Skip to content

Impeccable

Design guidance and practical commands that help AI coding agents shape, critique, audit, and refine frontend experiences.

Analysis revision ae388acGitHub snapshot Aug 12, 2026Metrics are a dated GitHub snapshot and are not used as a quality score.

Impeccable turns design into a system of authorities and rendered evidence. It is not a collection of prompts that merely asks an agent to make a page look better.

A session starts by resolving PRODUCT.md, DESIGN.md, the matching surface brief, and platform guidance. It then loads exactly one command playbook, and loads a separate craft floor immediately before UI edits. Deterministic detector rules and live browser variants supply evidence that model-only critique cannot.

The authority stack behind each UI change

Each layer answers a different question. Keeping them separate lets a redesign replace visual direction without rewriting product truth, and lets mechanical checks support rather than impersonate design judgment.

Product truth

PRODUCT.md

Users, jobs, evidence, operating context, constraints, and accessibility stay durable.

Visual truth

DESIGN.md + surface brief

The product-wide visual world and the decisions unique to one surface remain distinct.

One judgment

Command playbook

critique, audit, polish, harden, and other commands load only the method needed for the current task.

Before editing

Craft floor

A separate quality floor constrains typography, contrast, spacing, iconography, and common design shortcuts.

Rendered evidence

Detector + live browser

Deterministic rules catch mechanical defects; live mode compares real rendered alternatives.

EvidenceWhy ImpeccableCommands and examplesLive browser iteration

Where the method is more specific than “good design”

Product truth and visual direction stay separate

A redesign may replace the visual world while preserving users, product purpose, evidence, content, and constraints recorded as product truth.

Commands own different judgments

critique evaluates experience quality, audit checks technical quality, polish closes shipping gaps, and harden covers errors, i18n, overflow, and edge cases.

Tool-backed rendered evidence

Detectors catch mechanical defects, while live mode operates a real preview with selected elements, alternatives, annotations, and accept or reject actions.

Choose the command by the judgment you need

  1. Run impeccable init before other commands so later work has real brand and product context.

  2. Pair a command with a concrete target, such as audit the blog, polish settings, or harden checkout.

  3. Use critique for experience judgment and audit for deterministic technical checks; do not treat them as interchangeable.

Adopt the method without leaking private context

Use it for

Landing pages, portfolios, application UI, and dashboards that need shaping, UX critique, technical audit, or final polish.

Review before sharing

Installation can add project hooks. Codex users should review and approve them, and may need to approve again after hook definitions change.

Sources reviewed

ae388ac

Navigation

Type to search…

↑↓ navigate↵ selectEsc close