Agent-Skills.md

Agent Skills: Fetch index first

Download LLM instruction documents from the repository

UncategorizedID: yuheitomi/llm-instructions/set-up-llm-docs

Repository

yuheitomi/llm-instructions
yuheitomi

Install this agent skill to your local

pnpm dlx add-skill https://github.com/yuheitomi/llm-instructions/tree/HEAD/skills.bak/set-up-llm-docs

Skill Files

Browse the full folder contents for set-up-llm-docs.

Download Skill

Loading file tree…

skills.bak/set-up-llm-docs/SKILL.md

Skill Metadata

Name
set-up-llm-docs
Description
Download LLM instruction documents from the repository

Please help me download LLM instruction documents from the yuheitomi/llm-instructions repository. Here's what I need you to do:

  1. First, fetch the index from https://raw.githubusercontent.com/yuheitomi/llm-instructions/refs/heads/main/index.md using curl or wget (prefer these over WebFetch)
  2. Parse the index to understand available documents
  3. Use the '$ARGUMENTS' keyword to find the most relevant document(s)
  4. If '$ARGUMENTS' is provided, search for documents that match the keyword in:
    • File names
    • Document titles
    • Document descriptions
  5. If no keyword is provided or multiple matches are found, ask the user to clarify which document they want
  6. Download the selected document(s) to the ./docs directory (create if it doesn't exist)
  7. Provide a summary of what was downloaded

Search Strategy:

When matching keywords, consider these common patterns:

  • "react" → React Router documentation
  • "next" or "nextjs" → Next.js documentation
  • "daisy" or "daisyui" → daisyUI documentation
  • "svelte" → Svelte 5 documentation
  • "router" → React Router documentation
  • "advanced" → Advanced React Router documentation

Download Process:

  1. Create ./docs directory if it doesn't exist using mkdir -p ./docs
  2. Download raw markdown files using curl or wget from: https://raw.githubusercontent.com/yuheitomi/llm-instructions/refs/heads/main/frameworks/[file-name]
  3. Prefer curl -o filename.md URL or wget -O filename.md URL for downloads
  4. Save files with descriptive names that include the framework/topic
  5. Confirm successful download with file size and location

Command Examples:

# Fetch index first
curl -s https://raw.githubusercontent.com/yuheitomi/llm-instructions/refs/heads/main/frameworks/index.md

# Download specific documents
curl -o ./docs/react-router-v7.md https://raw.githubusercontent.com/yuheitomi/llm-instructions/refs/heads/main/frameworks/react-router-v7.md
wget -O ./docs/nextjs-v15.md https://raw.githubusercontent.com/yuheitomi/llm-instructions/refs/heads/main/frameworks/nextjs-v15.md
curl -o ./docs/svelte5.md https://raw.githubusercontent.com/yuheitomi/llm-instructions/refs/heads/main/frameworks/svelte5.md

Error Handling:

  • If the keyword is ambiguous (multiple matches), list all matches and ask user to choose
  • If no matches found, show available documents and ask for clarification
  • If curl or wget download fails, provide clear error message and suggest alternatives
  • If ./docs directory cannot be created, suggest alternative location
  • Only use WebFetch as a fallback if curl/wget are not available or fail

Tool Priority:

  1. First choice: Use curl or wget for all HTTP requests
  2. Fallback: Use WebFetch only if curl/wget fail or are unavailable
  3. Always: Use mkdir -p ./docs to ensure directory exists