Opencite

OpenCite Skill

Discovers papers for the manuscript pipeline using the opencite CLI, then hands off each result to /literature:card for full intake.

Triggers

Invoke when the user says any of:

• "find papers on", "search for papers", "look up papers"
• "opencite", "discover references", "find references"
• "what papers should I cite for", "find me papers about"
• "canonical papers on", "foundational papers"
• "who cites this", "papers citing"

API keys — check before doing anything

uvx opencite config show 2>/dev/null | grep -i "openalex\|semantic\|pubmed"

If openalex_api_key is not set: Stop and tell the user:

OpenAlex key not configured — search and canonical will fail. Get a free key at https://openalex.org, then run: uvx opencite config init and add the key to ~/.opencite/config.toml.

PDF fetch (uvx opencite batch-fetch) and identifier lookup (uvx opencite ids) still work without any keys — they use PMC and CrossRef.

Phase 1 — Identify the project root

Look for references/ and build_doc.py in the current directory tree. If found, note the project root — search results will be staged for card intake there. If not found, proceed without a project root (results shown only, no card prompt).

Phase 2 — Run the search

Choose the right command based on user intent:

Discovery — new topic, finding what to cite

uvx opencite canonical "{topic}" --max 10

Use canonical first for established topics — returns the most-cited foundational papers.

Recent work

uvx opencite search "{query}" --max 15 --sort citations

Citation graph — papers that cite a known paper

uvx opencite cite "{doi_or_pmid}" --direction both --max 20

Single paper lookup

uvx opencite ids "{title or doi or pmid}"

Returns canonical DOI. Then hand off to /literature:card doi:{doi}.

Phase 3 — Present results

After the search, show a numbered list:

Papers found ({N} results):

  [1] {first_author} ({year}) — {journal}
      {title}
      doi: {doi}    citations: {N}    OA: {yes/no}

  [2] ...

Ask the user: "Which of these should I retrieve full text for? (say 'all', '1,3,5', or 'none')"

Phase 4 — Fetch full text and hand off to card intake

For each selected paper, run /literature:card doi:{doi} (the card skill handles full-text retrieval and extraction automatically).

If the user wants batch fetch first (multiple PDFs at once):

uvx opencite search "{query}" --max 10 -f json -o /tmp/search_results.json
uvx opencite batch-fetch --from-json /tmp/search_results.json --convert \
    -o {project_root}/references/cards/raw/

Then run /literature:card doi:{doi} for each paper to complete intake into the structured card system.

Phase 5 — Report

Tell the user:

• How many papers found
• How many selected for intake
• For each: "Card ready — run /literature:card doi:{doi} to extract" (or confirm card intake already ran)

Quick reference

Notes on opencite commands used in /literature:card

The /literature:card skill already uses opencite internally for:

• uvx opencite ids "{input}" — resolve DOI from title/PMID (Phase 1)
• uvx opencite batch-fetch --from-stdin --convert -o {dir}/ — full-text fetch (Phase 2, Tier 1)

These work without API keys (PMC + CrossRef). This skill adds the discovery step that comes before card intake.