DEV Community: Ugo

unitmux: A Floating Desktop App for Claude Code and Codex in tmux

Ugo — Sat, 09 May 2026 08:27:00 +0000

Introduction

If you're running Claude Code or Codex in tmux, you've probably experienced this:

You want to give instructions while reading code in your editor, but every time you have to Cmd+Tab back to the terminal.

The AI asks "1. Yes / 2. No" and you have to switch back to the terminal just to type 1.

When running multiple Claude or Codex sessions in parallel, it's hard to keep track of which pane is in what state.

These small context switches add up and repeatedly break your flow of thought. It's surprisingly stressful.

To solve this, I built a desktop app called unitmux.

What is unitmux?

A desktop app that lets you send instructions directly to Claude Code or Codex running in tmux, from a floating window.

It stays always-on-top, auto-detects tmux panes, and eliminates the need to switch focus to the terminal. You can give instructions to your AI without interrupting your editor workflow.

It also detects panes running the ai wrapper command in addition to claude and codex.

Install with one line:

brew install --cask yugo-ibuki/tap/unitmux

Features

One-Click Responses to Choices

When Claude Code or Codex presents numbered choices, unitmux automatically displays clickable buttons.

You can also respond instantly via keyboard with Ctrl+1-Ctrl+9, so you don't even need to reach for the mouse.

That minor annoyance of switching to the terminal just to type a single character is gone.

Update: Choice detection now supports multiple patterns beyond simple numbered lists: marker-prefixed choices (❯, ›, >, ●), colon-separated inline options such as 1: staging 2: production 3: dev, and multi-line labels such as permission prompts with full command text.

It also intelligently filters out session rating feedback such as "How is Claude doing?" and CLI footers to avoid showing false positives.

Codex support has also been improved. unitmux can detect Codex-style letter choices and uses separate heuristics for Codex prompts, including English and Japanese choice context.

Global Shortcut: Access from Anywhere

Press Cmd+Shift+H to bring focus back to unitmux from any app.

Focus lands directly on the text area, so you can start typing right away and send with Cmd+Enter.

Reading docs in Chrome while instructing the AI now takes just a few seconds, practically seamless.

Update: The focus shortcut is now fully customizable from the Sidebar settings. It also auto-expands compact mode when triggered from another app, so you're immediately ready to type.

Session Status at a Glance

Panes are listed as tabs, with a colored dot indicating status:

Green: Accepting input
Orange: Processing
Gray: Waiting for response / choices available

Each pane also shows a small badge for the tool:

CC: Claude Code
CX: Codex

When running multiple Claude or Codex sessions in parallel, you no longer have to wonder "how far along is that pane?"

Switch between panes with Cmd+↑ / Cmd+↓ or Ctrl+H / Ctrl+L.

Update: Panes are now grouped by tmux session in the header, making it easier to navigate when you have many panes across different sessions. Cross-session navigation is also available via Ctrl+Cmd+H / Ctrl+Cmd+L.

Status detection has been refined. For Claude Code, unitmux reads the pane title for idle markers and spinner characters, and parses recent pane output for waiting prompts and choices. Codex status detection uses its own heuristics, such as Working, Thinking, Executing, and enter to send.

unitmux also detects panes running the ai wrapper command in addition to claude and codex.

Git Operations in a Popup

Ctrl+G opens a Git operations popup.

It only covers simple operations, but when you want to commit frequently during the early stages of personal development, being able to do it seamlessly is a real advantage.

You can:

Stage all changes
Select individual files with Space
Stage selected files with Enter
Type a commit message and press Enter
Push with Ctrl+P

The popup supports vim-style navigation with j / k.

Being able to commit right when the AI finishes a chunk of work, without going back to the terminal, is quietly satisfying.

Update: The Git popup now displays the current branch name in its header and shows detailed Git status: modified files, untracked files, deleted files, and so on. It also auto-refreshes pane detail after each operation and shows success/error feedback that auto-dismisses after 2 seconds.

Git Diff Viewer

Ctrl+F opens a Git diff viewer.

This is separate from the Git operations popup. The Git popup is for staging, committing, and pushing. The diff viewer is for quickly checking what changed.

It supports:

Unstaged / staged diff switching
File sidebar
Directory folding
Collapsible file sections
Hunk navigation
Vim-style scrolling

Useful keys:

j / k: scroll
d / u: half-page scroll
g / G: top / bottom
]c / [c: next / previous hunk
s: toggle staged / unstaged
q: close

This makes it easier to review AI-generated changes without switching back to the terminal to run git diff.

Semi-Transparent Window to Overlay on Your Editor

You can adjust the window opacity between 50% and 100% with a slider.

Combined with Always on Top, you can see through to your editor's code while giving instructions to the AI.

You can keep an eye on Claude or Codex activity while reading code.

Pair this with Cmd+Shift+H and you can shift focus to the app, write a prompt or pick a choice, all without touching the mouse.

View Current Session Content

Ctrl+P lets you check the session conversation without looking at the terminal.

When you just want to see what question was asked or glance at the last answer, this lets you quickly check and move on to the next prompt.

When the agent runs, it produces a lot of log output, but usually what you actually want to see is the final conclusion. Being able to quickly check just the part you care about and move right into the next prompt is the real benefit.

Update: The preview overlay now has two modes.

The first press of Ctrl+P shows a static snapshot. Pressing Ctrl+P again while the overlay is open switches to live streaming mode, which polls pane content every 500ms and shows a LIVE badge.

Vim-style navigation is available inside the preview:

j / k: line scroll
d / u: half-page scroll
g / G: top / bottom

When available, unitmux also supplements tmux pane output with Claude's JSONL conversation history under ~/.claude, so older conversation context is easier to inspect.

Image Attachment

Images can be attached from the + button or by dragging and dropping them onto the window.

Thumbnails are displayed in unitmux, and the image paths are sent to the AI CLI using bracketed paste so the CLI can detect them as image inputs.

This is useful when you want to ask about screenshots, UI states, diagrams, or local image files without manually typing paths into the terminal.

Compact Mode

Ctrl+W shrinks the window to a minimal bar of about 70px height.

This is useful when you want unitmux to stay visible without taking up screen space.

Expanding back restores the previous window size and position. The shortcut key is customizable.

Session Detail Overlay

Ctrl+D shows detailed information about the selected pane:

Model name
Session ID
Git branch
Git status
Working directory
Pane dimensions
Start command
PID / tty

It also includes a close button with a confirmation dialog to terminate the pane.

Create New Session

unitmux can create new Claude or Codex sessions from the UI.

There are two workflows:

Ctrl+N: quickly add a Claude session to the current tmux session
Ctrl+Shift+N: open a dialog to create a new tmux session or add a Claude/Codex window to an existing session

In the dialog, you can select the target tmux session and choose between claude and codex.

Shell Mode

Ctrl+B toggles shell mode.

In shell mode, unitmux sends input to a dedicated unitmux-shell tmux window instead of the selected AI pane.

The shell pane is created on demand and uses your default shell. If it is manually closed, unitmux recreates it on the next send or preview.

Preview and live streaming also work for shell output.

This is handy when you want to run quick commands near your AI sessions without switching back to the terminal.

Sidebar Settings Panel

A dedicated settings panel is accessible from the header, offering:

Always on Top toggle
Opacity slider
Theme switch: dark / light
Send key config: Enter or Cmd+Enter
Vim mode toggle
Font size control
Customizable shortcut keys for compact, preview, detail, Git, diff, and focus
Choice modifier: Ctrl or Cmd
Slash command manager

Slash Commands & Skills Integration

Type / in the input area to filter and select from registered commands.

User-defined slash commands can be managed in the Sidebar. You can create, edit, and delete reusable commands such as /fix, /review, or /test.

In addition, unitmux auto-discovers skill commands from:

~/.claude/skills/
.claude/skills/ in the selected project

It parses SKILL.md files for names and descriptions and shows them as read-only slash command candidates.

Input History

The last 10 unique inputs are stored.

Navigate with ↑ / ↓ in the textarea.

Your current draft is preserved when navigating up and restored when you come back down.

Vim Mode

When enabled, unitmux sends Escape followed by i before your text to enter insert mode.

This is useful if you've accidentally left the pane in normal mode while using Claude CLI's vim editor mode.

This is skipped when responding to choices, since single digit choices work directly.

Help Overlay

Ctrl+, opens a quick reference for keyboard shortcuts, organized by category.

It supports the same vim-style scrolling as other overlays.

Installation & First Launch

brew install --cask yugo-ibuki/tap/unitmux

If you'd rather install from DMG, grab it from Releases.

If you see a Gatekeeper warning on first launch, go to:

System Settings → Privacy & Security → Open Anyway

Closing Thoughts

I built this for people who use AI coding assistants in tmux and wanted to simplify the world of "too many terminal windows."

The guiding principle was "simple, then simpler."

No unnecessary mouse operations. Everything is designed to work with Vim-like keybindings so it feels natural to operate.

Rather than trying to replace tmux or become a full terminal manager, unitmux focuses on making prompt input and agent monitoring as low-effort as possible.

tmux is still the base. unitmux just strips away the context switches around it.

Feedback and PRs are very welcome.

👉 https://github.com/yugo-ibuki/unitmux

A star would be really motivating!

Notes

Mac is the primary supported platform.

Linux builds exist, but Linux support has not been verified thoroughly, so please keep that in mind.

Also, since I primarily use Claude Code, there may still be gaps in Codex support.

Share Claude-created Documents Across Worktrees with dot-claude-sync

Ugo — Sun, 16 Nov 2025 10:47:12 +0000

About dot-claude-sync

TL;DR

A Go-based CLI tool to synchronize .claude directories across multiple projects/worktrees
Flexible sync strategies with group management and priority system
Dramatically improves Claude Code workflow in git worktree environments

Introduction

My workflow with Claude Code involves having it create investigation documents and TODO lists first, which I then review before having it write code. I want these documents to be within the project so I don't have to open a separate editor to view them.

However, placing them in the project root causes them to appear in git diffs, and setting up global ignore is cumbersome. Additionally, placing them at the root risks polluting Claude's context with unnecessary prompts. That's when I discovered the .claude folder - a directory that Claude doesn't automatically read.

I tried organizing documents in .claude and found it worked well. For example, when fixing a dashboard, I create a folder like .claude/custom-documents/dashboard-fix-XXX to consolidate all task-related content there.

When working with Claude Code, you often save useful prompts, skills, and project context in the .claude directory.

However, when developing with git worktrees across multiple branches, .claude isn't synced between worktrees since it's gitignored.

dot-claude-sync was developed to solve this problem.

Background and Challenges

Using .claude with Claude Code

When developing with Claude Code, it's useful to save the following in the .claude directory as documentation:

Task-specific prompts
Frequently used skills and commands
Implementation specifications and TODO lists
Project context information

These are not managed by git (gitignored), allowing you to maintain Claude-specific long-term context without polluting the repository.

While you can share with your team, individual documentation quality varies, and unorganized growth leads to context pollution.

Therefore, it's better to manage personal documents individually and share skills through Plugins on the marketplace.

Git Worktree Challenges

Issues arise when using git worktrees:

my-project/
├── main/           # Main worktree
│   └── .claude/
│       └── prompts/useful-prompt.md
├── feature-a/      # feature-a worktree
│   └── .claude/    # Empty!
└── feature-b/      # feature-b worktree
    └── .claude/    # Empty!

Even if initially synced, .claude contents diverge as you work across worktrees
Each worktree requires separate setup
Manual copying of useful prompts between worktrees

dot-claude-sync Solution

dot-claude-sync manages projects through groups and controls sync strategy with a priority system.

Basic Usage

1. Installation

Go install:

go install github.com/yugo-ibuki/dot-claude-sync@latest

Short alias version (dcs command) also available:

go install github.com/yugo-ibuki/dot-claude-sync/cmd/dcs@latest

After installation, you can use the dot-claude-sync command (or dcs).

:::message
$GOPATH/bin (usually ~/go/bin) must be in your PATH.
:::

Build from source:

git clone https://github.com/yugo-ibuki/dot-claude-sync.git
cd dot-claude-sync
go build                    # Generates dot-claude-sync binary
go build -o dcs ./cmd/dcs   # Generates dcs alias binary

2. Initial Setup (Interactive)

dot-claude-sync init
# or short version
dcs init

Running this command starts an interactive setup:

Enter group name: Choose a name for your project group (e.g., my-app, web-projects)
Enter project paths: Input paths to .claude directories to sync
- Multiple paths supported (empty line to finish)
- Relative paths with ~ supported
Auto-create config file: Creates ~/.config/dot-claude-sync/config.yaml

Example execution:

$ dot-claude-sync init
Enter group name: my-app
Enter project paths (empty line to finish):
Path 1: ~/projects/my-app/main/.claude
Path 2: ~/projects/my-app/feature-a/.claude
Path 3: ~/projects/my-app/feature-b/.claude
Path 4: [Enter]

Configuration file created: ~/.config/dot-claude-sync/config.yaml

3. Auto-detect Worktrees (Optional)

Instead of manual path entry, use the detect command to auto-detect worktrees:

dot-claude-sync detect ~/projects/my-app --group my-app
# or
dcs detect ~/projects/my-app --group my-app

This command:

Runs git worktree list to detect worktrees
Checks if .claude directory exists in each worktree
Automatically adds to config file if found

4. View Configuration

dot-claude-sync list
# or
dcs list

View current configuration showing group names and project paths.

5. Run Sync

dot-claude-sync push my-app
# or
dcs push my-app

Syncs .claude directories across all projects in the specified group.

Safe execution (dry-run mode):

dot-claude-sync push my-app --dry-run
# or
dcs push my-app --dry-run

Previews what will happen without actually copying files. Recommended for first run with --dry-run.

Configuration File Example

~/.config/dot-claude-sync/config.yaml:

groups:
  my-app:
    paths:
      main: ~/projects/my-app/main/.claude
      feature-a: ~/projects/my-app/feature-a/.claude
      feature-b: ~/projects/my-app/feature-b/.claude
    priority:
      - main  # Highest priority (master configuration)

Running dot-claude-sync push my-app with this configuration:

Collects files from all project .claude directories
When duplicate filenames exist, uses file from highest priority project (main)
Distributes collected files to all projects

Key Features

🔍 detect - Auto-detect Worktrees

dot-claude-sync detect ~/projects/my-app --group my-app
# or
dcs detect ~/projects/my-app --group my-app

Batch adds multiple worktree .claude directories to configuration in git worktree environments.

What it does:

Runs git worktree list in specified directory
Checks if .claude directory exists in each detected worktree
Automatically adds to config file (~/.config/dot-claude-sync/config.yaml) if found
Uses worktree branch name as alias (e.g., main, feature-a)

Benefits:

No manual entry of worktree paths
Quick response to worktree configuration changes (add/remove)
Reduces typo and path error risks

Example execution:

# Detect worktrees in ~/projects/my-app and add to my-app group
dcs detect ~/projects/my-app --group my-app

# After execution, config.yaml is auto-updated:
# groups:
#   my-app:
#     paths:
#       main: ~/projects/my-app/main/.claude
#       feature-a: ~/projects/my-app/feature-a/.claude
#       feature-b: ~/projects/my-app/feature-b/.claude

Use cases:

Quick setup of worktree environment in new project
Update configuration when adding new worktree to existing project
Easy reproduction of same worktree structure by team members

📤 push - Run Sync

dcs push my-app

Collects .claude files from all projects in group and distributes based on priority.

💾 backup - Create Backup

dcs backup my-app

Creates timestamped backups of .claude directories for all projects in group.
Backups are saved to ~/.local/share/dot-claude-sync/backups/.

Use cases:

Safety before first sync
Protection before major changes
Regular backup operations

🗑️ rm - Batch Delete

dcs rm my-app prompts/old-prompt.md

Deletes specified file from all projects in group.

📝 mv - Batch Rename

dcs mv my-app old-name.md new-name.md

Renames/moves files in all projects in group.

⚙️ config - Configuration Management

# Add group
dcs config add-group new-group

# Add project
dcs config add-project my-app feature-c ~/projects/my-app/feature-c/.claude

# Set priority
dcs config set-priority my-app main feature-a feature-b feature-c

Edit configuration file directly from command line.

Command Aliases

dot-claude-sync and dcs provide identical functionality. Both aliases work for all commands:

# These work exactly the same
dot-claude-sync init
dcs init

dot-claude-sync push my-app
dcs push my-app

dot-claude-sync list
dcs list

dcs is recommended for daily use due to its shorter length.

Use Cases

Case 1: Quick Worktree Environment Setup

# Auto-detect and add worktree .claude directories
dcs detect ~/projects/my-app --group my-app

# Create backup (for safety)
dcs backup my-app

# Start syncing immediately
dcs push my-app

Case 2: Distribute Shared Configuration

groups:
  web-projects:
    paths:
      shared: ~/projects/shared-config/.claude  # Shared config master
      frontend: ~/projects/frontend/.claude
      backend: ~/projects/backend/.claude
    priority:
      - shared  # Highest priority

# Distribute shared project settings to all projects
dcs push web-projects

Case 3: Client Project Template Management

groups:
  clients:
    paths:
      template: ~/templates/client/.claude
      client-a: ~/clients/a/.claude
      client-b: ~/clients/b/.claude
    priority:
      - template

Deploy template project settings to each client project.

Global Options

All commands support the following options:

--config <path>   # Specify configuration file path
--dry-run         # Preview execution (no actual changes)
--verbose         # Output detailed logs
--force           # Skip confirmation prompts

Usage examples:

# Safe confirmation with dry-run
dcs push my-app --dry-run

# Execute with detailed logs
dcs push my-app --verbose

# Use custom configuration file
dcs push my-app --config ~/custom-config.yaml

Summary

dot-claude-sync is a small tool that improves Claude Code workflow in git worktree environments.

Recommended for:

Users working with git worktrees
Those wanting shared .claude settings across multiple projects
Anyone looking to streamline prompt and skill management

If you're interested, please give it a try!

CVE Search Tools

Ugo — Sun, 29 Sep 2024 06:34:51 +0000

API

National Vulnerability Database (NVD) API

Official Website	Provider	Notes
NVD Vulnerabilities	NIST (National Institute of Standards and Technology)	API key required, usage restrictions apply

Features

Official and highly reliable
Provides both CVSSv2 and v3 scores
Free to use

Endpoint Example
https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-YYYY-XXXXXXX

CIRCL CVE Search API

Official Website	Provider	Notes
CIRCL CVE Search	CIRCL (Computer Incident Response Center Luxembourg)

Features

Open source, can be self-hosted
Provides both CVSSv2 and v3 scores

Endpoint Example
https://cve.circl.lu/api/cve/CVE-YYYY-XXXXXXX

Vulners API

Official Website	Provider	Notes
Vulners API	Vulners	Paid plans available, limited free usage possible

Features

Extensive vulnerability database
Provides CVSSv2 and v3 scores

Endpoint Example
https://vulners.com/api/v3/search/lucene/?query=CVE-YYYY-XXXXXXX

Red Hat Security Data API

Official Website	Provider	Notes
Red Hat Security Data API	Red Hat

Features

Specialized in CVEs related to Red Hat products
Provides CVSSv2 and v3 scores

Endpoint Example
https://access.redhat.com/labs/securitydataapi/cve/CVE-YYYY-XXXXXXX.json

MITRE CVE API

Official Website	Provider	Notes
MITRE CVE Search	MITRE Corporation	CVSS scores often not included

Features

Official source of CVE data
Provides basic CVE information

Endpoint Example
https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-YYYY-XXXXXXX

VulDB API

Official Website	Provider	Notes
VulDB API	VulDB	Paid service

Features

Provides extensive vulnerability information
Includes CVSS scores, affected products, and remediation information

Endpoint Example
https://vuldb.com/?kb.api

Shodan API

Official Website	Provider	Notes
Shodan API	Shodan	Primarily paid, limited free plan available

Features

Provides vulnerability information for internet-connected devices
Offers information on actual vulnerable systems associated with CVEs

Endpoint Example
https://api.shodan.io/shodan/host/CVE-YYYY-XXXXXXX

OpenCVE API

Official Website	Provider	Notes
OpenCVE API	OpenCVE (Open source project)	Community-driven project

Features

Collects and provides CVE information as an API
Can be self-hosted

Endpoint Example
https://opencve.io/api/cve/CVE-YYYY-XXXXXXX

NIST National Checklist Program Repository API

Official Website	Provider	Notes
NIST NCP Repository	NIST	Rich in information related to U.S. government systems

Features

Provides security configuration checklists and CVE information
Focuses on U.S. government systems

Endpoint Example
https://nvd.nist.gov/ncp/repository/CVE-YYYY-XXXXXXX

Cybersecurity and Infrastructure Security Agency (CISA) API

Official Website	Provider	Notes
CISA API	U.S. Cybersecurity and Infrastructure Security Agency	No specific restrictions mentioned, but it's recommended to check before use

Features

Provides vulnerability information related to critical infrastructure
Focuses on CVEs deemed important from the U.S. government perspective

Endpoint Example
https://www.cisa.gov/known-exploited-vulnerabilities-catalog

ExploitDB API

Official Website	Provider	Notes
ExploitDB API	Offensive Security	No specific restrictions mentioned, but it's recommended to check before use

Features

Provides information on publicly available exploit code and related CVEs
Aimed at penetration testers and security researchers

Endpoint Example
https://www.exploit-db.com/api

Rapid7 Open Data API

Official Website	Provider	Notes
Rapid7 Open Data API	Rapid7

Features

Provides data on vulnerabilities, attacks, and other security-related information
Offers detailed technical information related to CVEs

Endpoint Example
https://opendata.rapid7.com/

OSS

OpenCVE

Official Website	Provider	Notes	Stars
OpenCVE	opencve	v1 will soon be closed, and v2 will be released	~1800

Features

OSS for collecting, analyzing, and displaying CVE information
Provides Web interface and REST API
Written in Python

CVE-Search

Official Website	Provider	Notes	Stars
CVE-Search	cve-search		~2300

Features

Imports CVE, CPE, and CWE data and makes it searchable
Uses MongoDB to store data
Written in Python, provides Web interface and API

Dependency-Track

Official Website	Provider	Notes	Stars
Dependency-Track	DependencyTrack	No specific restrictions mentioned, but it's recommended to check before use	~2600

Features

Component analysis platform for software supply chain
Provides vulnerability data including CVE information
Written in Java

nvdtools

Official Website	Provider	Notes	Stars
nvdtools	Facebook

Features

Vulnerability database written in Go
Parses NVD data and provides it in a user-friendly format
Offers both CLI tools and libraries

OWASP Dependency-Check

Official Website	Provider	Notes	Stars
OWASP Dependency-Check	Individual		~6300

Features

Scans project dependencies and detects known vulnerabilities
Uses NVD database
Written in Java but supports many languages and build tools

Grype

Official Website	Provider	Notes	Stars
Grype	Anchore		~8500

Features

Vulnerability scanner for container images and filesystems
Uses multiple vulnerability databases
Written in Go

VulnerableCode

Official Website	Provider	Notes	Stars
VulnerableCode	Individual		~500

Features

Aggregates vulnerability data from multiple sources
Provides REST API and Web UI
Written in Python/Django

Thinking of how we should follow AI.

Ugo — Sun, 19 May 2024 06:32:34 +0000

Introduction

The emergence of GitHub Copilot Workspace (GCW) is expected to have a significant impact on the future of engineering work. Even the introduction of GitHub Copilot, a plugin that offers AI-powered code suggestions, has already brought substantial changes, which is evident from its current usage.

I have my own thoughts on this matter and would like to share them.

Background Information about GCW

For example, suppose you want to prepare some mock data. Ordinarily, you would have to painstakingly create arrays, generate data, and consider the content of the mock data by yourself. However, with GitHub Copilot, you can simply write a comment like this:

// Describe data that fits the following content:
// - Number of data: 100
// - Data structure: {name: string, age: number, address: string}
// - Names should be fictional general names
// - Ages should be random numbers between 20 and 50
// - Addresses should be real addresses but up to the city level

Copilot will then automatically generate 100 data points that meet your specifications. You can also easily modify the generated data by simply providing a comment like "change this and that, and recreate the data with these changes" to the data generated above, and it will regenerate it for you.

This reduces the human workload to less than one-tenth. Additionally, providing specifications generates corresponding code and tests, showcasing a significant shift in engineering work.

According to GitHub Articles that say GCW is evolving even further, it is said to go beyond code-level and data creation, understanding the context of the project and creating code.

For example, you can create an issue with specifications and GCW will generate code that meets those specifications and even allow you to check its functionality. We can expect it to be quite powerful, if you actually use it once.

The Future of Engineering Work

So, how will the work of engineers change as a result of these developments? I believe that the job of an engineer will not disappear, but the definition of the job will change.

The current work of engineers like:

Defining specifications
Writing code
Ensuring quality

This will transform dramatically to:

Determine the direction of the product
Breaking down into specifications
Having AI make adjustments and works with the prompt you write

However, the problem here is, how far should we go with these revisions and specifications?, what level of engineering skills are required to do so?, and where can these skills be acquired?

From the current perspective, everything is unknown. The way we work will change drastically.

For now, if you want to consider API interfaces, you need to enforce those interfaces with libraries, and you also need to consider library selection and need to think of how you should operate it. To make these decisions, you need operational experience and experience with failures.

However, if you use AI (GCW), the interfaces will not be operated by us. If they are misoperated, it means that the specifications are wrong. In other words, there is no need to think about a mechanism to enforce them. Even if you manage interfaces with documents, AI can complement them so that there are no omissions or mistakes in changes.

Today, there are still many experienced engineers and the AI era has just begun, so there are many people with a refined sense of direction and land sense. However, in 3-4 years, a new generation of engineers who do not need this sense of direction will emerge. In 10 years, more engineering skills have rusted because they were no longer needed.

The definition of an engineer will shift from “an engineer checks what AI generates” to a new role. While some may think that the job of engineers will not disappear because they will verify AI outputs, the skills needed to judge whether these outputs are correct.

Currently, we manage with the wisdom of those without AI, but people who have a kind of experience will gradually disappear and those kind of experiences become unnecessary. Finally, a new definition of engineering will emerge.

Conclusion

Many engineers say they are afraid of their jobs being taken over by AI.

I am one of them.

I have tried to put into words my thoughts on where that fear comes from.

Not only engineers but everyone will face unknown world in the future. The year 2023 seems to have become a turning point for this world. We need to think about what skills will be required for this new world and what abilities we should pass on to the next generation, and adapt our work styles accordingly.

DEV Community: Ugo

unitmux: A Floating Desktop App for Claude Code and Codex in tmux

Introduction

What is unitmux?

Features

One-Click Responses to Choices

Global Shortcut: Access from Anywhere

Session Status at a Glance

Git Operations in a Popup

Git Diff Viewer

Semi-Transparent Window to Overlay on Your Editor

View Current Session Content

Image Attachment

Compact Mode

Session Detail Overlay

Create New Session

Shell Mode

Sidebar Settings Panel

Slash Commands & Skills Integration

Input History

Vim Mode

Help Overlay

Installation & First Launch

Closing Thoughts

Notes

Share Claude-created Documents Across Worktrees with dot-claude-sync

About dot-claude-sync

TL;DR

Introduction

Background and Challenges

Using .claude with Claude Code

Git Worktree Challenges

dot-claude-sync Solution

Basic Usage

1. Installation

2. Initial Setup (Interactive)

3. Auto-detect Worktrees (Optional)

4. View Configuration

5. Run Sync

Configuration File Example

Key Features

🔍 detect - Auto-detect Worktrees

📤 push - Run Sync

💾 backup - Create Backup

🗑️ rm - Batch Delete

📝 mv - Batch Rename

⚙️ config - Configuration Management

Command Aliases

Use Cases

Case 1: Quick Worktree Environment Setup

Case 2: Distribute Shared Configuration

Case 3: Client Project Template Management

Global Options

Summary

Links

CVE Search Tools

API

National Vulnerability Database (NVD) API

CIRCL CVE Search API

Vulners API

Red Hat Security Data API

MITRE CVE API

VulDB API

Shodan API

OpenCVE API

NIST National Checklist Program Repository API

Cybersecurity and Infrastructure Security Agency (CISA) API

ExploitDB API

Rapid7 Open Data API

OSS

OpenCVE

CVE-Search

Dependency-Track

nvdtools

OWASP Dependency-Check

Grype

VulnerableCode

Thinking of how we should follow AI.

Introduction

Background Information about GCW