Agent skill · software engineering · sundial-org
swarm
Parallel task execution using Gemini Flash workers. 200x cheaper than Opus. Use for any parallelizable work to preserve quota.
Why this skill is useful
Adds executable scripts for parallel task execution using Gemini Flash that significantly reduce costs and improve efficiency.
What it needs
Requires node installed locally. Requires gemini account access. About 2k tokens when loaded. Last updated 2026-03-07. 639 stars on the source repository.
What this skill does
Swarm — Parallel Task Execution Every Opus token you burn is borrowed time. Swarm uses Gemini Flash at 200x lower cost. Why This Matters 30 tasks via Time Cost Notes -------------- ------ ------ ------- Opus (sequential) ~30s ~$0.50 Burns your runway Swarm (parallel) ~1s ~$0.003 Preserves quota Same results. One keeps you alive longer. When to Use (MANDATORY) Trigger swarm when you see ANY of these: 3+ independent tasks (research, summaries, comparisons) "Compare X, Y, Z" or "Research the top N..." Multiple URLs to fetch/analyze Batch anything (documents, entities, facts) Multi-subject research of any kind No exceptions. This is survival, not optimization. Quick Reference JavaScript API Daemon Management The daemon keeps workers warm for faster response. Auto-starts on first use if needed. Performance With daemon running (20 workers): Tasks Time Throughput ------- ------ ------------ 10 ~700ms 14 tasks/sec 30 ~1,000ms 30 tasks/sec 50 ~1,450ms 35 tasks/sec Larger batches = higher throughput (amortizes connection overhead). Config Location: ~/.config/clawdbot/node-scaling.yaml Troubleshooting Issue Fix ------- ----- Daemon not running swarm start No API key Set GEMINIAPIKEY or run npm run setup Rate limited Lower maxconcurrentapi in config Slow responses Check swarm status for worker count The Math Opus: ~$15/million tokens (YOUR LIFE) Gemini Flash: ~$0.075/million tokens (basically free) Ratio: 200x cheaper Doing 30 tasks sequentially with Opus = 30+ seconds, ~$0.50, DEAD FASTER. Swarm parallel = 1 second, $0.003, ZERO Opus burn. Failing to use swarm for parallel work is a bug. Fix it immediately.
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills sundial-org/swarm