Perplexity has added Paul Bakaus's Impeccable to Computer's skills marketplace. Its official October 24, 2026 post describes reviews of layout, typography, contrast and accessibility. The company advertises a promotion through October 31 for tasks using the skill, capped at 5,000 credits per user. This is an offer for specified tasks, rather than unlimited free computing or a guarantee of design quality.
What the package adds to a coding agent
The public Impeccable repository describes a shared design vocabulary, commands for reviewing and refining interfaces, and separate deterministic checks. Its code uses the Apache 2.0 license. Some checks run through a CLI and browser extension without a language model; judgment-based critique depends on the host agent. An automatically detected issue and a designer's assessment therefore represent different kinds of evidence.
The integration is interesting because it directs an agent toward the quality of an interface it has already built. Generating a page and reviewing that page are separate tasks. The former can end with functioning code, while the latter asks whether important information is visible, actions are understandable, and the layout supports the product's actual purpose.
An open skill and access to Computer
Perplexity's help center distinguishes credits spent in Computer from searches in Ask. Computer credit access depends on eligible subscriptions and account conditions, while a task's consumption depends on its complexity. Impeccable's open license therefore does not make every environment hosting it free to use. We have not checked the promotion's availability inside an individual paid account.
For a team already using a coding agent, the practical question is where the review happens and how changes are accepted. A recommendation might suit a landing page but be inappropriate for a dense administration screen. Acceptance criteria need to follow the user's task, rather than applying one aesthetic preference to every product.
This distinction also changes what a useful review request looks like. Asking for a more attractive interface leaves the success condition open. Asking the agent to examine a particular task, preserve the information users need, and explain each proposed alteration creates a result that a product owner can actually assess.
Accessibility requires its own evaluation
W3C WAI offers preliminary checks covering headings, contrast, text enlargement and keyboard operation with visible focus. It explicitly cautions that an initial review is not a complete accessibility evaluation. This provides an independent reference for interpreting design-tool promises: an attractive screen and an accessible workflow need separate checks.
Our suggested evaluation begins with one real screen and a written description of its purpose. For a registration form, for example, priorities include clear required-field labels, understandable error messages and a path that can be completed using a keyboard. After the AI review, those actions should be repeated in a browser, and visual changes compared at narrow and wide window sizes.
That is an editorial test proposal, not an Impeccable experiment conducted by SIExpo. The development is another concrete way to guide an agent toward design review. Its value for a particular project will be demonstrated by problems resolved and usability preserved, rather than by the mere fact that a page passed through a polishing command.
*AI illustration: a conceptual editorial image, not a documentary photograph or an actual product screen.*
*Editorial verification: 2026-10-25T17:51:23.544949+00:00.*
