paper-harvester-grounded-zotero

name	paper-harvester-grounded-zotero
description	Trigger this skill for literature search/review/research-design tasks; harvest DOI-grounded references, verify discipline coverage in config first, and optionally sync to Zotero after explicit user confirmation.

Paper Harvester Grounded Writing + Zotero

Auto-trigger conditions (must call this skill)

Call this skill whenever the user asks for any of the following:

• literature search / paper search / evidence collection
• literature review / related-work writing
• research proposal / research design / hypothesis development
• Chinese intents such as: “搜索文献”“文献检索”“文献综述”“研究设计”“研究框架”“提出假设”

Use this skill for:

• research proposals
• literature reviews
• related-work sections
• evidence-grounded citation workflows

Pre-flight checks (required)

1. Discipline coverage check

   • Before harvesting, verify the requested discipline can be mapped to presets in:
     • config/presets.json
     • (and, if needed) config/journals.json
   • If the discipline is not covered, stop and ask the user to update config (or provide specific journal IDs), then continue.
   • Do not silently guess an unrelated preset.

2. Zotero sync decision check

   • Ask the user whether they want references written to Zotero.
   • If yes, request/confirm Zotero parameters:
     • --zotero-library-id
     • --zotero-api-key
     • --zotero-library-type (user or group)
     • optional --zotero-collection
   • If no, run harvest without --zotero.
   • Never hardcode secrets in prompts or files.

Mandatory anti-hallucination rules

1. Never invent papers, DOIs, authors, venues, years, or URLs.
2. Harvest first, write later.
3. Prefer DOI-backed records (--require-doi).
4. If evidence is missing, explicitly state insufficient evidence.
5. Every citation must map to generated verified artifacts.

Required workflow

1) Harvest verified references

Run from repository root:

# Without Zotero sync
python grounded_harvest.py \
  --preset marketing_comprehensive \
  --mode last_n_months \
  --n 12 \
  --max-results 120 \
  --require-doi \
  --citation-formats apa,gbt,ieee

# With Zotero sync (only after user confirms Zotero write)
python grounded_harvest.py \
  --preset marketing_comprehensive \
  --mode last_n_months \
  --n 12 \
  --max-results 120 \
  --require-doi \
  --citation-formats apa,gbt,ieee \
  --zotero \
  --zotero-collection "LitReview-Seed"

2) Cite only generated outputs

• verified_references_*.json (source of truth)
• verified_references_*.md (human-readable list)
• verified_references_apa_*.txt
• verified_references_gbt7714_*.txt
• verified_references_ieee_*.txt

3) Draft with explicit evidence links

• Tie each key claim to verified references.
• Include DOI or URL where available.
• Mark uncertain claims as gaps to verify.

Zotero credentials

The script reads credentials from ~/.paper_harvester_settings.json or CLI overrides:

• --zotero-library-id
• --zotero-api-key
• --zotero-library-type

Never hardcode secrets in prompts or files.