Shmakk Desktop

Shmakk Desktop is a graphical desktop client for [shmakk](https://www.twine.net/signin the AI-supervised terminal wrapper. Built with React 19, TypeScript, and Electron, it wraps the CLI’s core capabilities — command correction, tool-driven task execution, safety confirmations, and profile-based runtime modes — inside a rich, multi-pane GUI. It allows for preview in the app for both server applications, html files and markdown. The server apps and the html files are interactively editble. You dont like something delete it, you want to change the css and see it in real time, do it, you want to highlight an element and have AI make some changes; of course!

Relationship to shmakk

Shmakk (the CLI tool) is the engine: it intercepts terminal sessions, corrects mistyped commands, brokers tool calls to AI providers, and enforces safety gating. shmakk-desktop is the front-end: it connects to Shmakk’s AI backend over the same provider interface but presents conversations, diffs, workflows, skills, sts, and settings through windows, panels, and editors instead of a raw tty.

Capabilities

Chat — Streaming AI conversation with collapsible thinking panels, tool-call cards (safe/uncertain/unsafe), syntax-highlighted code blocks, artifact panels, and model selection.

Code — File tree with status indicators, breadcrumb navigation, line numbers, in-file search with highlighting.

Session Search — Full-text search across past sessions with date-range filters, timeline grouping, and session detail previews.

Edit Viewer — Changed-files sidebar, unified and split diffs with syntax coloring, per-hunk and per-file accept/reject.

Memory & Rules — User-editable memory and rule entries with category filtering and search.

Skills Browser — Installable skill cards with enable/disable toggles, status indicators, and filtering.

Workflows — Multi-step workflow runner with step status dots, type badges, and per-step execution logs.

Plans & Tasks — Markdown plan editor with auto-derived task list, dependency tracking, progress bar.

Cowork — Autonomous agent dashboard with active agent cards, scheduled jobs, activity feed, and proactive suggestion cards.

Design Studio — Viewport design tool with layer tree, frame presets, properties panel (layout, typography, background, border), and canvas.

Settings — Provider/endpoint management, API key config, workspace root, theme/font, and keyboard shortcut tables.

Electron Shell — Native titlebar, collapsible sidebar, status bar with connection indicator, system dialogs for directory selection.

Tech Stack

React 19, TypeScript, Zustand (state), React Router v7, Monaco editor, react-markdown + remark-gfm, Lucide icons, Electron 35, Vite 6, plain CSS with design tokens.

ScreenLink

The Problem Nobody Solved Properly

I run three operating systems — not because I enjoy suffering, but because I need all three. A Linux desktop as my main machine, a Windows laptop, and a MacBook. Each has a role. None of them are going anywhere.

The laptops are both well over ten years old. They sit there with perfectly good screens that are dark most of the day. Meanwhile I’m squinting at one monitor wishing I had more space. I searched for years for a setup that actually worked. Nothing did — not cleanly, not cheaply, not without proprietary hardware or a monthly subscription.

So I built one.

NVIDIA Said No

The first wall was NVIDIA. They don’t support creating virtual displays — no official path, no workaround in the docs. Most people stop there.

I didn’t.

Seven days later I had something working. What I learned is that NVIDIA isn’t actually the blocker people think it is. You just have to be stubborn enough to find the path they didn’t document.

What ScreenLink Does

ScreenLink turns those unused laptop screens into real extended monitors over LAN using VNC. No new hardware. No paid software. No cloud.

From a tray icon in my toolbar I can set each laptop independently — extend the screen, or switch to remote control. That’s it. One decision per machine, instantly.

In practice this means I run one keyboard and one mouse across all three computers. I can have my Linux desktop with two extended screens and one machine remote controlled. Or all three screens extended. Or any combination. My main monitor stays in the center where it belongs and everything else falls into place around it.

Honest About the Hard Parts

Getting around NVIDIA’s limitations while keeping the main monitor centered and everything stable — that was genuinely hard. Frustrating in the way that only good problems are frustrating. The kind where you’re cursing at 2am and then something clicks and you feel like you invented electricity.

I use it every single day.

Why Not Just Buy a Screen?

Part of what drives me is making IT accessible to people who don’t have the budget to throw hardware at every problem. Not everyone can just go buy a monitor. But a lot of people have an old laptop collecting dust — and that laptop is an asset waiting to be used.

ScreenLink turns nothing into something. That’s the whole point.

**[Read the story here](https://www.twine.net/signin

Shmakk

**[Demo →](https://www.twine.net/signin

I spend most of my day in the terminal. Switching windows, typing commands, fixing typos—it adds up. So I built Shmakk—an AI assistant that lives in your terminal and understands natural language, not just commands. The result: I can control my entire workflow without leaving the terminal or thinking about syntax.

Core Features

Natural Language Terminal Control

Type natural language directly into your terminal:

create a new branch and call it formatting and use black to format 
all the python code and then push it and merge it to main

Shmakk understands intent, breaks it into steps, and executes them—no need to remember syntax or orchestrate workflows manually.

Intelligent Auto-Correction

Typos don’t stop you. Shmakk uses a deterministic auto-corrector that learns from your environment:

• Builds a glossary from your installed commands
• Weights corrections against your bash/zsh/fish history
• Runs locally—fast and reliable

Example 1: Typo correction

You type:  gti commit -m "add: openai provider"
Shmakk fixes:  git commit -m "add: openai provider"

Example 2: Context-aware correction

You type:  lsof -i :3000
Shmakk understands:  lsof -i :3001
(reads your workspace, corrects based on context)

Multi-Step Workflows with Auto-Coordination

Shmakk’s coordinator system breaks complex projects into steps, validates each one, and maintains context. It uses a plan-first approach: think before acting. Create a plan, get approval, then execute methodically.

Skill Registry & Auto-Loading

Dozens of built-in skills cover code review, backend, frontend, devops, research, logs, design, documents, marketing, and more. Skills load on demand and adapt to your task. Install new skills directly from GitHub:

install skill <github-url>

Self-Aware Commands

Quick commands for introspection:

show help
show skills
find skill design

Instantly access documentation and skill lists without leaving the terminal.

Voice Support (TTS, STT, STS)

Speak naturally. No push-to-talk required:

• STT: Mic input, text responses
• TTS: Text input, spoken responses
• STS: Full speech-to-speech mode
• Interruption: Add info mid-thought without breaking the task

Permission Handling with Explanations

Before running risky commands, Shmakk prompts you:

Run this command? [y/n/?]

Press ? to see a detailed explanation. You stay in control.

Visual Review of Changes

Request a visual diff after execution:

show me your edits

A webpage opens with side-by-side diffs and file changes.

Did it for me

I use shmakk every day all day and it’s everything I always needed, versatile, simple, not a stickler for semantics and very powerful.