Keep Linear for product work. Add Wagglet when AI work changes hands.

Assignment and delegation

Linear lets a workspace member own an issue and delegate it to an installed agent. The delegate can be changed or removed, and the activity stays visible on the issue. See Linear's assignment and delegation documentation.

Agent context

Linear documents workspace- and team-level agent guidance, issue context, comments, and agent activity. It also lets people open issues in Claude Code, Codex, Cursor, and other configured coding tools with issue context and an optional custom prompt. See Linear's AI Agents and Code & reviews guides.

Linear coding sessions

Linear also documents collaborative coding sessions tied to an issue, powered by a workspace-selected Claude Code or Codex model and paid from workspace AI credits. Teammates can follow and steer the shared session. See Linear's coding sessions guide and its AI Credits documentation.

1. 01

   Keep the issue in Linear

   ENG-2844 remains the source for product priority, status, discussion, and the final pull request. Nothing is imported or synchronized into Wagglet.

2. 02

   Prepare the Wagglet task manually

   Create a Wagglet task that references ENG-2844. Put repository, branch, retry cases, scope, tests, and stop conditions in LLM instructions. Put the checks that require judgment in Human instructions.

3. 03

   Let the teammate claim and run it

   The teammate claims the task, starts a fresh Codex session with their own account and authorized repository access, supervises the run, and follows the Human instructions.

4. 04

   Return evidence to both records

   The runner records the outcome, branch, tests, warnings, and delivery evidence in Wagglet. A person then posts the relevant result or pull request back to ENG-2844 and updates Linear manually.

Can Linear transfer an in-progress AI task without sharing the original AI account?

Linear can reassign an issue and run workspace-funded coding sessions, but its current documentation does not describe exporting a personal-agent session as a teammate-to-teammate work package. Wagglet moves the prepared task to a fresh session in the recipient's own account.

Does Linear separate instructions for the AI agent and the supervising human?

Linear documents issue content, agent guidance, environment guidance, and custom prompts. Wagglet additionally provides separate per-handoff LLM instructions and Human instructions for the two audiences.

Does Linear support teammate-to-teammate AI task handoffs?

Linear supports issue reassignment and delegation to installed agents. Wagglet adds a claim, supervised run, separate delivery prompt, and independent review lifecycle for work executed through a teammate's own agent account.

How can a team redistribute AI execution when one developer reaches a provider limit?

Keep the issue in Linear, prepare the bounded execution task in Wagglet, and let another qualified teammate run it through their own authorized account. Return the outcome to Linear manually; no provider allowance is pooled or transferred.

For the underlying category, read AI work handoff. For the exact claim-to-delivery procedure, use the Task Handoff guide.

• Manual alongside workflow. Wagglet does not read from or write to Linear today. A person chooses what context may be copied and what result should be posted back.
• Fresh recipient-owned session. Wagglet transfers the prepared task, not a live Claude Code, Codex, Cursor, or Linear coding session.
• Not AI subscription sharing or a quota workaround. The runner uses their own authorized provider account and remains responsible for access, judgment, and the required checks.
• Recorded outcome and delivery evidence. Delivery reports what happened. It does not update Linear, approve a pull request, merge code, or deploy.

Adding Wagglet would create another record and a manual return step without solving a handoff problem. Introduce it only when transferable AI work between people, separate LLM instructions and Human instructions, or a claim-and-delivery record is actually needed.