Erduo Skills

中文

Empowering AI Agents with structured capabilities and intelligent workflows.

Erduo Skills is an AI Agent skill library — a collection of structured, composable workflow units that agents can invoke directly. Skills cover information retrieval, content processing, image tooling, and more.

Installation

Quick Install (Recommended)

npx skills add rookie-ricardo/erduo-skills

Skills at a Glance

Skill	Description	Invocation
Daily News Report	Multi-source scraping + smart filtering → tech daily report	Agent
AK RSS Digest	Curated RSS digest with 10-point scoring	Agent / CLI
Transcript Polisher	Speech transcript → readable article, preserving original voice	Agent
Translate Polisher	4-step publication-quality translation for ZH↔EN and ZH↔JA	Agent
Web To Markdown	URL routing + Readability cleanup for clean markdown extraction	Agent / CLI
Anthropic Style Diagram	Draw architecture and flow diagrams in SVG, audit the layout, export PNG	Agent / CLI
Gemini Watermark Remover	Reverse alpha blending to remove Gemini image watermarks	CLI

---

🗞 Daily News Report

npx skills add rookie-ricardo/erduo-skills --skill daily-news-report

Autonomously fetches, filters, and summarizes high-quality tech news from multiple sources into a structured daily report.

Uses a Master-Worker architecture: the master agent orchestrates and decides, while sub-agents fetch in parallel — including a headless browser worker for JS-rendered pages.

graph TD
    User((User)) -->|Start| Master[Master Agent<br>Orchestrate/Monitor/Decide]

    subgraph Execution Layer [SubAgent Layer]
        WorkerA[Worker A<br>WebFetch]
        WorkerB[Worker B<br>WebFetch]
        Browser[Browser Worker<br>Headless Chrome]
    end

    Master -->|Dispatch Tier 1| WorkerA
    Master -->|Dispatch Tier 2| WorkerB
    Master -->|Dispatch JS Pages| Browser

    WorkerA -->|Result| Master
    WorkerB -->|Result| Master
    Browser -->|Result| Master

    Master -->|Filter & Dedup| Report[Generate Report]
    Master -->|Update| Cache[Smart Cache]

• Aggregates HackerNews, HuggingFace Papers, ProductHunt, and more across tiered sources
• 10-point scoring + deduplication (URL + content hash)
• Early stopping: halts once 20+ quality items are collected
• Outputs Markdown reports to NewsReport/

---

📰 AK RSS Digest

npx skills add rookie-ricardo/erduo-skills --skill ak-rss-digest

Curates high-quality articles from a fixed RSS/Atom feed bundle, focused on AI agents, frontier AI commentary, deep interviews, and high-signal essays.

• Predefined feed list, defaults to the latest 7 days
• 10-point scoring, only outputs items above 7
• Filters out paper summaries, vendor marketing, and SEO content
• Chinese daily-brief style output: title, score, recommendation, summary, link

# Run the fetcher directly
python skills/ak-rss-digest/scripts/fetch_today_feed_items.py --format json

# Fetch a specific day
python skills/ak-rss-digest/scripts/fetch_today_feed_items.py --date 2026-03-18 --days 1

Prompt example:

"Use $ak-rss-digest to pull the latest week's RSS posts, keep only items above 7/10, and format as a concise Chinese daily brief."

---

✍️ Transcript Polisher

npx skills add rookie-ricardo/erduo-skills --skill transcript-polisher

Polishes speech transcripts (interviews, talks, podcasts, meetings) into highly readable articles. Core principle: text polisher, not content summarizer — preserves the speaker's original words, metaphors, and personality.

• Auto-detects solo speaker vs. multi-speaker dialogue mode
• Noise reduction: removes filler words and meaningless interjections
• Homophone correction + proper noun fixes
• Semantic breathing: restructures paragraphs by meaning groups, not mechanical length
• Auto-chunks long texts (~5000 chars), processes in parallel via sub-agents

Input format:

视频标题：xxx
视频作者：xxx
视频时长：xxx

--- 字幕内容 ---
<transcript text>

Output format:

## 视频信息
Title / Author / Duration

## 导读
Core ideas summary

## 正文
Polished full text

---

🌐 Translate Polisher

npx skills add rookie-ricardo/erduo-skills --skill translate-polisher

For high-quality article translation and localization, using a 4-step workflow: Analyze → Draft → Critique → Final. Supports only ZH↔EN and ZH↔JA, and does not support direct EN↔JA translation.

• Accepts a file path, URL, or pasted text as input; URLs are fetched via r.jina.ai, and the run stops if the full article body cannot be retrieved
• Supports --from, --to, --audience, --style, and --glossary
• Performs terminology extraction, figurative language mapping, and reader-friction analysis before translating
• Includes built-in EN↔ZH and ZH↔JA glossaries, and can merge a custom glossary
• Automatically chunks long texts and translates them in parallel with sub-agents before final review
• Includes 9 built-in style presets, defaults to auto, and also accepts custom style descriptions

/translate [--from <lang>] [--to <lang>] [--audience <audience>] [--style <style>] [--glossary <file>] <source>

Prompt examples:

"Translate this article https://example.com/article" "Translate this Chinese article into English for a technical audience --style technical"

---

🔗 Web To Markdown

npx skills add rookie-ricardo/erduo-skills --skill web-to-markdown

Routes each URL to a source-aware extraction strategy and returns clean markdown for downstream agent tasks such as summarization, translation, and review.

• General pages and X/Twitter use r.jina.ai by default
• YouTube URLs use defuddle.md for transcript-oriented extraction
• WeChat, Zhihu, and Feishu use cuimp Chrome-impersonated HTTP fetching, then auto-fallback to browser extraction when needed
• Built-in generic fallback chain: if r.jina.ai fails, try direct fetch + Mozilla Readability, then browser extraction
• Supports --json metadata output (strategy, source, normalized URL, and markdown)

cd skills/web-to-markdown
npm install
node scripts/url_to_markdown.mjs <url>
node scripts/url_to_markdown.mjs <url> --json

---

📐 Anthropic Style Diagram

npx skills add rookie-ricardo/erduo-skills --skill anthropic-style-diagram

Draw technical diagrams in the Anthropic / Claude visual language, then render and save them as an image automatically. Triggers only when that style is explicitly requested.

• You write geometry and semantic classes (th / ts / box / arr / c-blue …); the renderer injects colour, theme and typography
• Audits the three failure modes that ruin diagrams: text spilling out of its box, overlapping boxes, connectors crossing unrelated nodes
• Crops the viewBox to real ink bounds — no clipped content, no dead whitespace
• Bakes computed styles into literal attributes, so the exported SVG needs no CSS and opens correctly in Figma or Illustrator
• --theme both produces a light and a dark image from one source

cd skills/anthropic-style-diagram
npm install   # skip if playwright is already installed globally
node scripts/render.mjs diagram.svg -o ~/Downloads/architecture.png
node scripts/render.mjs diagram.svg -o out.png --theme both --scale 3 --svg
node scripts/render.mjs diagram.svg --check   # audit layout only, write nothing

Design tokens live in references/design-system.md; diagram-type routing and copy-ready snippets in references/diagram-types.md.

---

🖼 Gemini Watermark Remover

npx skills add rookie-ricardo/erduo-skills --skill gemini-watermark-remover

Removes the visible Gemini AI watermark from the bottom-right corner of generated images using reverse alpha blending. Pixel-perfect restoration.

• Pure Python, only depends on Pillow
• Pre-captured alpha masks: 48px (small images) / 96px (images >1024×1024)
• Algorithm: original = (watermarked - alpha × logo) / (1 - alpha)

python skills/gemini-watermark-remover/scripts/remove_watermark.py <input-image> <output-image>

For algorithm details, see skills/gemini-watermark-remover/references/algorithm.md

---

📂 Project Structure

erduo-skills/
├── .claude/
│   └── agents/                     # Agent definitions
├── skills/
│   ├── daily-news-report/          # Daily News Report
│   │   ├── SKILL.md
│   │   ├── sources.json
│   │   └── cache.json
│   ├── ak-rss-digest/             # RSS Digest
│   │   ├── SKILL.md
│   │   ├── scripts/
│   │   └── references/feeds.opml
│   ├── transcript-polisher/        # Transcript Polisher
│   │   ├── SKILL.md
│   │   └── references/
│   ├── translate-polisher/         # Translate Polisher
│   │   ├── SKILL.md
│   │   └── references/
│   ├── web-to-markdown/            # Web To Markdown
│   │   ├── SKILL.md
│   │   ├── scripts/
│   │   └── references/
│   ├── anthropic-style-diagram/    # Anthropic Style Diagram
│   │   ├── SKILL.md
│   │   ├── scripts/render.mjs
│   │   ├── assets/diagram.css
│   │   └── references/
│   └── gemini-watermark-remover/   # Gemini Watermark Remover
│       ├── SKILL.md
│       ├── scripts/
│       ├── assets/
│       └── references/
├── NewsReport/                     # Generated report archive
├── README.md                       # Documentation (Chinese)
└── README_EN.md                    # Documentation (English)

Claude Code Installation Supplement

This repository can be registered as a Claude Code plugin marketplace.

Native Claude Code Commands

Add the marketplace:

/plugin marketplace add rookie-ricardo/erduo-skills

Then install the plugin bundle you want:

/plugin install research-workflows@erduo-skills
/plugin install writing-workflows@erduo-skills
/plugin install image-tools@erduo-skills

Bundle contents:

• research-workflows: ak-rss-digest, daily-news-report
• writing-workflows: transcript-polisher, translate-polisher, web-to-markdown
• image-tools: anthropic-style-diagram, gemini-watermark-remover

For local testing, use the repository path directly:

/plugin marketplace add ./
/plugin install research-workflows@erduo-skills

If you use the skills CLI, you can also add this repository directly:

npx skills add rookie-ricardo/erduo-skills

🤝 Contributing

Contributions welcome! Each skill is a self-contained directory under skills/, with a SKILL.md (skill definition) and related scripts/resources.

📄 License

MIT © Erduo

---

Created with ❤️ by Erduo