Claude Code skills, ready to pour.

Every skill is a single readable markdown file. Nothing executes on install; you can (and should) read each one before copying it in.

A skill is a folder with a SKILL.md file: YAML frontmatter on top, instructions below. Claude Code scans three places for them: .claude/skills/ in your repo (shared with your team), ~/.claude/skills/ in your home directory (yours, everywhere), and any plugins you have enabled. Each skill costs nothing until it loads; only its one-line description rides along in the session.

Here is the anatomy, from this pack’s own commit skill:

---
name: commit
description: Write a clean conventional commit from the currently
  staged changes. Use when the user says "commit".
argument-hint: "[optional extra context]"
---

Stage state right now:

!`git status --short`

Write one commit for the staged changes.

Rules:
1. Format: type(scope): summary, max 72 chars, imperative mood.
2. Read the full staged diff before writing; describe the *why*.
3. If the changes are two unrelated units of work, propose
   splitting instead of writing one vague message.
...

Two details do the heavy lifting. The description is what Claude reads to decide a skill is relevant, so it should say when to use it, not what it is. And lines like !`git status --short` execute before Claude reads the skill, injecting live output, which is how a skill sees your actual repo state instead of guessing.

Unzip the pack, then copy the skill folders into your repo for team-wide use (they ship with your git history like any other file):

your-repo/
└── .claude/
    └── skills/
        ├── commit/SKILL.md
        ├── pr-description/SKILL.md
        ├── explain/SKILL.md
        └── ... (7 more)

Or copy them to ~/.claude/skills/to have them in every project you open. That’s the entire install: no registry, no build step. Open a new session (or /clear) and type / to see them in the menu.

Explicitly: type the slash command. /commit, /explain src/lib/auth.ts, /changelog 2.1.0. Anything after the name reaches the skill as $ARGUMENTS (and positionally as $0, $1).

Automatically: just describe the task. Claude matches your request against every skill’s description and pulls in the right one; “write me a commit message” activates commit without the slash. If you want a skill to fire only when you type it, add disable-model-invocation: true to its frontmatter.

When a skill won’t activate, it is nearly always one of three things: the file isn’t at exactly skills/<name>/SKILL.md, the frontmatter is missing a description, or the session predates the install. Fresh session, check /, done.

The ten skills use these deliberately: the audits restrict tools, the workflow skills inject live git state, everything keeps its description honest. Steal the patterns for your own.