note-tagger/CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Overview

Single-script tool that walks a hardcoded Obsidian notes folder and fills empty frontmatter fields (tags, slug, seo-title, seo-description, seo-keywords) by calling a local LM Studio LLM. Only empty fields are touched; everything else in frontmatter is preserved.

Run

pip install pyyaml requests
./tag-notes.py

Requires LM Studio running at LM_STUDIO_URL with MODEL_NAME loaded. There are no tests, linter, or build step.

Sanity-check the LLM endpoint with: curl http://localhost:1234/v1/models

Architecture

• tag-notes.py — all logic. Top-of-file constants (LM_STUDIO_URL, MODEL_NAME, NOTES_FOLDER, TAXONOMY_FILE) are the configuration surface — edit them in place.
• tag-taxonomy.yaml — tags: list injected into the LLM system prompt. Add tags here when the LLM's "new_tag_suggestions" are worth keeping.
• Flow per note: extract_frontmatter (regex + yaml.safe_load) → decide which fields are empty → slug is derived locally from filename → if any LLM-backed field is empty, send body[:20000] to call_llm, which expects a strict JSON response (with a fallback regex to unwrap ```json fences) → reconstruct_markdown writes the file back with sort_keys=False to preserve field order.
• The 20000-char cap lives inline at the call_llm invocation, not as a constant.
• Notes without --- frontmatter are skipped, not created.