Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/coretracker/agentswarm/llms.txt

Use this file to discover all available pages before exploring further.

Sequences let you chain multiple agent prompts into a single automated workflow. Instead of manually sending follow-up prompts after each agent run, you define all the steps upfront, optionally add template variables to make the sequence reusable, and choose whether the agent should apply each checkpoint automatically or pause for your approval between steps. AgentSwarm then orchestrates the full run from start to finish.

​
Sequence Structure

A sequence has a name, an execution mode, an ordered list of steps, and an optional list of variables.
​
name
string
required
Display name for the sequence. Maximum 120 characters.
​
executionMode
string
default:"auto_apply_changes"
Controls how checkpoints between steps are handled. Either auto_apply_changes or approve_before_continuing. See Execution Modes below.
​
steps
array
required
Ordered list of steps to execute. Minimum 1 step, maximum 100 steps. Each step is either an inline prompt or a reference to a saved snippet.
​
variables
array
Template variables that can be substituted into step prompts at run time. Maximum 100 variables. See Variables below.

​
Step Types

Each step in a sequence has a type field that determines where its prompt content comes from.
An inline step contains the prompt text directly. The prompt field is required and must not be empty. Maximum prompt length is 20,000 characters.
{
  "id": "step_1",
  "type": "inline",
  "prompt": "Refactor the {{component_name}} component to use TypeScript strict mode."
}
A snippet step references a saved snippet by its ID. The snippet’s content is resolved at sequence execution time, so updating the snippet automatically affects all sequences that reference it. The snippetId field is required.
{
  "id": "step_2",
  "type": "snippet",
  "snippetId": "snp_abc123"
}

​
Variables

Variables make sequences reusable across different contexts. Define variables on the sequence and reference them in step prompts using {{variable_name}} syntax. When a sequence is run, the caller supplies values for each variable, which are substituted into the prompts before execution.
​
name
string
required
Variable identifier. Must match the pattern /^[A-Za-z_][A-Za-z0-9_]*$/. Maximum 128 characters. Must be unique within the sequence.
​
type
string
required
Either text (single-line input) or multiline (multi-line input). text variables must not have a default value containing line breaks.
​
title
string
Human-readable label shown in the sequence run form. Maximum 200 characters.
​
description
string
Explanatory text shown alongside the variable input. Maximum 200 characters.
​
defaultValue
string
Pre-filled value used when the caller does not supply a value. Maximum 2,000 characters.

​
Variable Syntax

Reference a variable in any step prompt by wrapping its name in double curly braces: 
Add comprehensive tests for the {{component_name}} component located at {{file_path}}.
At run time, {{component_name}} and {{file_path}} are replaced with the values provided by the caller.

​
Execution Modes

auto_apply_changes

After each step’s agent run completes, any pending checkpoint is applied automatically and the next step starts immediately. The full sequence runs end-to-end without human intervention.If a checkpoint cannot be auto-applied (for example, because the task is in an unexpected state), the sequence pauses and waits for manual resolution before recovering.

approve_before_continuing

After each step completes, the sequence pauses with status waiting_for_approval. A human must explicitly approve before the next step is allowed to run. This lets you inspect intermediate changes between steps before the agent continues.

​
Creating a Sequence

1

Open Sequences

Navigate to Sequences in the main navigation and click New Sequence.
2

Set name and execution mode

Give the sequence a name (up to 120 characters) and choose whether steps should auto-apply or require approval between them.
3

Add steps

Add one or more steps. For each step, choose inline and write a prompt, or choose snippet and select a saved snippet. Step IDs are assigned automatically if not provided.
4

Add variables (optional)

Define template variables if you want the sequence to be parameterized. Variable names must match /^[A-Za-z_][A-Za-z0-9_]*$/.
5

Save

Click Save. The sequence is now available when creating tasks.

​
Running a Sequence

To run a sequence, create a task with task_source set to sequence and provide the sequence_id. If the sequence defines variables, pass their values in sequence_variables. 
{
  "title": "Refactor Button component",
  "repoId": "repo_xyz",
  "task_source": "sequence",
  "sequence_id": "seq_abc123",
  "provider": "claude",
  "providerProfile": "high",
  "baseBranch": "main",
  "branchStrategy": "feature_branch",
  "sequence_variables": {
    "component_name": "Button",
    "file_path": "src/components/Button.tsx"
  }
}
AgentSwarm creates a sequence run record that tracks the state of each step. You can retrieve the current run state at any time from the task detail page.

​
Sequence Run Statuses

StatusDescription
runningA step is currently executing.
waiting_for_approvalapprove_before_continuing mode: paused between steps, awaiting human approval.
waiting_for_checkpoint_resolutionAuto-apply mode: a checkpoint is pending and must be resolved before the sequence can continue.
succeededAll steps completed successfully.
failedA step failed; the sequence stopped.

​
Example Sequence

The following sequence adds TypeScript strict types to a component, then generates tests for it, using variables so it can be reused for any component. 
{
  "name": "Type + test component",
  "executionMode": "approve_before_continuing",
  "variables": [
    {
      "name": "component_name",
      "type": "text",
      "title": "Component Name",
      "description": "The name of the React component to update.",
      "defaultValue": ""
    },
    {
      "name": "file_path",
      "type": "text",
      "title": "File Path",
      "description": "Repo-relative path to the component file.",
      "defaultValue": ""
    }
  ],
  "steps": [
    {
      "id": "step_1",
      "type": "inline",
      "prompt": "Update {{file_path}} to add full TypeScript strict types to the {{component_name}} component. Do not change runtime behaviour."
    },
    {
      "id": "step_2",
      "type": "inline",
      "prompt": "Write a comprehensive Vitest test suite for the {{component_name}} component in {{file_path}}. Cover all props, edge cases, and accessibility."
    }
  ]
}

Build docs developers (and LLMs) love

Get started for freeTalk to us