Paper Finder

Use this skill when you want to discover recent arXiv papers for a specific research profile and write the results as Obsidian-compatible Markdown notes.

Run Retrieval

Run the CLI from the repository root:

python skills/paper-finder/find_papers.py \
  --profile path/to/research-interest.json \
  --output path/to/obsidian/inbox

Optional semantic ranking:

python skills/paper-finder/find_papers.py \
  --profile path/to/research-interest.json \
  --output path/to/obsidian/inbox \
  --semantic

Semantic settings can come from either source:

• config/api_keys.json with semantic_model.api_base_url, semantic_model.api_key, semantic_model.model
• Environment variables:
  • STABLE_JARVIS_SEMANTIC_API_BASE_URL
  • STABLE_JARVIS_SEMANTIC_API_KEY
  • STABLE_JARVIS_SEMANTIC_MODEL

You can still pass --semantic-config path/to/semantic-config.json for advanced overrides. The JSON file should match stable_jarvis.paper_finder.semantic.SemanticConfig.

Build or Refresh a Profile

If the user does not already have a profile JSON, use Zotero MCP read tools to collect evidence first:

1. zotero_list_collections
2. zotero_profile_evidence

Then write a JSON profile matching the same structure used by config/research-interest.example.json:

• profile_id
• profile_name
• zotero_basis
• retrieval_defaults
• interests[]

Prefer short method_keywords and only a small number of query_aliases per interest.

Write it in temp/paper-finder/research-interest.json, if there is no such directory, create it.

Enrich Candidate Notes

After the Python retrieval run finishes, read the generated Obsidian note and use Zotero MCP read tools to gather nearby library evidence. Then update the note's frontmatter and the Why It Matters, Quick Takeaways, and Caveats sections using the prompt in skills/paper-finder/prompts/enrich-candidate.prompt.txt.