How we hand off AI coding tasks without sharing accounts

AI coding agents are increasingly part of team workflows, but their work is often still coordinated through private chats and copied prompts.

We ran into a simple failure mode: one developer would prepare a detailed task, begin working with an agent, and then hit a subscription limit. Another teammate might have capacity, but transferring the work meant manually reconstructing the prompt, constraints, context, and expected result.

We built a Wagglet workflow around the task rather than the provider account.

The process is:

1. The developer asks Cursor to publish the exact prompt.
2. The Wagglet plugin creates a normal team draft.
3. The developer reviews and publishes it.
4. A coworker claims the task.
5. That coworker runs the prepared instructions in their own agent session.
6. The result is delivered through the original task.

No AI provider account, subscription, API key, or credential is shared. Each person continues using their own tools.

Under the hood, the Cursor plugin combines an MCP connection with a focused delegation skill. The task remains an ordinary Wagglet task, so it can use the same projects, permissions, discussion, and delivery process as other work.

This is still an evolving workflow, but it has made one distinction much clearer for us: teams should transfer work, not access.

Wagglet: https://wagglet.com/

Cursor plugin:
https://github.com/azakhary/wagglet-cursor-plugin