swarm

Coordinate parallel agent tasks with atomic SQLite claims and heartbeat liveness.

2|Updated Feb 3, 2026

Search Tags:#SQLite #concurrency #agents #task coordination #swarm #fault tolerance #distributed tasks

Ready to run

npx skills add https://github.com/chanhee-kang/oh-my-gemini --skill swarm-chanhee-kang

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: swarm
Source: https://github.com/chanhee-kang/oh-my-gemini/tree/main/skills/swarm
Command: npx skills add https://github.com/chanhee-kang/oh-my-gemini --skill swarm-chanhee-kang

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill enables multiple agents to collaborate on a shared task list, claim tasks atomically via SQLite transactions, and track progress with heartbeats to prevent deadlocks.

Core Features & Use Cases

Atomic task claiming using SQLite transactions to prevent race conditions
Heartbeat-based liveness and automatic stale-task release
Persistent state and task tracking via a local database; scalable to many tasks
Use Case: parallel code fixes, design reviews, and distributed task execution across a small team

Quick Start

Start a swarm by specifying the number of agents and a list of tasks. For example, initiate 5 agents to work on tasks like “fix tests in module A”, “document API surface”, and “refactor module B”.

Frequently Asked Questions about swarm

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

FAQPage Schema

How do I coordinate parallel tasks across multiple agents without race conditions?▼

Distributed task coordination for multiple agents uses a shared SQLite-backed task list with atomic claiming to prevent race conditions. Agents claim, execute, and track tasks concurrently via start, claim, complete, and fail API operations.

How does SQLite handle agent crash recovery and stale task reassignment?▼

Agent crash recovery and stale task reassignment rely on heartbeat-based liveness and lease timeouts within the SQLite database. When an agent crashes, lease timeouts automatically release stale tasks for other agents to claim.

What is the best way to parallelize code fixes across a small team of agents?▼

Parallelizing code fixes across a small team of agents is best handled by a swarm coordination approach. Specify the number of agents and a list of tasks, and they will atomically claim and execute work concurrently via a local SQLite database.

Do I need external servers to manage distributed task execution for concurrent agents?▼

No external servers are needed to manage distributed task execution for concurrent agents. The system relies entirely on a local SQLite database for persistent state tracking, transactional claims, and heartbeat-based liveness monitoring.

When should I use SQLite-backed task coordination for distributed work?▼

Use SQLite-backed task coordination for distributed work when you need to scale to many tasks across a small team of agents. It suits scenarios requiring transactional task claims, heartbeat-based liveness, and automatic recovery from crashes.