflow?▼

Workers must follow nine mandatory steps: initialize, query learnings, load skills, reserve files, do work, report progress, checkpoint, store learnings, and complete. Each step uses specific tools like swarmmail for communication, semantic_memory_find for discovery, and swarm_progress for tracking to ensure coordinated execution and result collection.

How do I decompose a complex task across frontend, backend, and test subsystems?▼

Use swarm_decompose to break tasks into parallel work units targeting specific subsystems. Validate decomposition with swarm_validate_decomposition, then spawn subtasks with swarm_spawn_subtask. Track cross-subsystem progress via swarm_status and swarm_progress, using swarmmail for inter-worker communication and file reservations.

Can I use swarm coordination for knowledge gathering and plan generation?▼

Yes. Swarm coordination supports knowledge gathering via semantic_memory_find, cass_search, and pdf-brain_search, then generates plans with swarm_plan_prompt. Workers execute in parallel, report findings through swarmmail checkpoints, and aggregate results before completing, enabling complex multi-step reasoning across agents.

What happens if a worker fails during task execution in a swarm?▼

Workers checkpoint progress regularly and store learnings to enable recovery. If a worker fails, swarm_status detects the issue, and swarm_review validates partial results. Checkpoints preserve state so remaining workers continue or retry—swarmmail reservations prevent conflicting file access during recovery.

Do I need to manually manage file access when multiple workers operate on the same codebase?▼

No. Swarm Mail handles mandatory file reservations and releases across workers, preventing conflicts. Each worker reserves files before work, reports progress through swarmmail, and releases after checkpointing, ensuring safe parallel execution without manual locking.