Impeccable gives coding agents a design vocabulary: 62 deterministic rules catch AI-slop interfaces before they ship
Synopsis
Impeccable turns design direction into commands a coding agent can call, so vague requests such as bolder, quieter or polish land on a specific target. It also supplies a review command that scores design against usability heuristics and a slop detector built from 62 deterministic rules, with no LLM or API key, that flags interface tells recurring across models. It is strongest on existing pages and components; building new pages is the harder mode.
Interpretation
A coding agent gains a set of nameable design moves instead of guessing from a single request to make it pop. With no other guidance, agents across models and projects keep producing the same defaults, such as a cream page with a thin serif and pill buttons, or cards nested in cards with gray text on colored backgrounds. With the same model and the same brief, the build without Impeccable is that tasteful default, while the build with it organizes the page around one picture of the house at night seen from the water.
Commands split by target: bolder loads a playbook written for that one move and points it at one target, while critique scores design against usability heuristics and ranks what to fix first. audit checks accessibility, performance, theming and responsive behavior and ranks findings from P0 to P3; the slop detector uses 62 deterministic rules to flag the tells above plus quality problems such as cramped padding and skipped headings, with no LLM and no API key. The workflow runs init once per project, which reads the code, asks only what it cannot infer and writes PRODUCT.md for later commands to read, then names a command and a target, such as critique on the pricing cards.
Perspective
It states it is strongest on interfaces that already exist, by pointing at a page or component and iterating. It also builds new pages, described as the harder mode. Live mode edits a local checkout only and should not be injected into a deployed site or used by weakening a production CSP.
The 62 rules flag the listed tells and quality problems, and the detector itself calls no LLM. Applying copy edits in live mode runs the optional impeccable:manual-edit-validate script from package.json with the user's permissions; if no session has polled in the last minute, it runs the codex CLI from PATH, or claude when codex is missing, with approvals, permission checks and sandboxing off.