Installing

A skill is just a directory with a SKILL.md in it. That means there is no installer to trust and nothing to uninstall — the whole operation is “put this directory where your agent looks for skills”. Everything below is a different amount of ceremony around that one idea.

If you only want the short version:

gh skill install calebstewart/ai-slop writing-skills --agent claude-code --scope user

Where skills live

Worth knowing before you pick a method, because every method below is just a way of writing to one of these directories.

.agents/skills/ is a shared convention that several agents read, which is why gh skill defaults to it. Check your own agent’s docs, or run gh skill install --help for the current list of supported hosts.

Using gh skill

The GitHub CLI skill command — currently in preview — handles discovery, placement, and update tracking. It is the method to prefer, because it is the only one that leaves behind enough metadata to update the skill later.

gh skill install calebstewart/ai-slop writing-skills

That works, but it probably does not do what you want. Non-interactively gh skill defaults to --agent github-copilot --scope project, so be explicit:

# Claude Code, available in every project
gh skill install calebstewart/ai-slop writing-skills --agent claude-code --scope user

# Claude Code, only inside the current repo
gh skill install calebstewart/ai-slop writing-skills --agent claude-code --scope project

Reading one before you commit to it

gh skill preview calebstewart/ai-slop writing-skills

Useful variations

# Faster: naming the exact path skips a full repo tree traversal
gh skill install calebstewart/ai-slop skills/writing-skills/SKILL.md --agent claude-code

# Pin to a tag or commit instead of the latest release / default branch HEAD
gh skill install calebstewart/ai-slop writing-skills@v1.2.3 --agent claude-code

# Drop it somewhere arbitrary, ignoring --agent/--scope
gh skill install calebstewart/ai-slop writing-skills --dir ~/some/other/place

# Overwrite an existing copy without being asked
gh skill install calebstewart/ai-slop writing-skills --agent claude-code --force

Keeping it updated

gh skill injects source-tracking metadata into the installed SKILL.md frontmatter. That is the whole reason updates work later:

gh skill update --dry-run          # report what's stale, change nothing
gh skill update --all              # update everything without prompting
gh skill update writing-skills     # just the one

gh skill update --force overwrites local edits to installed skill files with their upstream content. If you have been tweaking an installed skill in place, keep your version somewhere else — or use the symlink method below and stop fighting the tooling.

Cloning and copying

No gh extension required. Slightly more typing, no update tracking.

git clone https://github.com/calebstewart/ai-slop.git
cd ai-slop

Then copy the skill directory to wherever your agent looks:

# User scope — available in every project
mkdir -p ~/.claude/skills
cp -R skills/writing-skills ~/.claude/skills/

# Project scope — only inside one repo
mkdir -p /path/to/your/project/.claude/skills
cp -R skills/writing-skills /path/to/your/project/.claude/skills/

Symlinking, to track upstream

If you would rather follow changes than copy them:

ln -s "$PWD/skills/writing-skills" ~/.claude/skills/writing-skills

A git pull then updates the skill in place, and nothing can clobber local edits because there are none — you are reading the repo directly. The trade is that symlinked skills carry no gh skill source metadata, so gh skill update will not manage them. You are choosing git pull as your update mechanism instead.

The executable bit

artisanal-slop ships a driver at bin/artisanal-slop, and it has to stay executable. Some copy, archive, and sync paths do not preserve the mode. Whichever method you used, check that it survived:

test -x ~/.claude/skills/artisanal-slop/bin/artisanal-slop || \
  chmod +x ~/.claude/skills/artisanal-slop/bin/artisanal-slop

Troubleshooting

The skill does not appear in the / menu. The command comes from the directory name, not the name in frontmatter — so ~/.claude/skills/writing-skills/SKILL.md gives you /writing-skills. If you renamed the directory on the way in, that is your command name. A skill with user-invocable: false is hidden from the menu by design.

Claude never invokes it on its own. Check for disable-model-invocation: true in its frontmatter; artisanal-slop sets it deliberately, because something that opens and merges pull requests should not start itself. Those skills are user-invoked only.

It was invoked but seems not to have read its own reference files. Skills use progressive disclosure: the body loads on invocation, and linked reference files load only when the agent decides it needs them. That is intended, not a failure.

An installed skill stopped matching these pages. These pages are generated from the skill files in the default branch, so they describe main, not whatever tag you pinned. Compare against the SKILL.md page for the skill, which is the same file your agent reads.