Bottom Line: A pile of tiny, unglamorous AI utilities that solve a real neurological problem better than any $12/month productivity suite on the market — and then, almost defiantly, refuses to charge you for it. The gap between what goblin.tools does and what it could do is enormous, but what it does, it does with unusual clarity of purpose.
The Loop That Works
The genius of Magic ToDo is that it attacks the correct variable. Task paralysis is not a motivation deficit; it's an initiation deficit. The brain cannot find the entry point, so it bounces off the task entirely. Every conventional to-do app responds to this by making the task more visible — bolder, redder, higher in the list — which is roughly like treating a stutter by shouting.
Magic ToDo instead makes the task smaller until it's beneath the activation threshold. The spiciness slider is the mechanism, and it's brilliant for a reason that has nothing to do with AI: it hands the user control over their own granularity. Nobody knows what step size you need today except you, and that number changes hour to hour depending on medication, sleep, and mood. A static breakdown algorithm would be useless. A slider is honest about the variability. Pull it to four peppers on a bad day, one pepper on a good one. That's not a gimmick; that's a genuine insight about the user, encoded into a control.
The recursion matters too. Any step that's still too big gets its own breakdown. You can drill until you hit something with zero friction — "walk to the bedroom," "pick up the hamper" — and once you're moving, momentum does the rest. The tool doesn't need to carry you through the whole task. It only needs to get you past the first thirty seconds.
The Loop That Doesn't
Here's where the critique lands, and every reviewer who's used this thing seriously arrives at the same place: goblin.tools answers "what should I do?" but not "why can't I start?"
There is no timer. There is no guided, one-step-at-a-time execution mode. Once your task has exploded into fifteen micro-steps, you're handed the full list — all fifteen — and asked to just, you know, do them. For a user who was overwhelmed by one line of text ninety seconds ago, presenting fifteen is not obviously the win it looks like. The tool solves decomposition and then abandons you at execution, which is the second half of the same problem.
Worse, and this is the omission I can't get past: there is no saved history of past breakdowns. Think about that for a moment. This is a tool designed explicitly for people with working-memory deficits, and it forgets everything. You broke down "file taxes" into twelve steps last April. This April, you start from zero. You break down the same recurring chore every single week, paying the same small cognitive tax, and the app that exists specifically to reduce your cognitive load contributes nothing to that. The audience most likely to benefit from persistence is the audience least served by its absence. This isn't a missing "nice-to-have." It's a hole in the thesis.
There's no personalization either — the model doesn't learn that your four-pepper is someone else's two, or that you always forget the "gather materials" step. And there's no calendar or task-manager integration, so the output of the Estimator, which is genuinely useful time data, dies in the browser tab where it was born. It can't feed your calendar. It can't tell Todoist anything. Every goblin.tools session is a sealed room.
The Ecosystem Problem
The most frustrating criticism isn't even Skyhook's fault. goblin.tools has been swarmed by copycat apps with near-identical names charging monthly subscriptions for a worse version of a free thing, and some users have reported unclear trial pricing on those imposters. When your product is free and your clones aren't, discoverability becomes an attack surface. That the real app still holds 4.8 stars through this is a reasonable proxy for how much people like it.