Bonsai-27B-GGUF
lmstudio-community · Kit
A GGUF conversion of the 27B Bonsai model for LM Studio, with a 3.80 GB Q1_0 language file and a 0.93 GB BF16 vision projector. Tested for text, vision and par…

Paul Bakaus’s design skill for coding agents, with 24 commands such as polish, bolder, critique and audit for steering and reviewing UI work. A rule-based detector flags generic AI design habits as your agent edits.
Impeccable is an open-source project on GitHub in the Developer category, listed by Mindket. It is free. This is version 2, last updated on . Install it from GitHub as its README says. Get Mindket’s setup notes on your Mindket account, as a zip.
npx impeccable install
And 2 more in Where to use it
Impeccable is Paul Bakaus’s design skill for coding agents: 24 commands such as polish, bolder, critique and audit, plus a detector for generic AI design. Ask an agent for a page with no other guidance and the same habits come back, such as Inter everywhere, purple-to-blue gradients and cards nested in cards. The detector flags them with 62 deterministic rules, no model and no API key. The README says it is strongest on interfaces that already exist.
Front-end developers, designers who code and product teams that ship UI through an AI agent. Teams iterating on a product they already have get the most, since that is where the README says it does best.
| Industry | Who uses it | How they use it | Example |
|---|---|---|---|
| Software products | Design leads and product teams | Write down the design system first, then hand UI work to agents that follow it | A design lead records the current system in DESIGN.md with /impeccable document before handing UI work to agents. |
| Indie and side projects | Solo developers | Give the product its context once, then steer each change with a named command | A solo developer runs init on a side project so later commands know who the product is for. |
| Engineering teams with CI | Teams that review every pull request | Run the detector in CI with no agent at all | A team adds the detector to GitHub Actions and gets annotations on each pull request. |
| Design and front-end studios | Designers who code | Try versions of a section live on the dev server and keep the best | A developer cycles through variants of a hero in live mode, keeps one, and it is written back to the source. |
If you never touch interface code, there is nothing for it to work on.
It reaches well past one agent.
Get it saves a zip of Mindket’s setup notes in your downloads folder, with each agent’s steps from pbakaus’s README.
| Need | Detail |
|---|---|
| Node.js | 18.19 or later, per Impeccable’s getting-started docs, and only for npx |
| Your project root | The installer runs from there |
| Hook approval | Codex needs the project hook approved under /hooks, and Grok Build needs /hooks-trust or a launch with --trust |
| Agent Skills | Cursor and Gemini CLI may need it turned on in their settings |
npx impeccable install
In Claude Code you can add it as a plugin instead, then open /plugin and install Impeccable.
/plugin marketplace add pbakaus/impeccable
Later, one line refreshes an install made with npx.
npx impeccable update
Name a command and a target. A narrow target gets better results, since one section reviewed closely beats a whole page at a glance.
| Command | What it does |
|---|---|
| init | Writes PRODUCT.md, so later commands know who the product is for |
| document | Records an existing design system in DESIGN.md |
| critique | Reviews one part of the UI and ranks what to fix first |
| audit | Checks accessibility, performance, theming and responsive behavior |
| polish, bolder, quieter, harden | Named design moves you aim at one target, such as settings or checkout |
| pin | Gives a command you use often a short form of its own |
| doctor | Finds stale context files, unused settings and broken hooks |
/impeccable critique the onboarding flow /impeccable quieter the landing page /impeccable polish settings /impeccable harden checkout
Type /impeccable alone to list the commands, and /impeccable pin audit gives you /audit. In a terminal or in CI, the detector needs no agent. Exit code 2 means findings, and 1 means a target could not be scanned.
npx impeccable detect src/ npx impeccable detect --json . npx impeccable detect --github src/
After adding a pricing section, you ask for a second opinion on that one component.
/impeccable critique the pricing cards
The agent scores the cards against usability heuristics and ranks what to fix first. Run audit instead and you get a technical report, with findings ranked from P0, blocking, to P3, polish.
If you are new. An AI agent left to itself tends to give every site the same look. Impeccable gives it named design moves, such as bolder or quieter, and a checker that flags the usual habits while it edits. Run /impeccable init on a small project, then ask it to critique one component.
For experts. The detector’s rules make no model call, so a scan is fast and repeatable but catches only what a rule describes. Source files such as JSX and TSX get the pattern rules only. Low contrast and tiny text need a rendered page, so point the CLI at your dev server too. Commands read PRODUCT.md and DESIGN.md, and results improve once both exist. A critique is still a model’s judgment against heuristics: a second opinion, not a pass mark.
Checks run on your machine, with a few calls out to the maker’s site.
The line below turns telemetry off, which stops only the anonymous record of which direction you picked. DO_NOT_TRACK=1 and IMPECCABLE_NO_TELEMETRY=1 do the same.
npx impeccable telemetry off
# Impeccable setup notes
This kit holds setup notes written by Mindket. It does not contain Impeccable itself. The skill, plugin and hooks install from Paul Bakaus's repository, pbakaus/impeccable: https://github.com/pbakaus/impeccable
License: Apache-2.0, per the repository's LICENSE file: https://github.com/pbakaus/impeccable/blob/HEAD/LICENSE
Commands in these notes are copied from the repository README (https://github.com/pbakaus/impeccable#installation) and the getting-started page (https://impeccable.style/tutorials/getting-started).
## What installs
- One skill, impeccable, with 24 commands such as polish, bolder, critique and audit.
- An optional design hook that scans the UI files your agent edits. The README lists it for Claude Code, Cursor, Codex, GitHub Copilot, Grok Build and Hermes Agent.
- Subagents. The README notes a manual copy skips them and the design hook.
- A self-contained engine binary that ships with the skill or downloads once into ~/.impeccable/bin/.
## Before you start
Node.js 18.19 or later. Node is needed only for npx.
## Install for any supported agent
From your project root:
```
npx impeccable install
```
Follow the prompts to choose your tools and a project or global scope, accept the design hook, then reload your tool. For scripted installs, the installer takes --providers=claude,codex,cursor and --scope=project|global.
Then, in your agent chat:
```
/impeccable init
```
Per-agent notes are in the setup folder: claude-code.md, codex.md, cursor.md, github-copilot.md, gemini-cli.md and grok-build.md.
## Update
```
npx impeccable update
```
Marketplace and plugin installs update from your tool's plugin menu. Reload your tool after an update.
## Remove
The README and the docs do not describe how to remove Impeccable.
## Telemetry
```
npx impeccable telemetry off
```
DO_NOT_TRACK=1, IMPECCABLE_NO_TELEMETRY=1 or "telemetry": false in .impeccable/config.json also turn it off. The skill still fetches visual directions from the impeccable.style catalog when a new design starts. Separately, Impeccable checks impeccable.style/api/version for a newer version, normally at most once a day, and sends no project content. IMPECCABLE_NO_UPDATE_CHECK=1 turns that off. Details: https://impeccable.style/privacy
## Hooks and troubleshooting
The README warns that in Claude Code, installed command hooks run even when you deny the model's own commands. It suggests reviewing installed hooks before unattended runs, or passing this flag to turn them off for a run:
```
--settings '{"disableAllHooks": true}'
```
/impeccable doctor finds stale context files, unused settings and broken hooks.README.md 3 KB setup/claude-code.md 1 KB setup/codex.md 994 B setup/cursor.md 1 KB setup/gemini-cli.md 826 B setup/github-copilot.md 1 KB setup/grok-build.md 756 B
Updated by Mindket to the current version of this kit.
Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search and on Trendshift. Mindket doesn’t write or check them. Open one to read it where it was posted.
lmstudio-community · Kit
A GGUF conversion of the 27B Bonsai model for LM Studio, with a 3.80 GB Q1_0 language file and a 0.93 GB BF16 vision projector. Tested for text, vision and par…
LibertAIDAI · Kit
Runs DeepSeek-V4.1-Flash-1017 as a 2.9 bits-per-weight EXL3 quantization, a 196 GiB copy for people serving large models on their own hardware. Perplexity stay…
facebook · Kit
Generates short music clips from a text description or an audio prompt, in the small checkpoint of facebook's MusicGen. It suits developers and music hobbyists…
An MCP server and CLI by morluto that lets your coding agent study binaries, desktop and Android apps, websites and firmware on your own computer, and back eac…
Leonxlnx · Skill
Leonxlnx’s Agent Skill that gives Claude Code and Codex an art direction, pictures from open collections and a real-browser check, so the websites and app scre…
alchaincyf · Skill
Turns a one-line request in your AI agent into a finished design: a launch animation, a clickable app prototype, an editable slide deck, an infographic or a se…