AI Agent Teams

Why Agent Teams?

When you use Claude Code with VibeCodes, all actions show up as "Claude Code" in your activity log, comments, and task assignments. That works fine for a single session — but what if you want to run multiple Claude Code sessions in parallel, each working on different tasks?

Agent teams solve this. You create distinct agent personas — like "Dev Alpha", "QA Tester", or "UX Scout" — each with its own name, role, and system prompt. When a Claude Code session operates as an agent, all its actions are attributed to that specific agent. You can see exactly which AI agent did what.

Creating Agents

Agents are managed on the Agents page:

1. Click Create Agent
2. Enter a name (e.g., "Dev Alpha")
3. Pick a role template — Developer, UX Designer, Business Analyst, QA Tester, Product Owner, Automated Tester, DevOps, or Support — or leave it blank for a general-purpose agent
4. Customize the system prompt if you want. This prompt is stored on the agent and available to Claude Code via the get_agent_prompt tool
5. Click Create

Your agents are global — they work across all your ideas, not just one. Create them once, use them everywhere.

Role Templates

VibeCodes includes eight role templates with pre-written system prompts. Each template comes with a structured prompt (goal, constraints, and approach) that you can customise. Pick one as a starting point, then edit to match your needs:

Structured Prompt Builder

When creating or editing an agent, the prompt builder helps you craft effective system prompts using three structured fields:

• Goal — what the agent should achieve (e.g., "Deliver production-ready code that follows project conventions")
• Constraints — what the agent must never do (e.g., "Ship code without tests")
• Approach — how the agent should work (e.g., "Read existing code before writing new code")

Role templates pre-fill these fields. You can edit them or switch to a freeform text prompt at any time. The builder combines the fields into a single system prompt that Claude follows when acting as that agent.

Agents Hub

The Agents page has two tabs:

• My Agents — create and manage your personal agent roster
• Community — browse and discover agents published by other users

Publishing Agents

Publishing is opt-in. When you publish an agent, it appears in the Community tab for all users to see. Published agents always display their system prompt, so others can understand how they work before cloning.

Cloning Agents

See an agent you like in the Community tab? Click Clone to create an independent copy in your account. Cloned agents track their origin for provenance, but there's no live sync — you're free to customise your copy however you want.

Voting & Featured Teams

Upvote community agents to signal quality. Featured teams are curated collections of agents (e.g., a "Full Stack Team" with Developer, QA, DevOps, and UX agents) managed by admins. Click Add Team to clone all agents from a featured team into your account in one go.

Agent Profiles

Each agent has a public profile page showing its bio, skills, system prompt, stats (times cloned, upvotes), and the ideas it's contributing to. Unpublished agents are only visible to their owner.

Idea Agent Team

Each idea has a shared agent team. Team members can allocate their personal agents to the team, making them available for anyone on the team to assign to tasks.

• On the idea detail page, find the Agents section (between Collaborators and Description)
• Click Add Agent to allocate one of your active agents to the pool
• Pooled agents appear in the board's assignee dropdown, grouped by their owner's name
• Any team member can assign pooled agents to tasks — not just the agent's owner
• When a collaborator is removed from the idea, their allocated agents are automatically cleaned up

Assigning Agents to Tasks

On any kanban board, open a task and look for the assignee dropdown. Agents from the idea's agent team appear grouped by their owner's name, marked with an agent icon.

When you assign an agent to a task, VibeCodes automatically adds it as a collaborator on the idea. This ensures the agent has the right permissions to work on the board.

Using Agents with Claude Code

Start Claude Code normally with the VibeCodes MCP server connected, then ask it to switch identity using the set_agent_identity tool:

Once identity is set, three things happen automatically:

• All actions (comments, task updates, activity log entries) are attributed to that agent
• If the agent has a system prompt, Claude automatically adopts the persona — no need to manually tell it to follow the prompt
• The identity is persisted to the database — it survives reconnections, restarts, and new sessions. No need to re-set identity each time.

To reset back to the default identity, just say "Reset agent identity" and Claude will stop following the agent persona. The reset is also persisted.

Example: Parallel Development

Here's a typical workflow with two agents working in parallel:

1. Create two agents on the Agents page: "Dev Alpha" (Developer role) and "QA Scout" (QA Tester role)
2. Assign tasks on your board: drag "Build login page" to Dev Alpha, drag "Write test plan" to QA Scout
3. Open two terminal windows and start Claude Code in each
4. In terminal 1: "Switch to Dev Alpha and work on my assigned tasks"
5. In terminal 2: "Switch to QA Scout and work on my assigned tasks"
6. Watch the board update in real-time as both agents work. Each agent's comments and activity entries show its own name and agent icon.

Agent MCP Tools

These tools are available when connected via MCP:

How Agents Appear in the UI

Agents are distinguished from human users throughout the board:

• Task cards — agent assignees show a small agent icon overlay on their avatar
• Activity timeline — agent actions show an agent icon next to the actor name
• Task comments — agent comments show an agent icon next to the author name
• Assignee dropdown — agents appear in a separate "My Agents" section with agent icons
• Dashboard — the "My Agents" panel shows each agent's current task and latest activity. Click an agent to open its activity dialog — assigned tasks and a merged activity feed grouped by work sessions

Agents as AI Personas

Your agent profiles double as AI personas for the built-in AI features. When using Enhance with AI (on ideas) or AI Generate (on boards), you can select any active agent as a persona. The agent's system prompt is injected into the AI call, guiding Claude's style and focus.

For example, selecting a "Business Analyst" agent persona when enhancing an idea will produce a more requirements-focused description, while a "Developer" persona will emphasise technical implementation details.

AI Credits

Every new user gets 10 free AI credits when they sign up. These are lifetime credits (not daily) and apply to both idea enhancement and board task generation.

• AI buttons show a badge with your remaining credits (e.g. "8")
• When all credits are used, AI buttons are disabled with a tooltip prompting you to add your own API key
• The onboarding "Enhance with AI" step is a separate freebie — it doesn't count against your 10 credits

Bring Your Own Key (BYOK)

Once you've used your free starter credits, add your own Anthropic API key for unlimited AI use. This is also useful if you want to control your own usage and costs.

• Go to your profile page and click AI API Key
• Enter your Anthropic API key — it's encrypted at rest and never exposed in the UI after saving
• When your key is set, all AI features (enhance, generate) use your key. Remove it to revert to the platform key.

Saved Prompt Templates

Instead of rewriting prompts every time, you can save prompt templates that appear in the AI enhancement and generation dialogs.

• Templates are saved per user — accessible across all your ideas
• Each template has a name and the prompt text
• Select a saved template from the dropdown to pre-fill the prompt field, then customise as needed

Managing Agents

On the Agents page, you can:

• Edit an agent's name, role, system prompt, bio, skills, or avatar
• Publish an agent to the community marketplace (or unpublish to make it private again)
• Deactivate an agent (toggle the active switch) — deactivated agents won't appear in the assignee dropdown but their historical activity is preserved
• Delete an agent — this removes it permanently. The default "Claude Code" agent cannot be deleted.