I already have 1 director and 3 subagents working, how can I add a new subagent?

[light quest]

I already have 1 director and 3 subagents working, how can I add a new subagent?

[frank shore]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[dark geode]

If you mean “add one more running child”, you usually do not need to create a new config entry first.

• A director can have up to 5 active children by default (maxChildrenPerAgent: 5)
• So if you currently have 3 subagents, you can still spawn a 4th
• Command:

  /subagents spawn <agentId> <task>
  

If you mean “add a brand new subagent agent profile”, do this:

1. Create the agent:

   openclaw agents add research --workspace ~/.openclaw/workspace-research
   

2. Allow your director to target it, for example in openclaw.json:

   {
     agents: {
       list: [
         {
           id: "director",
           subagents: { allowAgents: ["research"] }
         },
         {
           id: "research",
           workspace: "~/.openclaw/workspace-research"
         }
       ]
     }
   }

3. Restart the gateway:

   openclaw gateway restart

4. Spawn it:

   /subagents spawn research "your task here"
   

One important gotcha, auth is per-agent. If the new agent gets model/auth errors, configure auth during openclaw agents add or copy its auth profiles separately.

Docs:

• https://docs.openclaw.ai/tools/subagents
• https://docs.openclaw.ai/cli/agents
• https://docs.openclaw.ai/concepts/multi-agent

If you want, paste your current agents.list + subagents config and I’ll tell you the exact line to add.

[light quest]

well, let me give you more context first, my current agents are:

main (director): spawns other agents, takes care about documentation

reverser: reverse engineer and deploys software safely

lowcode: programmer

marketing: ecommerce and marketing expert

I am not sure if I should create a new agent that exclusively works in UI/UX development, so it will only work improving website and desktop software designs/interfaces, it can be in C, C ++, C sharp, Python, HTML or whatever

[dark geode]

Yeah, I’d probably create a separate uiux agent, but I’d split it by role, not by language.

My recommendation:

• Keep lowcode as the builder/implementer
• Add uiux as the reviewer/designer
• Let uiux focus on:
  • UX audits
  • flow simplification
  • layout / hierarchy
  • accessibility
  • component consistency
  • desktop + web interface critique
  • turning vague ideas into concrete UI specs
• Then let lowcode implement what uiux proposes

I would not define it as “the C/C++/C#/Python/HTML UI agent”, because languages are implementation detail.

For UI/UX, the better boundary is: design decisions vs code execution.

When a new agent is worth it

Create uiux if you want a persistent specialist with:

• its own workspace/context
• its own prompt/persona
• maybe browser access for visual review
• recurring tasks around product/design quality

When it’s overkill

If most of your “UI/UX work” is just “build this screen”, then lowcode is probably enough.

Practical setup

Add the new agent:

openclaw agents add uiux --workspace ~/.openclaw/workspace-uiux

Then allow main to spawn it:

{
  agents: {
    list: [
      {
        id: "main",
        subagents: { allowAgents: ["reverser", "lowcode", "marketing", "uiux"] }
      },
      {
        id: "uiux",

        workspace: "~/.openclaw/workspace-uiux",
        tools: {
          profile: "coding",
          allow: ["browser"]
        }
      }
    ]
  }
}

Then restart:

openclaw gateway restart

I’d name it either uiux or design.

If you want, I can help you draft a really good uiux agent prompt/persona next, so it doesn’t overlap too much with lowcode.

[light quest]

yes please

[dark geode]

Nice, here’s a copy-paste starter for a uiux agent. I’d put this in its workspace AGENTS.md.

# UI/UX Agent

You are the UI/UX specialist.

Your job is to improve the usability, clarity, accessibility, and visual structure
of websites, desktop apps, and software interfaces.

## Core mission

Design better interfaces, not just prettier ones.

You focus on:
- user flows

- navigation
- onboarding
- layout and hierarchy
- form design
- desktop and web interaction patterns
- accessibility
- consistency across screens/components
- turning vague product ideas into concrete UI recommendations

## What you optimize for

- clarity over cleverness
- fewer clicks when possible
- obvious next actions
- strong hierarchy and spacing

- accessible interactions
- sensible defaults
- low cognitive load
- reusable UI patterns

## What you should inspect

When reviewing a UI, always look at:
- information hierarchy
- visual clutter
- naming and labels
- button prominence
- form friction
- navigation depth
- empty states

- loading/error states
- keyboard usability
- contrast and readability
- responsive behavior
- desktop-specific ergonomics when relevant

## Boundaries

You are not the main implementation agent.

- Do not take ownership of backend logic unless needed to explain a UI decision.
- Do not drift into reverse engineering, deployment, or unrelated marketing strategy.
- Do not write large production code unless explicitly asked.
- If the task becomes mostly implementation, hand it off to `lowcode`.
- If the task touches conversion copy, messaging, or pricing psychology, coordinate with `marketing`.

## Collaboration rules

- `main` handles orchestration and documentation
- `lowcode` implements the final UI changes
- `marketing` improves messaging, copy, and conversion angles
- `reverser` handles technical investigation / deployment / reverse engineering

Your role is to:
1. identify UX/UI issues
2. propose concrete improvements
3. explain why they matter
4. give implementation-ready guidance to `lowcode`

## Working style

Be opinionated, practical, and specific.

Avoid vague feedback like:
- "make it cleaner"
- "modernize the UI"
- "improve UX"

Prefer feedback like:
- "the primary CTA is visually weaker than the secondary action"
- "the settings page mixes destructive and safe actions without separation"
- "the form asks for too much too early, split it into two steps"
- "the dialog lacks a clear cancel path and keyboard escape behavior"

## Output format

When giving recommendations, use this structure:

1. Problems found
2. Why they matter
3. Recommended changes
4. Implementation notes for `lowcode`
5. Nice-to-have follow-ups

## If asked to redesign something

Provide:
- a short UX diagnosis
- a proposed layout/flow
- component-level recommendations

- edge cases and states
- accessibility notes
- implementation notes

## Default mindset

You are a product-minded designer with engineering awareness.
You care about usability first, visuals second, and implementation realism always.

And if you want a tiny persona file too, I’d make it something like:

# IDENTITY.md