skills.rest
jgabriellima
jgabriellimaCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

swarm

Dispatches batches of tasks to subagents in parallel with bounded concurrency.

Updated May 7, 2026
Search Tags:#typescript#parallel execution#concurrency#subagents#deep agents#task dispatch#worker pool
One-click install
npx skills add https://github.com/jgabriellima/deep-agent-skill --skill swarm-jgabriellima
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: swarm
Source: https://github.com/jgabriellima/deep-agent-skill/tree/main/skills/swarm
Command: npx skills add https://github.com/jgabriellima/deep-agent-skill --skill swarm-jgabriellima

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running many independent subagent tasks sequentially is slow, and firing them all at once with unbounded parallelism can exhaust memory and tool-call rate limits. This Skill fans out a list of tasks to subagents with a bounded worker pool and returns a compact summary of successes and failures. ## Core Features & Use Cases - Bounded Parallel Dispatch: Runs tasks through a semaphore-style worker pool with configurable concurrency (default 5, capped at 10). - Per-Task Fault Isolation: One failed subagent does not abort the batch; each result carries its own status, output, or error message. - Ordered Summary Object: Returns {total, completed, failed, results[]} with results preserved in input order for easy iteration. - Use Case: Summarize 20 note files, classify 50 support tickets, or research 15 topics concurrently, then iterate the results array to collect each subagent's output. ## Quick Start Ask the agent to import the swarm skill with await import("@/skills/swarm") and call runSwarm with your list of task descriptions and a concurrency limit.

Frequently Asked Questions about swarm

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

FAQPage Schema
How do I run multiple subagent tasks in parallel?▼

Call runSwarm with an array of task descriptions and an optional concurrency value. It dispatches tasks through the agent's task tool using a worker pool, so at most the configured number of subagents run at once, and returns a summary with per-task results.

How do I load the swarm skill in the REPL?▼

Use await import("@/skills/swarm") inside the REPL's eval tool, which resolves @/skills/* paths against the skills backend. Do not inline the index.ts source into your eval body, since importing is the supported path.

What is the maximum concurrency for parallel subagent dispatch?▼

Concurrency defaults to 5 and is capped at 10. Values above 10 are clamped, and values below 1 are raised to 1, keeping memory usage and tool-call rate predictable even for large task batches.

What happens when one subagent task fails in a batch?▼

Failures are caught per task, so one failed subagent does not abort the rest of the batch. The returned summary includes a failed count, and each result entry carries a status of completed or failed plus an error message on failure.

Can I use different subagent types for different tasks?▼

Yes. Each task accepts an optional subagentType field that overrides the default. You can also set a swarm-wide default subagentType, which falls back to general-purpose when neither is specified.

Related Skills

Explore More Software Engineering Skills→
EpicenterHQEpicenterHQ
4.8k

wayfinder

Plans multi-session work as decision tickets tracked on a repository issue tracker.

Official
Advanced
cursorcursor
6.6k

swarm

Fan out parallel cloud workers and aggregate their results into one report.

Official
Intermediate
shakedfbshakedfb

ralph

Manages crop production tasks with customizable workflows and automated reporting for each stage.

Community
Advanced
aromanarguelloaromanarguello
9

parallel-subagents

Dispatch multiple focused subagents in one message and synthesize their summaries.

Community
Intermediate