AI & Design
Impeccable: Give Your AI Coding Agent a Better Design Vocabulary
Turn vague requests for better UI into focused design work, with Impeccable's critique, polish, and audit vocabulary as a starting point.
Rommel Clarino 4 min read
At a glance
Impeccable gives coding agents named design operations and provides a deterministic detector for interface problems. Its value is easier to judge on one existing component: name the user problem, request a focused change, and inspect the result against explicit constraints.
Better UI starts with a more specific request
An interface can have clean code and still leave users wondering what to do. Asking a coding agent to make it look better does not explain whether the problem is a weak heading, competing actions, cramped spacing, or an unclear error message. The agent has to invent the goal before attempting the work.
Impeccable offers a vocabulary for directing that work. Its repository documents commands such as critique, polish, audit, and harden, alongside a deterministic detector for recurring interface issues. It appeared on GitHub Trending during the October 31, 2026 research snapshot. That attention makes it worth examining; it is not evidence that its changes improve every interface.
Give one component a job
Consider a hypothetical pricing card. Its title, price, description, features, and primary action all carry similar visual weight. The problem is not a shortage of decoration. A reader needs to identify the offer and its next step quickly.
A useful request names that problem and preserves the constraints. The example below is an illustrative prompt, not a measured result from running the tool. It is specific enough that a reviewer can later explain whether the edit helped.
Review this pricing card for hierarchy and action clarity.
The reader should identify the plan, price, and main action in that order.
Keep the existing copy, brand colors, and feature set.
Explain the three most useful changes before editing.
After editing, check narrow screens, keyboard focus, and long plan names.Use different passes for different decisions
The repository distinguishes design critique from technical auditing and final polish. Those are useful boundaries for any review process. First decide whether the component communicates the right thing. Then inspect whether people can operate it. Finally, adjust details that make the design feel consistent.
For the pricing example, a critique might question the order of information. A technical check might uncover a button without a visible focus state. A final visual review might align spacing across cards. Combining all three into one request makes it harder to see which change solved the original problem.
Automation needs an acceptance criterion
A detector can identify patterns it knows about. It cannot decide whether a particular customer understands an offer or trusts a checkout. Treat findings as inputs to a review, and keep a written reason for accepting or declining consequential suggestions.
Compare the original and edited component at the same viewport and with the same content. Include unusually long labels, missing optional content, and an error state. A polished default screenshot is weak evidence if the layout falls apart as soon as a real name or translated sentence appears.
- Define the intended reading order before changing the layout.
- Check keyboard operation and visible focus after edits.
- Review both narrow and wide layouts with realistic content.
- Keep changes that solve the stated problem and explain why.
Judge the workflow on a small, visible result
The project says its standalone deterministic detector requires no LLM or API key. Agent-driven editing is a separate activity and can still involve model or subscription costs. Keep those costs distinct when comparing the workflow with another design tool.
Start with a component you already understand. Give the agent a concrete user problem, review a bounded edit, and preserve the before-and-after evidence. The useful outcome is a clearer interface and a repeatable way to ask for the next improvement.