Generated banner for Swarm
Generated logo for Swarm

Swarm

LangChainOfficialPublished by LangChain

  • Skill
  • Tool integration
  • Automation and integrations
  • MIT
  • Low risk
  • Has scripts

From the author

Dispatches many independent items in parallel: create a table, fan out to subagents, aggregate results. One row = one unit of work.
— the description by LangChain

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

Swarm provides a structured way to process large volumes of independent items in parallel. By organizing data into tables, it allows users to fan out instructions to subagents or direct model calls, aggregate results, and perform targeted retries on failed items. It is designed to handle batching automatically, making it efficient for tasks ranging from simple data classification to complex, multi-step agentic workflows.

What it does

  • Creates tabular data structures from files, glob patterns, or custom record sets.
  • Dispatches instructions across rows in parallel using subagents or direct model calls.
  • Supports automatic and custom batching to optimize throughput and cost.
  • Enables chaining of multiple passes to accumulate data on existing rows.
  • Provides filtering capabilities for targeted processing and error recovery.
  • Allows for local JavaScript-based aggregation of results after processing.

When to use it

  • When you need to perform the same task on a large number of independent files or records.
  • When you want to classify, extract, or label data using structured output.
  • When a task requires complex, multi-step reasoning or tool usage across many items.
  • When you need to retry only the failed portions of a large batch operation.

Advantages

  • Significantly faster and cheaper for simple tasks by using direct model calls.
  • Handles large datasets by managing batching and concurrency automatically.
  • Maintains state across multiple passes, allowing for iterative data enrichment.
  • Flexible source handling supports both file-based and custom record-based inputs.

Usage examples

“Classify the sentiment of every JSON object in data.jsonl using the swarm skill.”
The agent creates a table from the records, runs a classification instruction across all rows, and returns a summary of the sentiment distribution.
“Use swarm to review all TypeScript files in src/ for security issues.”
The agent identifies files via glob, dispatches a 'reviewer' subagent to each file, and compiles a table of findings.

Install

  1. Download the skill as a ZIP:

    https://mp-staging.autogineer.com/agent-skills/langchain-ai/swarm/download
  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip swarm.zip -d .agents/skills

Documentation by LangChain

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

  1. Create. Build a table from a source — files, a glob pattern, or pre-parsed records. One row per item. Returns a handle.

  2. Run. Dispatch an instruction template across rows. Results are merged back into the table. Returns { completed, failed, skipped, failures }.

  3. Aggregate. Use rows() and plain JS to count, filter, or summarize. Do not spawn additional subagents for aggregation.

  4. 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.

tasks — pass pre-built records directly. Use when the data lives inside a file (JSONL, CSV, JSON array). Read and parse the file first inside eval, then pass the records. One record = one row — do not group multiple items into a single row.

For small files (under ~500 lines), parse and create in one block:

const { create } = await import("@/skills/swarm");
const raw = await tools.readFile({ file_path: "/data.jsonl" });
const records = raw.trim().split("\n").map(l => JSON.parse(l));
const table = await create({ tasks: records });
console.log(table);

For large files, read in chunks of 500 lines to avoid truncation:

const { create } = await import("@/skills/swarm");
let records = [];
let offset = 0;
while (true) {
  const chunk = await tools.readFile({ file_path: "/data.txt", offset, limit: 500 });
  const lines = chunk.split("\n").filter(l => l.trim());
  for (const l of lines) { records.push({ id: `r${records.length}`, text: l }); }
  if (lines.length < 500) break;
  offset += 500;
}
const table = await create({ tasks: records });
console.log(table);

When the file is too large to parse and dispatch in one eval call, split across two blocks. Only the block that calls swarm functions needs the import:

// eval 1: parse only — no swarm import needed
const raw = await tools.readFile({ file_path: "/data.jsonl" });
globalThis.records = raw.trim().split("\n").map(l => JSON.parse(l));
console.log(`Parsed ${globalThis.records.length} records`);
// eval 2: create and dispatch
const { create, run } = await import("@/skills/swarm");
const table = await create({ tasks: globalThis.records });
const result = await run(table.id, {
  instruction: "Classify {text}",
  responseSchema: {
    type: "object",
    properties: { label: { type: "string" } },
    required: ["label"],
  },
});
console.log(result);

Passing filePaths: ["/data.jsonl"] would produce a table with one row pointing at the file — not one row per record inside it.

When to use subagentType

Omit subagentType for classification, extraction, labeling, and any task where a single model call with structured output is sufficient. This is the default and is significantly cheaper and faster — each dispatch is a direct model call, no tools, no iteration.

Set subagentType when the task requires tools, file access, or multi-step reasoning. Each dispatch runs a full agentic loop with the named subagent.

// Direct model call — classification, no tools needed
await run(table.id, {
  instruction: "Classify {text}",
  responseSchema: { type: "object", properties: { label: { type: "string" } }, required: ["label"] },
});

// Subagent — needs to read files and reason over multiple steps
await run(table.id, {
  subagentType: "reviewer",
  instruction: "Review {file} for security issues.",
  responseSchema: { type: "object", properties: { finding: { type: "string" } }, required: ["finding"] },
});

Instruction + context

instruction is a per-item template with {column} placeholders. Placeholders are resolved by the framework — your column names appear in prompts as references to the values listed alongside, never as raw template syntax. Subagents do the work — do not process items yourself in JS and write the results into rows.

context is free-form prose prepended to every subagent prompt. Use it for shared background: domain terms, classification rules, examples, etc.

const { create, run } = await import("@/skills/swarm");

const table = await create({ glob: "src/**/*.ts" });
const r = await run(table.id, {
  subagentType: "reviewer",
  instruction: "Review {file} for security issues. List findings or write 'no issues'.",
  context: "TypeScript Express backend using Prisma ORM. Focus on injection, auth bypass, path traversal.",
  responseSchema: {
    type: "object",
    properties: { review: { type: "string" } },
    required: ["review"],
  },
});
console.log(r);
// → { completed: 45, failed: 2, skipped: 0, failures: [...] }

Structured output

responseSchema is required. Schema properties become top-level columns on each row and constrain what subagents can return.

const { run } = await import("@/skills/swarm");
await run(table.id, {
  instruction: "Classify: {text}",
  responseSchema: {
    type: "object",
    properties: {
      sentiment: { type: "string", enum: ["positive", "negative", "neutral"] },
    },
    required: ["sentiment"],
  },
});
// Row after: { id: "r1", text: "...", sentiment: "positive" }

Batching

By default, swarm auto-batches to keep total dispatches under 10. For small tables (≤10 rows) each row gets its own subagent call. For larger tables, rows are grouped automatically.

Set batchSize to control grouping:

  • Number — uniform batch size for all rows. batchSize: 1 forces per-row dispatch; batchSize: 20 groups in twenties.

  • Function — (row, rowCount) => number. Returns the desired batch size for each row. Rows with the same batch size are grouped together, then chunked. Allows mixed dispatch where some rows go solo and others batch.

const { create, run } = await import("@/skills/swarm");
const table = await create({ tasks: items });

// Complex items get individual attention; simple ones batch together
await run(table.id, {
  instruction: "Analyze {text}",
  responseSchema: {
    type: "object",
    properties: { analysis: { type: "string" } },
    required: ["analysis"],
  },
  batchSize: (row) => (row.token_count > 1000 ? 1 : 10),
});

Batch sizes are clamped to [1, 50] after evaluation.

Aggregation

After run(), use rows() and plain JS — no additional subagents needed.

const { rows } = await import("@/skills/swarm");
const data = await rows(table.id, { columns: ["sentiment"] });
const counts = {};
data.forEach(r => { counts[r.sentiment] = (counts[r.sentiment] || 0) + 1 });
console.log(counts);
// → { positive: 120, negative: 45, neutral: 35 }

Chaining passes

run updates the table in place — chain calls to accumulate columns.

const { create, run } = await import("@/skills/swarm");
const table = await create({ tasks: interviews });
await run(table.id, {
  instruction: "Classify sentiment of {text}",
  responseSchema: {
    type: "object",
    properties: { sentiment: { type: "string", enum: ["positive", "negative", "neutral"] } },
    required: ["sentiment"],
  },
});
await run(table.id, {
  filter: { column: "sentiment", equals: "negative" },
  instruction: "Summarize why {text} had negative sentiment.",
  responseSchema: {
    type: "object",
    properties: { summary: { type: "string" } },
    required: ["summary"],
  },
});

Action-only tasks

When subagents perform actions (write a file, apply a fix) rather than return data, use a simple schema with a status or marker field. The exists: false filter still works for retries.

const { create, run } = await import("@/skills/swarm");
const fixedSchema = {
  type: "object",
  properties: { fixed: { type: "string" } },
  required: ["fixed"],
};
const table = await create({ glob: "src/**/*.ts" });
await run(table.id, {
  subagentType: "fixer",
  instruction: "Add missing JSDoc to all exported functions in {file}.",
  responseSchema: fixedSchema,
});
// retry any that failed
await run(table.id, {
  subagentType: "fixer",
  instruction: "Add missing JSDoc to all exported functions in {file}.",
  responseSchema: fixedSchema,
  filter: { column: "fixed", exists: false },
});

Filtering

{ column: "status", equals: "done" }
{ column: "status", notEquals: "done" }
{ column: "category", in: ["A", "B"] }
{ column: "result", exists: false }      // not yet processed
{ and: [filter1, filter2] }
{ or: [filter1, filter2] }

Technical notes

  • Only import @/skills/swarm in blocks where you call swarm functions. Data preparation (reading files, parsing, storing in globalThis) does not need the import. Destructure only what you use: { create }, { run }, { create, run }, etc.

  • Console output is capped at ~5 KB. Never log raw file contents — log only counts and short samples.

  • readFile inside eval returns raw content — no line-number prefixes. Request at most 500 lines per call. For files with more than 500 lines, loop with incrementing offset.

  • When building a table from a file, read it inside eval. Data read inside the sandbox stays there; it never enters the agent's context window.

  • Never write to .swarm/ directly. Always use create().

  • Everything the subagent needs must be in instruction + context. Subagents can't see the agent's context.

  • Row ids must be unique. create() rejects sources that produce duplicate ids. For tasks, that's a caller-side responsibility; for glob / filePaths, ids are auto-disambiguated by parent directory.

  • Unknown columns fail fast. If instruction references {foo} and no matched row provides foo, run() throws before any subagent is dispatched.

API Reference

create(source)

Create a table. Returns a handle { id, count, columns }.

SourceDescription
{ glob: "src/**/*.ts" } or { glob: ["src/**/*.ts", "lib/**/*.ts"] }Match files by one or more patterns. Columns: id, file
{ filePaths: ["a.ts", "b.ts"] }Explicit file list. Columns: id, file
{ tasks: [{ id: "t1", text: "..." }] }Custom rows. Each must have id
run(tableId, options)

Dispatch work across rows. Returns { completed, failed, skipped, failures }.

OptionDefaultDescription
instruction(required)Template with {column} placeholders
responseSchema(required)JSON Schema (type: "object") — properties become row columns
context—Prose prepended to every subagent prompt
filter—Only dispatch matching rows
subagentType—Name of subagent to dispatch to. When set, runs a full agentic loop. When omitted, runs a direct model call
batchSizeautoNumber or (row, rowCount) => number. Auto caps dispatches at 10; 1 = per-row; function = per-row sizing
concurrency10Max concurrent subagent dispatches (clamped to 1–10)
rows(tableId, options?)

Retrieve rows. Use for inspection and JS-based aggregation.

OptionDescription
filterOnly return matching rows
columnsProject to specific columns
limitMax rows returned

Files

scripts/types.ts

/**
 * Lightweight handle returned by `create()`.
 *
 * Contains only metadata — actual row data stays on the backend.
 * The agent uses this handle to reference the table in subsequent
 * `run()` and `rows()` calls. Handles are stable across evals.
 */
export interface SwarmHandle {
  /**
   * Unique table identifier (e.g. `"t_a1b2c3"`).
   */
  id: string;

  /**
   * Number of rows in the table at creation time.
   */
  count: number;

  /**
   * Column names present in the first row (e.g. `["id", "file"]`).
   */
  columns: string[];
}

/**
 * Source specification for `create()`.
 *
 * Exactly one of `glob`, `filePaths`, or `tasks` must be set.
 * Providing zero or more than one source throws an error.
 */
export interface CreateSource {
  /**
   * Glob pattern(s) to match files. Each match becomes a row with
   * `{ id: <basename>, file: <full path> }` columns. Requires a
   * `glob` tool in the PTC configuration.
   */
  glob?: string | string[];

  /**
   * Explicit list of file paths. Same row structure as `glob`
   * (`{ id, file }`) but skips pattern resolution.
   */
  filePaths?: string[];

  /**
   * Custom row data. Each object must include a string `id` field.
   * All other fields become table columns.
   */
  tasks?: Array<Record<string, unknown>>;
}

/**
 * Per-row batch size function.
 *
 * Returns the desired batch size for a given row. Rows that return
 * the same batch size are grouped together, then chunked into
 * batches of that size.
 */
export type BatchFn = (
  row: Record<string, unknown>,
  rowCount: number,
) => number;

/**
 * Options for `run()`.
 *
 * Controls how rows are selected, how instructions are templated,
 * and how subagent dispatch is configured.
 */
export interface RunOptions {
  /**
   * Instruction template with `{column}` placeholders that are
   * interpolated per-row (e.g. `"Review {file} for security issues"`).
   */
  instruction: string;

  /**
   * Context prose prepended to every subagent prompt. Use for shared
   * background that applies to all rows (e.g. project description).
   */
  context?: string;

  /**
   * Filter clause to select a subset of rows. Rows that don't match
   * are skipped (counted in `RunResult.skipped`).
   */
  filter?: SwarmFilter;

  /**
   * Name of the subagent type to dispatch to. When set, each dispatch
   * runs a full agentic loop with tools. When omitted, each dispatch
   * is a direct model call with structured output (no tools, no iteration).
   */
  subagentType?: string;

  /**
   * JSON Schema (type: "object") for structured output. Each property
   * in the schema becomes a top-level column on the row.
   */
  responseSchema: Record<string, unknown>;

  /**
   * Controls how rows are grouped into subagent calls.
   *
   * - **Number**: uniform batch size for all rows.
   * - **Function**: called per-row, returns desired batch size. Rows with
   *   the same batch size are grouped together, then chunked.
   *
   * Batch sizes are clamped to [1, MAX_BATCH_SIZE] after evaluation.
   *
   * @default auto-batch based on table size to cap total dispatches.
   */
  batchSize?: number | BatchFn;

  /**
   * Maximum concurrent subagent dispatches. Clamped to [1, MAX_SUBAGENTS].
   * Defaults to MAX_SUBAGENTS (10) when omitted.
   */
  concurrency?: number;
}

/**
 * Summary returned by `run()`.
 *
 * Contains counts and deduplicated failure groups. The agent uses
 * this to decide whether to retry, inspect, or proceed.
 */
export interface RunResult {
  /**
   * Number of rows where the subagent succeeded and a result was merged.
   */
  completed: number;

  /**
   * Number of rows where the subagent failed or interpolation failed.
   */
  failed: number;

  /**
   * Number of rows excluded by the filter (not dispatched).
   */
  skipped: number;

  /**
   * Failures grouped by error message, sorted by count descending.
   */
  failures: FailureGroup[];
}

/**
 * A group of rows that failed with the same error message.
 *
 * Deduplication keeps the failure list compact even when hundreds of
 * rows hit the same error (e.g. rate limiting).
 */
export interface FailureGroup {
  /**
   * The error message shared by all rows in this group.
   */
  error: string;

  /**
   * Number of rows that hit this error.
   */
  count: number;

  /**
   * IDs of all rows that hit this error.
   */
  ids: string[];
}

/**
 * Options for `rows()`.
 *
 * Controls filtering, column projection, and row limiting when
 * retrieving table data for inspection or aggregation.
 */
export interface RowsOptions {
  /**
   * Filter clause — only rows matching the filter are returned.
   */
  filter?: SwarmFilter;

  /**
   * Project to specific columns. Omit to return all columns.
   */
  columns?: string[];

  /**
   * Maximum number of rows to return. Omit for no limit.
   */
  limit?: number;
}

/**
 * Filter clause for selecting rows. Can be a leaf predicate or a
 * combinator (`and`/`or`) composing multiple clauses.
 *
 * Leaf predicates operate on a single column (supports dot-paths
 * for nested access, e.g. `"meta.score"`).
 */
export type SwarmFilter =
  | {
      /**
       * Column path to compare.
       */
      column: string;

      /**
       * Row matches if column value deeply equals this value.
       */
      equals: unknown;
    }
  | {
      /**
       * Column path to compare.
       */
      column: string;

      /**
       * Row matches if column value does NOT deeply equal this value.
       */
      notEquals: unknown;
    }
  | {
      /** Column path to compare. */
      column: string;

      /**
       * Row matches if column value deeply equals any item in this array.
       */
      in: unknown[];
    }
  | {
      /**
       * Column path to compare.
       */
      column: string;

      /**
       * When true, matches non-null/non-undefined. When false, matches null/undefined.
       */
      exists: boolean;
    }
  | {
      /**
       * All sub-filters must match for the row to match.
       */
      and: SwarmFilter[];
    }
  | {
      /**
       * At least one sub-filter must match for the row to match.
       */
      or: SwarmFilter[];
    };

/**
 * A single dispatch unit for the executor.
 *
 * Represents one subagent call — either a single row's interpolated
 * prompt or a batch prompt covering multiple rows.
 */
export interface TaskSpec {
  /**
   * Row ID (single dispatch) or batch ID (batched dispatch).
   */
  id: string;

  /**
   * Fully interpolated prompt to send to the subagent.
   */
  prompt: string;

  /**
   * Name of the subagent type to dispatch to. When omitted, the
   * dispatch is a direct model call (invoke mode).
   */
  subagentType?: string;

  /**
   * Optional JSON Schema to constrain the subagent's response.
   */
  responseSchema?: Record<string, unknown>;

  /**
   * Dispatch mode for this task.
   *
   * - `"agent"` — Full agentic loop with tools and middleware.
   * - `"invoke"` — Direct model call, no tools or iteration.
   *
   * @default "agent"
   */
  mode?: "agent" | "invoke";
}

/**
 * Result of a single subagent dispatch.
 *
 * Returned by the executor in the same order as the input `TaskSpec[]`.
 */
export interface TaskResult {
  /**
   * Row ID or batch ID that this result corresponds to.
   */
  id: string;

  /**
   * Whether the dispatch succeeded or failed.
   */
  status: "completed" | "failed";

  /**
   * The subagent's response string (present when `status` is `"completed"`).
   */
  result?: string;

  /**
   * Error message (present when `status` is `"failed"`).
   */
  error?: string;
}

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Needs Node.js

The author says: Requires @langchain/quickjs code interpreter with swarm_task PTC tool

Security

What our scanner noticed· Low risk

  • Scripts that your agent may run

This skill includes scripts that your agent may run. Read them before you use it.

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Published by
LangChain
Publisher
LangChain (Official)
Source
langchain-ai/langchain-skills · config/skills/swarm · commit 126c384
Licence
MIT License (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by LangChain.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download swarm.zip