Citation Management

Manage the full lifecycle of citations in a LaTeX paper.

Input

$0 — Action: harvest, validate, add, format
$1 — Path to .tex or .bib file

Scripts

Validate citations (check all cite keys resolve)

python ~/.claude/skills/citation-management/scripts/validate_citations.py \
  --tex paper/main.tex --bib paper/references.bib --check-figures --figures-dir paper/figures/

Reports: missing citations, unused bib entries, duplicate keys, duplicate sections, duplicate labels, undefined references, missing figures.

Generate BibTeX from paper database

python ~/.claude/skills/deep-research/scripts/bibtex_manager.py \
  --jsonl paper_db.jsonl --output references.bib

Search for a specific paper to add

python ~/.claude/skills/deep-research/scripts/search_semantic_scholar.py \
  --query "attention is all you need" --max-results 5 \
  --api-key "$(grep S2_API_Key /Users/lingzhi/Code/keys.md 2>/dev/null | cut -d: -f2 | tr -d ' ')"

Harvest missing citations automatically

python ~/.claude/skills/citation-management/scripts/harvest_citations.py \
  --tex paper/main.tex --bib paper/references.bib --output candidates.bib --max-rounds 10

Scans .tex for uncited claims, searches Semantic Scholar, outputs candidate BibTeX entries. Key flags: --dry-run (preview only), --verbose, --api-key

Auto-fix missing citation placeholders

python ~/.claude/skills/citation-management/scripts/validate_citations.py \
  --tex paper/main.tex --bib paper/references.bib --fix

Generates references_fixed.bib with placeholder entries for all missing citation keys.

Action: `harvest` — Iterative Citation Harvesting

Based on AI-Scientist's 20-round citation harvesting loop. For each round:

Read the current .tex draft
Identify the most important missing citation
Search Semantic Scholar via script
Select the most relevant paper from results
Extract BibTeX and generate a clean key (lastNameYearWord)
Append to .bib (skip if key exists)
Insert \cite{key} at the appropriate location
Stop when no more gaps or 20 rounds reached

Key rules:

DO NOT add a citation that already exists
Only add citations found via API — never fabricate
Cite broadly — not just popular papers
Do not copy verbatim from prior literature

Action: `validate` — Pre-Compilation Check

Run validate_citations.py to catch all issues before compilation. Fix any reported problems.

Action: `add` — Add Specific Paper

Search Semantic Scholar for the paper, extract BibTeX, clean the key, append to .bib.

BibTeX key format: firstAuthorLastNameYearFirstContentWord (e.g., vaswani2017attention)

Action: `format` — Standardize .bib

Sort entries alphabetically by key
Ensure consistent indentation (2 spaces)
Remove empty fields
Protect proper nouns with {Braces} in titles
Ensure required fields per entry type

Related Skills

Upstream: literature-search, deep-research
Downstream: paper-compilation, latex-formatting
See also: related-work-writing

citation-management