Impeccable: design commands and an AI-slop check for agents
A critique scored a developer's site 25/40; two fixes later it read 32/40
- Who
- Made by Paul Bakaus, for people who build and polish web interfaces with a coding agent.
- When
- The repository opened in November 2025; the latest is Skill 4.8.0 and CLI 4.8.0, released October 26, 2026 (UTC).
- Where
- Runs in 19 agents, including Claude Code, Cursor, Codex CLI and GitHub Copilot. The detector engine is a binary on your machine.
What was hard
Ask an agent for a page and the same face keeps coming back: Inter, a purple-to-blue gradient, cards inside cards, a rounded icon tile over every heading. Tell it to make the page pop and it can only guess what you mean.
What it does about it
Impeccable splits design work into named commands. Give it a command and a target, such as bolder on the pricing cards, and the agent loads a playbook written for that one move and changes only that part. critique scores the design against usability heuristics and ranks what to fix, and a 62-rule detector that needs no LLM or API key checks each UI file the agent edits.


What people made with it
How it runs inside
- 1
You → Claude Code
/impeccable initOnce per project. It reads your code and asks only what it can't infer, like who the product is for..claude/skills/impeccable/reference/init.md - 2
Claude Code → PRODUCT.md
writes PRODUCT.mdThe answers go into PRODUCT.md, which every later command reads..claude/skills/impeccable/reference/init.md - 3
You → SKILL.md and playbooks
/impeccable bolder the pricing cardsA command and a target; the command table in SKILL.md routes it to that command's playbook..claude/skills/impeccable/SKILL.md - 4
impeccable engine → Claude Code
impeccable context: PRODUCT.md, DESIGN.mdOnce a session, the engine finds the context files and the surface brief and hands them over.crates/context/src/context.rs - 5
SKILL.md and playbooks → Claude Code
reference/bolder.mdThe playbook written for that one move; right before editing, craft-floor.md and its bans..claude/skills/impeccable/reference/bolder.md - 6
Claude Code → impeccable engine
impeccable hook after Edit or WriteEvery UI edit gets a quick pass; when the agent stops, a deep pass runs every rule..claude/settings.json - 7
impeccable engine → Claude Code
findings as JSONWhat it found goes back to the agent before it calls the work done.crates/hook/src/hook.rs - 8
Claude Code → The fixed page
the fixed componentThe page, changed to answer the findings, stays in your source.
What goes in, what comes out
Is it for you
Good for
- Pages your agent builds all look alike
- You want to work one existing component over closely
- You want accessibility, performance and responsive checks ranked P0 to P3 before a launch
- You want CI to catch design anti-patterns
Not for
- Per the README, it is strongest on interfaces that already exist
- Source files like JSX and TSX get the pattern rules only; contrast and tiny-text rules run only on a rendered URL
- A new design fetches directions from the impeccable.style catalog; turning telemetry off does not stop that request
- Live mode is for a local checkout only, never a deployed site
Reviewed by Devalley’s editor before publishing. How the column is made
