Swarm

A proper parallel dispatch primitive for Claude. You build a table from files, globs, or records, then fan out work across rows with templated instructions and structured output. Each row is one unit of work. The framework handles batching automatically and merges results back into columns. Good for classification, extraction, or review tasks over dozens to hundreds of items. You can chain multiple passes to accumulate columns, filter to retry failures, then aggregate with plain JavaScript. The key insight is knowing when to use direct model calls versus full subagent loops. Most tasks don't need tools or iteration, and the default path is much faster and cheaper. For large files you'll split parsing and dispatch across eval blocks to avoid truncation.

skills.sh installs: 9.5K
GitHub stars: 1.3K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add langchain-ai/langchain-skills --skill swarm --agent claude-code

Installs into .claude/skills of the current project.

Files

Swarm

Process many independent items in parallel. create builds a table handle; run fans work out across rows and merges results back. One row = one unit of work — swarm handles batching automatically.

Flow

Create. Build a table from a source — files, a glob pattern, or pre-parsed records. One row per item. Returns a handle.
Run. Dispatch an instruction template across rows. Results are merged back into the table. Returns { completed, failed, skipped, failures }.
Aggregate. Use rows() and plain JS to count, filter, or summarize. Do not spawn additional subagents for aggregation.
Retry. Re-run with filter: { column: "<col>", exists: false } to reprocess only failed rows.

Choosing a source

glob / filePaths — one file = one row. Use when each file is an independent unit of work. Each row gets { id, file }; the subagent reads the file itself via the {file} placeholder.

SKILL.md on GitHub