pkit

Multi-source AI prompt bookmark manager

Features • Quick Start • Installation • Usage • Configuration

pkit is a CLI tool for subscribing to, organizing, and using AI prompts from multiple GitHub sources. It provides full-text search, interactive browsing, and seamless piping to execution tools like claude, llm, fabric, and mods.

Status: WIP - Phase 1 MVP in active development

---

Interactive TUI and Web for browsing and searching prompts

---

Features

• Multi-source aggregation: Subscribe to GitHub repositories (Fabric, awesome-chatgpt-prompts, etc.)
• Full-text search: Fast bleve-powered indexing and search
• Interactive TUI: Real-time filtering with Bubbletea
• Web Interface: Browse prompts in your browser with a clean, responsive UI
• Bookmarking: Save prompts with custom aliases and tags
• Unix pipe-friendly: Clean output protocol for piping to execution tools
• Tool-agnostic: Works with claude, llm, fabric, mods, and any CLI tool
• Private GitHub Repos supported: include your own prompt databases

Quick Start

Installation

# Clone repository
git clone https://github.com/whisller/pkit.git
cd pkit

# Build binary
make build

# Install to GOPATH/bin
make install

# Or download pre-built binary from releases
# https://github.com/whisller/pkit/releases

Usage

Basic Usage

# Subscribe to a prompt source
pkit subscribe fabric/patterns

# or multiple sources at once
pkit subscribe fabric/patterns f/awesome-chatgpt-prompts

# Interactive browser (TUI) where you can interactively do all actions
pkit find

# Start web server, where you can use your browser to manage your prompts
pkit web --port 8080
# Open http://127.0.0.1:8080 in your browser

# Search for prompts
pkit search "code review"

# Bookmark a prompt
pkit save fabric:code-review --as review --tags dev,security

# Get prompt content (pipe to execution tools)
pkit get review | claude -p "analyse me ~/main.go"

# Shorthand form
pkit review | claude -p "analyse me ~/main.go"

Bookmarks and Tags

# Add bookmark with alias and tags
pkit bookmark add fabric:code-review --alias review --tags dev,security

# List all bookmarks
pkit bookmark list

# Tag an existing prompt
pkit tag add fabric:summarize productivity,writing

# Search by tags
pkit search --tags dev,security

Web Interface

# Start web server on custom port
pkit web --port 3000 --host 0.0.0.0

# Access at http://localhost:3000
# Features:
# - Browse all prompts with filters
# - Search in real-time
# - Manage bookmarks and tags
# - Copy prompt content with one click

Piping to AI Tools

# With Claude CLI
pkit get fabric:summarize | claude < article.txt

# With Simon Willison's llm
cat code.go | pkit review | llm -m claude-3-opus

# With fabric
pkit get fabric:extract-wisdom | fabric --stream

# With mods
echo "Explain Docker" | pkit get teacher | mods

What Problem Does This Solve?

Current workflow requires managing multiple repositories manually:

cd ~/repos/fabric && git pull
cd ~/repos/awesome-chatgpt-prompts && git pull
grep -r "summarize" ../fabric ../awesome-chatgpt-prompts
# Copy-paste prompt manually

With pkit:

pkit subscribe fabric/patterns f/awesome-chatgpt-prompts
pkit search summarize
pkit save fabric:summarize --as sum
cat article.txt | pkit sum | llm -m claude-3-sonnet

Configuration

pkit stores all data in ~/.pkit/:

~/.pkit/
├── config.yml          # Main configuration
├── bookmarks.yml       # Bookmarked prompts
├── aliases.yml         # Custom aliases
├── tags.yml           # Prompt tags
├── sources/           # Cloned repositories
│   ├── fabric/
│   └── awesome-chatgpt/
└── index/             # Bleve search index
    └── prompts.bleve/

Configuration File

Default ~/.pkit/config.yml:

sources:
  - id: fabric
    url: https://github.com/danielmiessler/fabric
    format: fabric_pattern
    prompt_count: 150
    commit_sha: abc123...
    last_indexed: 2024-01-15T10:30:00Z

display:
  table_style: rounded    # simple, rounded, unicode
  date_format: relative   # relative, rfc3339, short

search:
  default_max_results: 50

Architecture

How It Works

1. Subscribe: Clone GitHub repositories to ~/.pkit/sources/
2. Parse: Extract prompts based on repository format (Fabric patterns, awesome-chatgpt CSV, etc.)
3. Index: Build full-text search index using Bleve
4. Search: Query indexed prompts with fuzzy matching
5. Bookmark: Save frequently used prompts with aliases and tags
6. Execute: Pipe prompt content to AI tools

Supported Sources

Source	Format	Example
Fabric	Markdown patterns	pkit subscribe fabric/patterns
awesome-chatgpt-prompts	CSV	pkit subscribe f/awesome-chatgpt-prompts
Custom Markdown	Frontmatter-based	Any GitHub repo with markdown files

Development

Prerequisites

• Go 1.23+
• Git 2.x
• Make (optional)

Build & Test

# Build
make build

# Run tests
make test

# Run linters
make lint

# Generate coverage report
make coverage

# Clean artifacts
make clean