Skills Repository

Overview

This repository contains a collection of skills — modular, self-contained packages that extend AI agent capabilities (Claude Code, Gemini CLI, CodeBuddy, etc.) with specialized workflows, domain knowledge, and reusable tools.

Repository Structure

skills/
├── README.md                  # This file
├── Makefile                   # Build & install automation
├── flake.nix                  # Nix dev environment (optional)
├── <skill-name>/              # One directory per skill
│   ├── SKILL.md               # Required: metadata + instructions
│   ├── agents/openai.yaml     # UI metadata for OpenAI-compatible agents
│   ├── scripts/               # Optional: Python/Bash scripts
│   ├── references/            # Optional: domain documentation
│   └── assets/                # Optional: templates, images, fonts
└── <skill-name>.zip           # Distributable package (generated)

Available Skills

Skill	Description
build-android	Compile, build, and run Gradle tasks for Android projects
code-quality-review	Multi-language code review with structured prioritized findings
dig	DNS lookup via unfiltered DoH (snd.kagurach.uk) when local dig is missing — public Internet and DN42
find-java	Locate Java installations, configure JAVA_HOME, resolve JDK version mismatches
framework-rev-engineer	Reverse-engineer Android framework — analyze vendor-customized services and framework APIs
jadx	Decompile and reverse-engineer Android/Java binaries (.apk, .dex, .jar, .aar)
law	Legal consultation, document analysis, and regulatory interpretation
mobile-control	ADB-based Android device control with visual verification and ReAct loop
python-dev	Python development workflow (lint, venv, type check)
skill-creator	Guide for creating effective skills
tmux-spawn	Spawn interactive CLI agents in tmux sessions with multi-agent orchestration
whistle	HTTP proxy capture/inspect/mock via whistle — network traffic analysis and mock rules
zellij-control	Programmatic session inspection, tab/pane management, screen dumping, and command execution inside Zellij

Deprecated Skills

These directories stay in the repo as historical reference. They are excluded from make all and every install target. Remove leftover copies from agent skill directories with make remove-deprecated.

Skill	Status
claude-configure	Deprecated — do not load or use
ollama-vision	Deprecated — do not load or use
prompt-enhance	Deprecated — do not load or use
verify-kmm-build	Deprecated — do not load or use
whypua	Deprecated — do not load or use

---

Installation

Prerequisites

• make (standard on macOS/Linux; on Windows use WSL or Git Bash)
• zip command-line tool

One-Key Install (for AI Agents)

AI agents can clone the repository to a temporary directory and install all skills in one command:

git clone https://git.kagurach.uk/kagura/skills.git /tmp/skills && cd /tmp/skills && make install

Install to Claude Code

make install-claude

Copies all skill directories to ~/.claude/skills/ and sets correct file permissions (files: 644, dirs: 755).

Install to Gemini CLI

make install-gemini

Copies all skill directories to ~/.gemini/skills/.

Install to CodeBuddy

make install-codebuddy

Copies all skill directories to ~/.codebuddy/skills/.

Install to Codex

make install-codex

Copies all skill directories to ~/.codex/skills/.

Manual Install (single skill)

To install only one skill, copy its directory to the agent's skills folder:

# Example: install only the python-dev skill for Claude Code
cp -R python-dev ~/.claude/skills/
chmod -R 644 ~/.claude/skills/python-dev
find ~/.claude/skills/python-dev -type d -exec chmod 755 {} +

---

Packaging Skills (for distribution)

To generate .zip packages for all skills:

make all

To package a single skill:

cd skill-creator
python scripts/package_skill.py ../python-dev

The packaging script:

1. Validates SKILL.md frontmatter, structure, and agents/openai.yaml when present
2. Creates a .zip named after the skill (e.g., python-dev.zip)

---

Creating a New Skill

Use the skill-creator skill or follow these steps manually:

Step 1 — Initialize

python skill-creator/scripts/init_skill.py <skill-name> --path . \
  --resources scripts,references \
  --interface display_name="Skill Name" \
  --interface short_description="A concise 25-64 character summary" \
  --interface default_prompt='Use $skill-name to handle this task.'

This creates <skill-name>/SKILL.md, agents/openai.yaml, and only the requested empty resource directories.

Step 2 — Edit SKILL.md

Every SKILL.md requires YAML frontmatter at the top:

---
name: my-skill
description: One sentence explaining what this skill does and when to use it.
---

Write instructions in imperative form (verb-first). Target audience is another AI agent instance.

Step 3 — Add Resources (optional)

Folder	Purpose	Example
agents/	Agent UI metadata	openai.yaml
scripts/	Deterministic Python scripts	rotate_pdf.py
references/	Domain docs loaded on demand	schema.md, api_docs.md
assets/	Output files (templates, images)	logo.png, template.html

Use only Python standard library; maintain Python 3.9+ compatibility.

Step 4 — Package

python skill-creator/scripts/package_skill.py ./<skill-name>

---

Development Environment (Nix, optional)

A flake.nix is provided with Python linters pre-installed:

nix develop
# Activates: flake8, pyright, ruff, shellcheck

---

Makefile Targets

Target	Description
make all	Package active skill directories into .zip files (excludes deprecated skills)
make install	Install active skills to all detected target agent directories (never copies deprecated skills)
make install-claude	Install active skills to ~/.claude/skills/
make install-gemini	Install active skills to ~/.gemini/skills/
make install-codebuddy	Install active skills to ~/.codebuddy/skills/
make install-codex	Install active skills to ~/.codex/skills/
make remove-deprecated	Delete leftover deprecated skill copies from detected agent directories
make clean	Remove all generated .zip files
make validate	Validate every directory containing SKILL.md, including deprecated skills

---

How Skills Are Loaded by the Agent

Skills use a three-level progressive disclosure system:

1. Metadata (name + description in frontmatter) — always in context (~100 words)
2. SKILL.md body — loaded when the skill triggers (keep under 5 000 words)
3. Bundled resources (scripts/, references/, assets/) — loaded by the agent only as needed

This design keeps context lean while making full detail available on demand.

---

Quick Validation

To validate a skill without packaging:

python skill-creator/scripts/quick_validate.py ./<skill-name>
make validate

Validation checks the exact name and description frontmatter, naming and body-size rules, TODO markers, and required agents/openai.yaml interface fields when the file exists.