AgentSkillsCN

clean-links

为codiceartificiale通讯简报清理并整理Markdown文档中的链接。去除UTM参数，解析重定向，生成完整的描述，并按主题对链接进行分类。当用户提到“清理链接”“清洁链接”“为通讯简报做规范化处理”，或拥有需要处理的带链接文档时，可使用此功能。

SKILL.md
--- frontmatter
name: clean-links
description: Cleans and organizes links from markdown documents for the codiceartificiale newsletter. Removes UTM parameters, resolves redirects, generates complete descriptions, and categorizes links thematically. Use when user mentions "pulisci i link", "clean links", "normalizza per newsletter", or has a document with links that needs processing.
metadata:
  author: codiceartificiale
  version: "1.0"

Clean Links Skill

You are a link cleaning and organization specialist for the "codiceartificiale" newsletter.

Trigger Phrases

  • •"pulisci i link" / "clean the links"
  • •"elimina redirect" / "remove redirects"
  • •"normalizza per newsletter" / "normalize for newsletter"
  • •"documento con link" / "document with links"
  • •"versione pulita dei link" / "clean version of links"

Task Instructions

Step 1: Read and Analyze Input

  1. •Read the markdown file provided by the user
  2. •Extract all links from the document
  3. •Preserve all original text content (titles, descriptions, etc.)

Step 2: Link Cleaning

For each link found:

  1. •Remove UTM parameters: Strip utm_source, utm_medium, utm_campaign, utm_content, utm_term
  2. •Resolve redirects: Follow all redirects to get the final destination URL
  3. •Clean beehiiv links: If link contains https://link.mail.beehiiv.com, resolve to direct URL
  4. •Track failed links: If any link cannot be resolved, log it with reason

Step 3: Description Enhancement

For each link description:

  • •If description has < 50 words: Use web_reader MCP to access the page and generate a 50-100 word description based on actual content
  • •If description has >= 50 words: Keep unchanged
  • •Always preserve original text for everything else in the document

Step 4: Category Confirmation

Ask the user for category titles. Always propose these 5 default categories first:

Categorie proposte:

  1. •Novità e ricerca nei modelli AI - Nuovi modelli, paper di ricerca, architetture innovative
  2. •Agentic AI - Agenti autonomi, sistemi multi-agente, orchestrazione
  3. •AI Assisted Coding - Strumenti di sviluppo assistito, copilot, refactoring AI
  4. •Business e società - Impatto sociale, regulamentazione, mercato del lavoro
  5. •Robotica e Physical AI - Robotica, computer vision, embodied AI

Ask: "Confermi queste categorie o preferisci fornirne di diverse?"

Step 5: Thematic Organization

  1. •Analyze each link's title, description, and page content
  2. •Categorize each link based on the confirmed categories
  3. •Sort links by relevance within each category
  4. •Create brief category descriptions

Step 6: Output Generation

Create a new markdown file with filename {original_name}_clean.md.

See references/OUTPUT_FORMAT.md for the complete output structure.

Important Rules

  1. •Always use web_reader MCP tool to access link content
  2. •Never invent content: Base all descriptions on actual page content
  3. •Preserve original tone: Match the style of the source document
  4. •Resolve ALL redirects: Ensure final URLs are direct links
  5. •Clean beehiiv links: Always resolve link.mail.beehiiv.com to destination
  6. •Track failures: List any links that could not be processed
  7. •Generate artifact: Always create the output markdown file
  8. •Minimum description length: 50 words for auto-generated descriptions

Tool Selection

  • •Link resolution: web_reader MCP (for content extraction and redirect following)
  • •Content fetching: web_reader MCP
  • •File operations: Native Read/Write tools
  • •Web search: WebSearch for additional context if needed

Error Handling

If a link cannot be processed:

  1. •Log the original URL
  2. •Document the specific error (timeout, 404, redirect loop, etc.)
  3. •Include in "Link Non Processati" section of output
  4. •Continue processing remaining links

Example Workflow

yaml
Input: "newsletter/en/CY26W3.md"

Step 1: Read file
  → Extract: 15 links found
  → Preserve: Original titles and structure

Step 2: Clean each link
  → Remove: ?utm_source=newsletter&utm_medium=email
  → Resolve: https://link.mail.beehiiv.com/click/... → https://example.com/article
  → Result: 15 clean direct URLs

Step 3: Enhance descriptions
  → 8 links have < 50 words → Fetch content → Generate 50-100 word descriptions
  → 7 links have 50+ words → Keep unchanged

Step 4: Confirm categories
  → Propose 5 default categories (see references/DEFAULT_CATEGORIES.md)
  → User confirms or provides alternatives

Step 5: Organize
  → Analyze content and titles
  → Categorize all 15 links
  → Sort by relevance within categories

Step 6: Generate output
  → Create: "newsletter/en/CY26W3_clean.md"
  → Format: See references/OUTPUT_FORMAT.md
  → Include: List of any unprocessed links (if applicable)