Claude Skills by Yrzhe

yrzhe Claude Code Skills

A collection of powerful Claude Code skills and plugins by @yrzhe.

Installation

Option 1: Via Plugin Marketplace

Add this marketplace to your Claude Code:

/plugin marketplace add yrzhe/claude-skills

Then install any plugin you want:

/plugin install intelligent-web-scraper@yrzhe-skills

Option 2: Manual Installation

If the plugin command doesn't work, you can manually copy the skill files:

1. Clone or download this repository
2. Copy the skill folder to your Claude skills directory:

macOS / Linux:

cp -r plugins/intelligent-web-scraper/skills/intelligent-web-scraper ~/.claude/skills/

Windows (PowerShell):

Copy-Item -Recurse plugins\intelligent-web-scraper\skills\intelligent-web-scraper $env:USERPROFILE\.claude\skills\

3. Restart Claude Code to load the new skill

Available Plugins

translate-view

Translate outgoing prompts into the agent's target language and stream reply translations into a right-hand pane while keeping the original reply. Includes separate language targets, local/LLM/Jev detection, custom translation prompts, and private API configuration.

/plugin install translate-view@yrzhe-skills
/reload-plugins
/translate settings

Requires Claude Code 2.1.287+, Python 3, a Chat Completions-compatible translation endpoint, and fullscreen mode for right docking. Setup and data-flow guide.

reply-view

Preview pasted images and the images/videos linked in Claude's last reply above the terminal prompt. Open websites, localhost URLs, or media with a button; copy all visible assistant text from the last turn with one click.

/plugin install reply-view@yrzhe-skills
/reload-plugins

Requires Claude Code 2.1.287+, Python 3, and a kitty-graphics terminal for thumbnails. FFmpeg enables video posters; videos open in your default player. Setup, data access, and development guide.

intelligent-web-scraper

Self-learning intelligent web scraper agent that automatically analyzes page structure, handles pagination, anti-blocking, and discovers article series. No user configuration needed - AI decides everything.

Features:

• Intelligent page analysis and data extraction
• Smart pagination handling (page numbers, infinite scroll, load more)
• Detail link following for complete data
• Anti-blocking with adaptive delays
• Series/chapter discovery
• Self-learning system that remembers successful patterns
• Resume capability for interrupted scrapes
• Concurrent scraping with rate limiting
• Local browser support (preserve login sessions)

Usage:

/intelligent-web-scraper

Then provide a URL to scrape and let the AI handle everything.

lenny-advisor

Product and business diagnostic advisor powered by distilled wisdom from 289 Lenny's Podcast guests and 348 newsletter articles. Not a knowledge dump — an active advisor that diagnoses your real problem before delivering expert frameworks.

Features:

• Diagnose → Probe → Deliver methodology (asks before answering)
• 18 topic areas: growth, pricing, PMF, positioning, hiring, leadership, metrics, fundraising, marketplace, AI strategy, and more
• 40 deep expert profiles (Shreyas Doshi, Elena Verna, April Dunford, Rahul Vohra, etc.)
• 3-layer progressive loading (minimal token usage)
• Companion workflow with dbskill and gstack

Install:

/plugin install lenny-advisor@yrzhe-skills

Manual install:

cp -r plugins/lenny-advisor/skills/lenny-advisor ~/.claude/skills/

The skill activates automatically when you discuss product decisions, business strategy, growth, pricing, or any product/business topic.

persona-sim

Simulate feedback from census-grounded virtual populations. Panel-score your product / copy / pricing, predict votes with IPF post-stratification, or run what-if social-sandbox experiments — before paying for real user research. Backed by NVIDIA Nemotron-Personas (1M US Census-aligned synthetic people) and methodology from Park et al. 2024.

Features:

• 4 skills in one plugin: persona-sim (core engine) + product-feedback-sim (SGO A/B ranking) + vote-predict (policy/voting with IPF) + social-sandbox (what-if experiments)
• SGO (Semantic Gradient Optimization) with anchored counterfactuals — causal attribution, not independent re-scoring
• Persuadable-middle identification (score 4-7 only) to avoid wasting LLM calls on extremes
• IPF post-stratification to reweight panels to real-population marginals (PUMS-ready)
• Bias audit with 4 probe types (framing, acquiescence, order, authority) — flags when the simulation under-represents human biases
• 6-suite eval score card (GSS attitudes, Big Five norms, test-retest, diversity, demo-correlation, bias-audit)
• HuggingFace streaming — no local dataset download required for MVP
• Multi-provider LLM routing: native Anthropic, Anthropic-compatible gateways, OpenAI-compatible gateways

Install:

/plugin install persona-sim@yrzhe-skills

Manual install:

cp -r plugins/persona-sim/skills/persona-sim ~/.claude/skills/
cp -r plugins/persona-sim/skills/product-feedback-sim ~/.claude/skills/
cp -r plugins/persona-sim/skills/vote-predict ~/.claude/skills/
cp -r plugins/persona-sim/skills/social-sandbox ~/.claude/skills/

First-time setup: see plugins/persona-sim/skills/persona-sim/SETUP.md — you need to create ~/.claude/data/personas/config.json (from the provided config.example.json) with your LLM API key, and a venv for Python dependencies.

Usage examples:

• "Score this landing page copy with 30 software developers" → auto-triggers product-feedback-sim
• "Predict US support for a $22 minimum wage, by age and education" → auto-triggers vote-predict
• "If AI copilots got regulated tomorrow, what would developers do?" → auto-triggers social-sandbox
• Direct Python use: from persona_sim import sampler, sim_engine

design-distiller

Scrape any website's design system into a structured Design MD + decomposed design tokens. Ships with 55 pre-analyzed brand references (Vercel, Stripe, Linear, Notion, Claude, Figma, Airbnb, Spotify, and more) and an 8-dimension Digest Pool for mix-and-match composition.

Features:

• 4-phase pipeline: Scrape → Analyze → Generate → Digest
• Multi-tier browser support: Browser Use Cloud / Playwright / Chrome Headless / WebFetch fallback
• 9-module Design MD format following awesome-design-md standard
• 55 pre-loaded brand references with full design system documentation
• 8-dimension Digest Pool (typography, colors, spacing, components, depth, motion, layouts, philosophy)
• Compose command: mix-and-match from Digest Pool to generate new design systems
• Machine-friendly cross-reference tags for programmatic matching
• Confidence tagging: High (CSS var) / Medium (computed) / Low (visual estimate)

Install:

/plugin install design-distiller@yrzhe-skills

Manual install:

cp -r plugins/design-distiller/skills/design-distiller ~/.claude/skills/

Usage:

/design-distiller https://vercel.com              # Full pipeline
/design-distiller compose "Vercel typography + Stripe colors + Linear components"
/design-distiller compare vercel stripe            # Side-by-side comparison
/design-distiller list                             # List all 55 references

seed

Build-in-public activity recorder. A Stop hook mechanically logs every Claude Code turn (user prompt + tools used + full assistant output) to a per-session markdown file. When you're ready to tweet, /seed reads the log and helps you synthesize draft tweets from real evidence — no more "wait, what did I actually do today?"

Features:

• Zero-cost Stop hook — mechanical turn capture, no LLM calls, no scoring, zero latency
• De-duped by user-prompt uuid (Stop hook fires per-turn, so dedup matters)
• Per-session markdown logs with full assistant output + tool summaries
• /seed shot — screenshot capture bound to the current session (interactive window-pick OR headless Chrome URL mode)
• /seed — Claude reads the log and proposes tweet drafts tied to real evidence (skips routine sessions)

Install:

/plugin install seed@yrzhe-skills

Manual install:

cp -r plugins/seed/skills/seed ~/.claude/skills/seed
chmod +x ~/.claude/skills/seed/scripts/*.py

Hook setup (one-time, see plugins/seed/skills/seed/README.md for full instructions):

cp ~/.claude/skills/seed/hooks/capture-session-seed.sh ~/.claude/hooks/
chmod +x ~/.claude/hooks/capture-session-seed.sh

Then add to ~/.claude/settings.json:

{ "hooks": { "Stop": [{ "matcher": ".*", "hooks": [{ "type": "command", "command": "~/.claude/hooks/capture-session-seed.sh" }] }] } }

Usage:

/seed                                # synthesize current session → tweet drafts
/seed shot                           # interactive window pick
/seed shot localhost:3000            # headless Chrome screenshot
/seed list                           # list all session logs
/seed done                           # archive current session

chef

Cooking assistant for Chinese and Western cuisine — recipes are pulled from real sources, never fabricated. LLMs have no taste buds; specific quantities, timings, and heat levels must come from real recipe data or be fetched live. Ships with 604 pre-fetched recipes as evidence.

Sources (all real): xiachufang, douguo, meishij (下厨房/豆果美食/美食杰 — Chinese), allrecipes, BBC Good Food, Food Network, TheMealDB, Serious Eats, Epicurious, and 10+ others. Zero fabrication — if a source doesn't have it, the agent says "let me fetch it" before answering.

Features:

• Hard rule: no data = no answer. Every quantity/technique traceable to a source URL
• Chinese cuisines: 川/粤/鲁/苏/闽/浙/湘/徽/家常
• Western cuisines: 意/法/美式/地中海/英伦 + categories (mains/baking/soups/salads/breakfast)
• Recipe walkthroughs with ingredients (name + gram weight), steps (timing/heat/key moments), doneness cues, common pitfalls
• Ingredient-based suggestions (prioritize recipes covering ≥2 of what you have)
• Pairing analysis via local co-occurrence + theory (not vibes)
• Nutrition lookup (Open Food Facts) + substitution advice
• 604 pre-cached recipes in data/recipes/*.md — search before fetch

Install:

/plugin install chef@yrzhe-skills

Manual install:

cp -r plugins/chef/skills/chef ~/.claude/skills/chef

Usage:

"怎么做西湖醋鱼"            → local search → walkthrough with source URL
"家里有鸡蛋西红柿土豆能做啥"  → ingredient-match suggestions
"X 和 Y 搭不搭"            → co-occurrence + pairing theory
"how do I make carbonara"   → allrecipes/BBC Good Food lookup
"这菜多少卡路里"            → nutrition from frontmatter or OpenFoodFacts

Contributing

Feel free to open issues or submit pull requests to improve these skills.

License

MIT License - see individual plugins for details.

Author

yrzhe - @yrzhe_top