MakerSkills: Add AI Agent Skills to Claude Code on Linux

If you want to go deeper, the Claude Code for Linux Sysadmins course covers plugins, skills, and hooks from start to finish.

When the Clone Fails with Permission Denied (publickey)

GitHub’s owner/repo shorthand uses SSH by default. On a fresh Linux workstation without an SSH key loaded into ssh-agent, the clone fails immediately. The error can make it look like there’s a problem with the plugin when it’s actually an SSH authentication issue.

Force HTTPS instead:

export CLAUDE_CODE_PLUGIN_PREFER_HTTPS=1

Git operations have a 120-second timeout by default. On a slow connection, that may not be enough, and you’ll see Git clone timed out after 120s without much useful information about the repository. You can increase the timeout; the value is in milliseconds:

export CLAUDE_CODE_PLUGIN_GIT_TIMEOUT_MS=300000

Set the Environment Variables in Bash

The upstream documentation tells you to put these variables in ~/.zshenv. On most Linux systems, you’re using Bash, so put them in ~/.bashrc instead.

MAKERSKILLS_CONFIG has a sensible default that works as-is. The other variables point to directories you choose.

export MAKERSKILLS_CONFIG="$HOME/.config/makerskills"
export SECOND_BRAIN_VAULT="$HOME/Documents/SecondBrain"
export COMPANY_BRAIN_VAULT="$HOME/Documents/CompanyBrain"
export COMPANY_CFO_ROOT="$HOME/code/company-cfo"
export SLIDE_DECK_REPO="$HOME/code/"

Anything inside angle brackets is a placeholder that you need to replace with your own value. For example, should become the actual name of your site directory.

Reload the file so the current shell picks up the variables:

source ~/.bashrc

Next, create the configuration directories used by the skills. These directories hold your actual data, which is why makerskills keeps them outside the repository:

mkdir -p ~/.config/makerskills/{jab-hook,pm,decide,business-brainstorm,deep-research,personal-cfo,slide-deck}

The brace expansion at the end creates all seven subdirectories in one command:

• mkdir creates the directories.
• -p creates parent directories as needed and doesn’t complain if they already exist.
• {jab-hook,pm,...} expands into seven separate paths, so Bash treats it as if you had typed each path separately.

Install the Runtime Dependencies on Linux

The skills call real command-line tools. watch-video needs yt-dlp and ffmpeg, read-book needs pandoc and Calibre’s ebook-convert, and domain needs whois.

Install only the dependencies for the skills you actually plan to use.

On Ubuntu/Debian:

sudo apt update
sudo apt install ffmpeg pandoc git-filter-repo calibre whois -y

sudo runs the command with root privileges, which package installation needs because it writes to system directories. If you get a Permission denied error, check that you included sudo.

On RHEL/Rocky Linux:

sudo dnf install https://mirrors.rpmfusion.org/free/el/rpmfusion-free-release-$(rpm -E %rhel).noarch.rpm -y
sudo dnf install ffmpeg pandoc git-filter-repo calibre whois -y

ffmpeg isn’t available in the default RHEL repositories, which is why the RPM Fusion release package is installed first. You don’t need that extra step on Ubuntu, where ffmpeg is available in the main repositories.

The yt-dlp package in distribution repositories can lag behind upstream releases, and an outdated version can stop working when YouTube changes. Install the Python tool runner and get the current version through it:

curl -LsSf https://astral.sh/uv/install.sh | sh
uv tool install yt-dlp

The Whisper Substitution

The upstream instructions install mlx-whisper for transcription. MLX is Apple’s framework for Apple Silicon, so there isn’t a Linux equivalent of that package.

Instead, use the reference implementation, which runs on the CPU and can use CUDA when available:

uv tool install openai-whisper

If you’re on a laptop without a GPU and the reference model is too slow, whisper.cpp compiled locally provides a faster CPU option, but it requires an additional build step.

For the PDF rendering used by read-book and second-brain, the Mac instructions install BasicTeX. On Ubuntu/Debian, use: sudo apt install texlive-latex-recommended texlive-fonts-recommended, and on RHEL/Rocky Linux it’s sudo dnf install texlive-scheme-basic.

Run Your First Skill

Open Claude Code and type a skill name as a slash command. decide is a good one to start with because it doesn’t need any configuration or API keys:

/decide

Claude Code should respond by asking for the decision context. If the short name doesn’t resolve, use the namespaced version instead. Claude Code prefixes plugin skills with the plugin name to prevent two plugins from using the same command name:

/makerskills:decide

Now check how many skills were actually installed. Plugins are copied into a versioned cache under your home directory rather than staying in the directory you cloned:

find ~/.claude/plugins/cache -name "SKILL.md" | wc -l

Here’s what the command does:

• find ~/.claude/plugins/cache searches through the plugin cache directory.
• -name "SKILL.md" matches only files with that exact name. Each skill has one SKILL.md file.
• | passes the output to the next command.
• wc -l counts the lines, giving you the number of skills found.

The number you get is the actual skill count on your system, so it’s more reliable than the number listed in the README or install file. Each skill also has its own SKILL.md. You can open those files directly to see what a skill expects before you run it.

Where Your Personal Data Lives

Nothing you configure is pushed into the repository. Personal configuration lives in ~/.config/makerskills//, while generated output is stored under ~/Documents//. Both locations sit outside the repo and are gitignored by design.

This separation also makes moving to a new machine straightforward. Copy ~/.config/makerskills/ and the export lines from your ~/.bashrc to the new machine, reinstall the plugin, and you should be back where you left off.

A few skills support optional API keys for services such as Notion, Typefully, and Brave Search. These skills still work without API keys by using free fallbacks, so you can treat the API key list as something to revisit later rather than a required installation step.

Conclusion

You’ve installed makerskills as a Claude Code plugin, pointed its environment variables to Linux paths in ~/.bashrc instead of ~/.zshenv, replaced the Homebrew dependencies with apt and dnf packages, and swapped the Apple Silicon transcription tool for one that works on your hardware.

The plugin itself was never the Mac-specific part. The dependency chain around it was. Now run the find command against your plugin cache and open one of the SKILL.md files it finds.

Reading how a skill is structured is the quickest way to decide whether you want to use it as-is or fork it into your own namespace with adapt-skill.

Which skill are you planning to run first? And did you encounter a dependency that doesn’t have a clean Linux equivalent on your distro? Drop it in the comments, and I’ll take a look.