promptkit

promptkit

A collection of Claude Code skills, agents, and commands for enhanced AI-assisted development workflows.

Overview

promptkit provides reusable AI tooling components that extend Claude Code's capabilities. These components encode domain expertise, enforce best practices, and automate repetitive tasks across projects.

Key Features

• Skills: Reusable capabilities for documentation, code quality, database setup, and file analysis
• Agents: Specialized AI personas for complex multi-step tasks like data profiling
• Commands: Slash commands for scaffolding and project automation

Quick Start

# Clone the repository
git clone https://github.com/austinogilvie/promptkit.git

# Copy skills to your user-level Claude Code config
cp -r promptkit/.claude/skills/* ~/.claude/skills/

# Or symlink for easy updates
ln -s $(pwd)/promptkit/.claude/skills/* ~/.claude/skills/

Contents

Skills

Skill	Description
docs-writer	Generate comprehensive technical documentation—API references, user guides, architecture docs, ADRs, and runbooks. Includes templates and validation scripts.
readme-updater	Keep README files synchronized with code changes. Detects dependency updates, new features, and configuration changes that require documentation updates.
file-categorization	Classify project files using an 8-category taxonomy (Config, Tests, Docs, Scripts, Source Code, Data, AI Tooling, Other). Includes automated categorization script.
python-dunder-enforcer	Ensure Python classes include appropriate dunder methods (__repr__, __str__, comparisons). Enforces best practices and provides field ranking heuristics.
duckdb-setup	Initialize and configure DuckDB databases. Handles extensions, connection settings, and project structure with security best practices for .gitignore.

Agents

Agent	Description
data-profiler	Perform comprehensive dataset profiling before modeling or schema design. Analyzes field structures, type distributions, null percentages, nested objects, and generates schema recommendations.

Commands

Command	Description
/create-skill	Scaffold a new Claude Code skill with proper structure (SKILL.md, scripts/, references/, assets/).
/build-research-doc	Convert a directory of interlinked Markdown research files into a single DOCX with preserved structure, links, and auto-generated TOC via Pandoc.

Structure

.claude/
├── agents/
│   └── data-profiler.md          # Dataset profiling agent
├── commands/
│   ├── create-skill.md           # Skill scaffolding command
│   ├── build-research-doc.md     # Markdown-to-DOCX converter
│   └── scripts/
│       └── build_research_doc.py # Python script for build-research-doc
└── skills/
    ├── docs-writer/
    │   ├── SKILL.md              # Main skill definition
    │   ├── workflows/            # Task-specific workflows
    │   ├── templates/            # Documentation templates
    │   ├── examples/             # Output samples
    │   └── scripts/              # Validation tools
    ├── readme-updater/
    │   ├── SKILL.md
    │   ├── references/           # Patterns and templates
    │   └── scripts/              # README checking tools
    ├── file-categorization/
    │   ├── SKILL.md
    │   ├── references/           # Pattern definitions
    │   └── scripts/              # categorize.py
    ├── python-dunder-enforcer/
    │   ├── SKILL.md
    │   └── references/           # Examples and cheatsheets
    └── duckdb-setup/
        ├── SKILL.md
        ├── references/           # Extension docs
        ├── scripts/              # init_duckdb.py
        └── assets/               # SQL templates

Usage

Using Skills

Skills activate automatically based on context when Claude Code recognizes a matching task:

User: "Create a README for this project"
→ Claude activates docs-writer skill

User: "Categorize the files in this repo"
→ Claude activates file-categorization skill

User: "Add __repr__ to my Python class"
→ Claude activates python-dunder-enforcer skill

Using Agents

Agents are invoked via the Task tool for complex, multi-step operations:

User: "Profile this JSONL dataset before I build models"
→ Claude spawns data-profiler agent for comprehensive analysis

Using Commands

Commands are invoked with the slash prefix:

/create-skill api-client "Generate typed API clients from OpenAPI specs"

/build-research-doc ~/research/my-project
/build-research-doc ./docs/analysis --output final_report.docx

Installation

User-Level (Recommended)

Install skills globally for use across all projects:

# Create Claude config directory if needed
mkdir -p ~/.claude/skills ~/.claude/agents ~/.claude/commands