Wagglet

Hey👋

Wagglet came out of a problem that kept happening on our own team: a

couple of people would hit their weekly Claude Code limits by Wednesday,

while other people's allowance sat almost completely unused. The work was

there. The capacity was there. They were just in different people's

accounts.

The obvious fix is to share the login, and it's a bad fix. You lose

attribution — every prompt and every generated diff is now logged against

one identity, so when you need to know who told the agent to touch the

payments module, the data doesn't exist. You inherit the other person's

blast radius. And you can't revoke cleanly, because a credential that

lives in two heads stays live long after anyone needed it.

So Wagglet moves the work instead of the access.

An Author writes a task spec — outcome, context, limits, evidence. A

Runner claims it based on the capacity they actually have. They supervise

their own agent through it. The Agent delivers a record of what changed.

A Reviewer accepts, sends back, or reopens. The Runner gets a read-only

credential scoped to exactly that one claim, and it expires within seven

days rather than living forever.

The thing that surprised us most: writing the spec is the hard part, not

running it. A task description that works for a human colleague falls

apart when handed to someone else's agent, because your colleague fills

gaps from context and asks you when something doesn't make sense — the

agent does neither, and will confidently do the wrong thing at speed. The

section people skip is "limits", and it's the one that saves you.

Second surprise: the specs kept exposing knowledge that had never been

written down. Every time someone couldn't finish the context section, it

was because the requirement only existed as an assumption shared between

two people.

Docs are here if you want the mechanics: https://wagglet.com/docs

And we wrote up the task-design side properly here:

https://wagglet.com/blog/dual-prompt-human-agent-task-design

Built by the team at Rockbite Games. Genuinely keen to hear how other

teams are handling the capacity-distribution problem — I don't think

anyone has a clean answer yet.