Pocket-Knife

84 AI skills for Claude Code using your own API keys — no paid proxy, no middleman.

Generate images, create videos, synthesize speech, search the web, post tweets, and more. Each skill calls the provider API directly via curl using keys you configure once in ~/.claude/.env.

Inspired by inference.sh/skills (MIT) — same capabilities, zero intermediary costs.

---

Quick Start

1. Install the plugin

/plugin install madeinlowcode/pocket-knife

Or via marketplace:

/plugin marketplace add madeinlowcode/pocket-knife

2. Configure your API keys

Option A — Interactive CLI (recommended for first setup):

npx github:madeinlowcode/pocket-knife

The CLI asks which skill categories you want, prompts only the relevant API keys, validates them, and writes ~/.claude/.env.

Option B — Conversational skill (inside Claude Code):

/pocket-knife:setup

Claude guides you through creating or updating ~/.claude/.env step by step.

Option C — Manual:

Create ~/.claude/.env and add the keys you need:

# Google (Gemini, Imagen, Veo)
GOOGLE_API_KEY=your-key-here

# ElevenLabs (TTS, STT, voice cloning, dubbing, music)
ELEVENLABS_API_KEY=your-key-here

# fal.ai (FLUX, Kling, Wan, Seedance, upscaling, background removal)
FAL_KEY=your-key-here

# Alibaba DashScope (Qwen image models)
DASHSCOPE_API_KEY=your-key-here

# Tavily (web search)
TAVILY_API_KEY=your-key-here

# Exa (web search — alternative to Tavily)
EXA_API_KEY=your-key-here

# X/Twitter (OAuth 1.0a — all 4 required for posting)
X_CONSUMER_KEY=your-key
X_CONSUMER_SECRET=your-secret
X_ACCESS_TOKEN=your-token
X_ACCESS_TOKEN_SECRET=your-token-secret

# Google Cloud (required only for Google Veo video generation)
GOOGLE_CLOUD_PROJECT=your-project-id

3. Use any skill

/pocket-knife:nano-banana
/pocket-knife:elevenlabs-tts
/pocket-knife:web-search
/pocket-knife:flux-image

---

Skills (84 total)

Image Generation (9 skills)

Skill	Provider	Key Required
ai-image-generation	Google Imagen 4	GOOGLE_API_KEY
nano-banana	Google Gemini 3 Pro Image	GOOGLE_API_KEY
nano-banana-2	Google Gemini 3.1 Flash	GOOGLE_API_KEY
flux-image	fal.ai FLUX Dev LoRA	FAL_KEY
image-upscaling	fal.ai Topaz Upscaler	FAL_KEY
background-removal	fal.ai BiRefNet	FAL_KEY
p-image	fal.ai Pruna P-Image	FAL_KEY
qwen-image-2	Alibaba DashScope Qwen	DASHSCOPE_API_KEY
qwen-image-2-pro	Alibaba DashScope Qwen Pro	DASHSCOPE_API_KEY

Audio (9 skills)

Skill	What it does	Key Required
elevenlabs-tts	Text to speech	ELEVENLABS_API_KEY
elevenlabs-stt	Speech to text (transcription)	ELEVENLABS_API_KEY
elevenlabs-dialogue	Multi-speaker audio generation	ELEVENLABS_API_KEY
elevenlabs-sound-effects	Sound effect generation	ELEVENLABS_API_KEY
elevenlabs-voice-cloner	Clone a voice from audio sample	ELEVENLABS_API_KEY
elevenlabs-voice-changer	Transform voice in audio file	ELEVENLABS_API_KEY
elevenlabs-music	Music generation from prompt	ELEVENLABS_API_KEY
elevenlabs-dubbing	Automatic audio dubbing/translation	ELEVENLABS_API_KEY
elevenlabs-voice-isolator	Isolate voice from noisy audio	ELEVENLABS_API_KEY

Video (4 skills)

Skill	Provider	Key Required
ai-video-generation	fal.ai Kling 1.6	FAL_KEY
google-veo	Google Veo 3.1 (Vertex AI)	GOOGLE_CLOUD_PROJECT + gcloud
image-to-video	fal.ai Wan 2.2	FAL_KEY
p-video	fal.ai Seedance 1.0 Lite	FAL_KEY

LLM & Search (2 skills)

Skill	What it does	Key Required
llm-models	Multi-provider LLM (Claude, Gemini, Kimi, GLM)	Provider-specific key
web-search	Web search via Tavily or Exa	TAVILY_API_KEY or EXA_API_KEY

Social (1 skill)

Skill	What it does	Key Required
twitter-automation	Post tweets, like, retweet via X API v2	4 X_* OAuth keys

Tools (2 skills)

Skill	What it does	Key Required
python-executor	Execute Python code inline (data analysis, file processing)	None (local)
remotion-render	Render videos with React/Remotion	None (local, npx)

SDK & UI Guides (6 skills)

Skill	What it covers
javascript-sdk	Building AI apps with JS/TS — fetch, streaming, error handling
python-sdk	Building AI apps with Python — requests, async, providers
agent-ui	Building AI agent interfaces with React
chat-ui	Chat interface patterns and components
tools-ui	Tool result rendering and display
widgets-ui	Reusable AI-powered widget components

Design Guides (10 skills)

Skill	Topic
app-store-screenshots	App store screenshot composition and A/B testing
book-cover-design	Book cover design by genre, typography, color psychology
character-design-sheet	Character turnaround, expressions, proportions
data-visualization	Chart selection, color accessibility, data storytelling
email-design	Responsive email templates, dark mode, deliverability
landing-page-design	Hero sections, CTAs, fold strategy, conversion optimization
logo-design-guide	Scalability, color variations, typography pairing
og-image-design	Open Graph images (1200x630), platform previews
pitch-deck-visuals	Slide structure, data presentation, visual language
youtube-thumbnail-design	Face prominence, text contrast, CTR optimization

Video Guides (5 skills)

Skill	Topic
ai-marketing-videos	Script structure, hooks, platform-specific formats
explainer-video-guide	Problem-solution-benefit scripts, animation styles
storyboard-creation	Panel layout, shot composition, timing annotations
talking-head-production	Lighting, framing, audio, teleprompter, editing
video-ad-specs	Platform specs (YouTube/Meta/TikTok/LinkedIn)

Writing Guides (5 skills)

Skill	Topic
case-study-writing	Challenge/solution/results structure, metrics
newsletter-curation	Content sourcing, subject lines, growth tactics
press-release-writing	Inverted pyramid, headline formulas, AP style
seo-content-brief	Keyword research, search intent, meta tags
technical-blog-writing	Problem/context/solution/code structure

Social Media Guides (4 skills)

Skill	Topic
ai-social-media-content	Platform formats, content calendar, hashtag strategy
linkedin-content	Post types, hook formulas, algorithm tips
social-media-carousel	Slide structure, swipe hooks, platform dimensions
twitter-thread-creation	Hook formulas, thread structure, engagement tactics

Product Guides (4 skills)

Skill	Topic
competitor-teardown	Feature matrix, pricing analysis, SWOT
customer-persona	Demographics, jobs-to-be-done, buying triggers
product-changelog	Version formatting, audience-appropriate language
product-hunt-launch	Timing, assets, launch day checklist

AI & Automation Guides (13 skills)

Skill	Topic
prompting-guide	Prompt engineering (chain-of-thought, few-shot, etc.)
video-prompting-guide	Video model prompts (Kling, Veo, Wan, Seedance)
design-guide	AI-assisted design workflows
video-guide	Video production with AI models
writing-guide	AI content writing frameworks
social-guide	Social media strategy with AI
product-guide	AI in product development
content-guide	Content repurposing pipelines
content-repurposing	Atomic content model, format adaptation
ai-content-pipeline	One piece to multiple formats workflow
ai-podcast-creation	Script generation, voice synthesis, distribution
ai-product-photography	Background removal + image generation workflow
ai-automation-workflows	Chaining skills together, batch processing

Agent & Developer Guides (7 skills)

Skill	Topic
agent-browser	Browser automation with Playwright/Puppeteer
agentic-browser	Autonomous browser navigation with AI
agent-tools	Tool use patterns, MCP servers, function calling
ai-rag-pipeline	RAG pipeline: chunking, embeddings, vector stores
ai-voice-cloning	Voice sample preparation, ethical considerations
ai-music-generation	Genre prompting, duration control
ai-avatar-video	AI avatar creation, image-to-video workflow
dialogue-audio	Multi-speaker dialogue production
text-to-speech	Voice selection, SSML, pronunciation control
speech-to-text	Audio preparation, accuracy tips

---

How It Works

~/.claude/.env          (your API keys — configured once)
       |
       v
hooks/hooks.json        (SessionStart hook)
       |
       v
scripts/load-env.sh     (loads keys into environment — runs once per session)
       |
       v
skills/*/SKILL.md       (each skill uses curl to call provider API directly)

1. When Claude Code starts a session, the SessionStart hook runs load-env.sh
2. load-env.sh reads ~/.claude/.env and exports all variables
3. Each skill uses curl with the appropriate API key to call the provider directly
4. No proxy, no middleman — your key, your cost, your data

Security

• API keys live in ~/.claude/.env (outside any git repository)
• load-env.sh never uses set -x (would leak keys to stderr)
• load-env.sh never echoes key values
• All API-calling skills use disable-model-invocation: true — Claude won't trigger them without your explicit command
• Skills use allowed-tools: Bash(curl *) — scoped to curl only

---

Requirements

• Claude Code >= 1.0.33
• curl (macOS, Linux, WSL, Git Bash)
• bash 4+ (for the env loader hook)
• Node.js 18+ (only for CLI installer and Twitter OAuth helper)

Windows

Scripts use LF line endings (enforced via .gitattributes). Use WSL or Git Bash.

If curl resolves to PowerShell's Invoke-WebRequest, install real curl:

winget install curl.curl

---

API Keys — Where to Get Them

Provider	Key	Get it at
Google AI	GOOGLE_API_KEY	aistudio.google.com
ElevenLabs	ELEVENLABS_API_KEY	elevenlabs.io/app/settings/api-keys
fal.ai	FAL_KEY	fal.ai/dashboard/keys
DashScope	DASHSCOPE_API_KEY	dashscope.console.aliyun.com
Tavily	TAVILY_API_KEY	app.tavily.com
Exa	EXA_API_KEY	dashboard.exa.ai
X/Twitter	4 OAuth keys	developer.x.com
Google Cloud	GOOGLE_CLOUD_PROJECT	console.cloud.google.com

---

Project Structure

pocket-knife/
  .claude-plugin/
    plugin.json             # Plugin manifest (namespace: pocket-knife)
    marketplace.json        # Marketplace distribution config
  hooks/
    hooks.json              # SessionStart hook to load env
  scripts/
    load-env.sh             # Loads ~/.claude/.env into environment
    validate-plugin.sh      # Structural validation script
  commands/
    setup.md                # /pocket-knife:setup conversational wizard
  skills/
    image/                  # 9 image generation skills
    audio/                  # 9 ElevenLabs audio skills
    video/                  # 4 video generation skills
    llm/                    # LLM multi-provider skill
    search/                 # Web search (Tavily + Exa)
    social/                 # Twitter automation + OAuth helper
    sdk/                    # JS and Python SDK guides
    ui/                     # UI component guides
    guides/                 # 7 reference guides (no API key needed)
  cli/
    package.json            # npx pocket-knife init
    bin/init.js             # CLI entry point
    lib/                    # prompts, validation, env writer
  .gitattributes            # LF line endings for scripts
  LICENSE                   # MIT
  README.md

---

Contributing

1. Fork the repo
2. Create a skill in the appropriate category directory
3. Follow the SKILL.md pattern (frontmatter + instructions)
4. Ensure disable-model-invocation: true for any skill that makes API calls
5. Keep descriptions under 100 characters
6. Submit a PR

Adding a new provider

1. Add the env var to cli/lib/categories.json
2. Add validation logic to cli/lib/validate.js
3. Create skills in the appropriate category
4. Update this README

---

Comparison with inference.sh

	inference.sh	Pocket-Knife
Cost	Paid proxy (per-request billing)	Free — you pay providers directly
API calls	Via infsh app run CLI	Direct curl to provider
API keys	Managed by Inference	Your own keys in ~/.claude/.env
Skills	85+	84
License	MIT	MIT
Vendor lock-in	Yes (infsh CLI required)	None (standard curl)

---

License

MIT — see LICENSE

---

Built with Claude Code. Inspired by inference.sh.