Wink Pings

AI interfaces finally get a designer to review them: Impeccable comes to Computer, free through the end of October

Computer has listed the open-source skill Impeccable, which specifically checks layout, typography, contrast, and accessibility. Until October 31, every user gets 5,000 credits free. It uses 62 detection rules and 24 commands to fix the generic feel of AI interfaces, but whether taste can be outsourced was hotly debated in the replies.

Computer's skill marketplace now has a skill called Impeccable. It was built by Paul Bakaus, is open source, and is meant to review interfaces and fix the issues a designer would point out: layout, typography, contrast, and accessibility. Computer officially says that until October 31, tasks using /impeccable are free, with up to 5,000 credits per user.

The interesting part of this news is not the closed loop of AI fixing AI-generated interfaces. Hekate added the key details in a reply: Impeccable is Apache-2.0 on GitHub, with 86k+ stars; the README lists 24 commands, from audit and critique to polish; it also has 62 deterministic detection rules and can run both as a CLI and as a browser extension, with no LLM and no API key required. In other words, it does not rely on a model improvising taste on the spot; at least some issues are determined directly by rules. This design explains why it is well suited to being a review pass.

When AI coding agents build pages, the first version of the layout is usually not wildly off. John Kovacs's observation is more specific: layout is easy to get roughly right; what really goes off track is type scale and contrast, and it eventually slides toward a generic feel, while users rarely write these things into the prompt in advance. So the interface does not look bad, but it looks like the same template. Purple gradients became a target for complaints: Ricardo said he hopes not every page will have a purple gradient, and Hongji Lee said directly that the purple gradient hero section can rest in peace.

Alek's self-deprecation is more accurate: his interface does not look like AI slop, it looks like his own slop, which is worse. Asia Gigi's cold water is also worth including: taste cannot be packed into an open-source skill; if there is no aesthetic judgment at the foundation, de-slopping will only produce carefully polished beige. What Impeccable can fix are detectable, describable design problems; it cannot decide brand personality for people.

Computer's usage is straightforward: call /impeccable within a task and let it review and fix after generation. Daniel Priscu thinks a review that can directly submit fixes saves a round of back-and-forth with a designer. Charlie Grant says open-sourcing a de-slop skill is the right move; the default AI look will disappear only when fixes spread as fast as generation. Will also points out that making it a nameable skill makes sense, because "make it not look like AI slop" is already the input many people use, and now it has become a command.

There are still several questions the official tweet did not answer. Moriarty asked how many 5,000 credits it takes to run /impeccable once on a real app screen; 5,000 could be three runs or fifty. Abhi asked how well it works when an existing interface already looks like slop. Danylo asked whether it handles motion; static screens have gotten much better this year, but transitions still look generated. Harshit asked whether it can adapt to an existing design system without flattening its personality. Before using it, it is best to run it on a real page once and see what it changes and what it does not.

Treating Impeccable as aesthetic outsourcing will lead to disappointment. It is more like a cheap inspection layer: it goes over easily missed details such as layout, typography, contrast, and accessibility before delivery. For indie developers and small teams, this layer of checking is valuable. For large teams, the real test is whether it can respect an existing design system. It is free until October 31, with 5,000 credits, so it is worth trying on an old page. Do not expect it to install taste; first see whether it can leave a few fewer purple gradients.

Original: Aravind Srinivas tweet
Quote: Computer tweet

发布时间: 2026-10-25 13:41