Sub-Agents

LongerAgent can spawn parallel sub-agents within a session. Sub-agents run concurrently, each with their own context and tool access, and report back to the main agent when done.

Spawning Sub-Agents

There are two ways to spawn sub-agents:

From Chat

Simply tell the agent what you want parallelized:

You: investigate the auth module and the payment module at the same time

The agent will use the spawn_agent tool to create sub-agents for each task.

From YAML Call Files

For more control, define tasks in a YAML call file:

# tasks.yaml
tasks:
  - name: research
    template: explorer
    prompt: "Investigate how authentication works in this codebase"
  - name: refactor
    template: executor
    prompt: "Rename all legacy API endpoints to v2"

The agent writes this file to its session artifacts directory and calls spawn_agent with the filename.

Templates

Each sub-agent is created from a template that defines its system prompt and available tools. Three built-in templates are available:

main

The full-capability template. Has access to all 15 built-in tools plus orchestration tools. Use this when the sub-agent needs to do everything the main agent can do.

explorer

A read-only template. Has access to file reading, searching, and analysis tools, but cannot edit files or run destructive commands. Use this for investigation, code review, and research tasks.

executor

A task-focused template. Has file editing and shell access but is scoped to completing a specific task. Use this for implementation work like refactoring, renaming, or applying changes.

Orchestration Tools

The main agent has several tools for managing sub-agents:

Tool	Description
spawn_agent	Spawn sub-agents from a YAML call file
kill_agent	Kill one or more running sub-agents by ID
check_status	Check the status of running sub-agents
wait	Wait for specific sub-agents to complete

How It Works

1. The main agent creates a YAML call file defining the tasks.
2. spawn_agent reads the file and creates a sub-agent for each task.
3. Sub-agents run concurrently, each in their own agent loop with their own context.
4. When a sub-agent finishes, its results are delivered back to the main agent.
5. The main agent synthesizes the results and continues.

Sub-agents share the same model and provider as the main agent. They have access to the same filesystem and project, but maintain separate conversation contexts.

Custom Templates

You can override the built-in templates by placing custom versions in ~/.longeragent/agent_templates/:

~/.longeragent/agent_templates/
├── main/
│   └── system_prompt.md
├── explorer/
│   └── system_prompt.md
└── executor/
    └── system_prompt.md

Custom templates follow the same format as the built-in ones. The system prompt is a markdown file that defines the agent's behavior, available tools, and constraints.

Practical Tips

• Use explorer for investigation. When you want the agent to look at something without changing it, the explorer template is safer and faster.
• Use executor for scoped changes. Give it a clear, well-defined task. The executor template is optimized for completing a single task and reporting back.
• Let the main agent coordinate. The main agent decides when to spawn, how to split work, and how to synthesize results. You just describe what you want done.
• Check on progress. You can ask the main agent about sub-agent status at any time, and it will use check_status to report back.