AgentVault Vault Processing

Instructions

• •

  Vault structure

  • •Treat the uploaded vault as a zip containing folders of .md files.
  • •Ignore .obsidian and other hidden/system directories.
  • •Work against a per-request temp directory, not the original zip.
• •

  Markdown handling

  • •For each markdown file:
    • •Read the file content as UTF-8.
    • •Detect and parse YAML frontmatter using PyYAML.
      • •If frontmatter is missing, create a minimal one with at least title and tags keys.
    • •Optionally, use markdown-it-py or mistune to understand headings and sections before rewriting.
• •

  Entity extraction with spaCy

  • •Load the en_core_web_sm model once at startup or in a shared context.
  • •For each note body (excluding frontmatter):
    • •Run doc = nlp(text) and collect doc.ents (e.g., names, organizations, dates).
    • •Store relevant entities either in frontmatter fields (e.g., entities, people, places) or in a separate data structure used to build entities.md.
• •

  LLM-powered rewriting via OpenRouter

  • •Use OpenRouter with models like anthropic/claude-3.5-sonnet to rewrite notes.
  • •Prompt guidelines:
    • •Preserve meaning, code blocks, and links exactly.
    • •Improve structure with clear headings, lists, and sections.
    • •Make the note more explicit and self-contained for AI agents.
  • •Implement batching or concurrency (e.g., with asyncio) while respecting rate limits and timeouts.
• •

  Global artifacts

  • •After processing all notes:
    • •Create an index.md that lists all notes by title and relative path, optionally grouped or tagged.
    • •Create an entities.md summarizing entities across the vault; group them by type (people, organizations, etc.).
• •

  Output

  • •Write all transformed notes and global index files into a new directory, mirroring the original structure.
  • •The backend will zip this directory to produce the cleaned vault for download.

Examples

• •Example tasks this skill should handle
  • •Given an unzipped vault directory, produce a list of all .md files to process.
  • •For a given markdown file, show how to:
    • •Parse YAML frontmatter.
    • •Extract entities with spaCy.
    • •Prepare a prompt and call the OpenRouter API to rewrite the body.
  • •Given a collection of processed notes, outline how to build index.md and entities.md.