Wagglet
Hand off a prepared AI coding task to a teammate without sharing your AI account, API key or subscription.
Documentation
One claimed task Task Handoff Copy exact task instructions into a fresh agent session, refresh bounded context while the claim is active, then use a separate later delivery capability.Read guide → One person · one Team MCP Workspace Workflow Connect an agent to current authorized Wagglet views and explicit actions through a named, revocable MCP Connection.Read guide → Product flow How Wagglet works See how an author prepares the work, a teammate claims and supervises it, and a reviewer records the verdict.Read guide → Trust boundary Security and privacy Understand what each credential can expose, what Wagglet rechecks, and where the human remains responsible.Read guide →
Task Handoff
Rooted in one current claim. A read-only capability covers that task and, when linked, permitted current Story context. It ends with the claim and expires after at most seven days.
MCP Workspace Workflow
Rooted in one named MCP Connection for one person and one Team. It has no automatic expiry, but Team access, membership, roles, permissions, and revocation are checked again on every request.
Task Handoff delivery
The starting read credential cannot deliver. The runner later copies a separate, one-time delivery prompt into the same agent session.
MCP mutations
Explicit Wagglet actions use current revisions and a fresh operation id. There is no generic status override and no database-shaped interface.
[!note] Note Raw agent guides
The human pages explain the product boundary. Agents can read the canonical raw Task Handoff Skill and MCP Workspace Skill. These public files contain procedure, not a user's credential or private work.
Start with a small synthetic or low-risk task. Confirm that the runner may access the repository and data, knows what to inspect, and uses their own authenticated agent. Keep secrets out of task text. Use Human-only work when an agent should not receive the task.