Enrich zotero abstracts

Enrich Zotero Abstracts

Overview

Find Zotero items with empty or short abstracts in the configured libraries, web-search for plausible abstract content, present each candidate to the user for approval, and write back to Zotero with version-conflict protection. Items updated this way receive a claude-enriched-YYYY-MM-DD tag for traceability.

When to Use

• User says "fix abstracts", "fill missing abstracts", "/zotero-context:enrich-zotero-abstracts"
• User wants to make a library more useful for downstream search

Prerequisites

• /zotero-context:setup-zotero-references has been run in this repo (.zotero-context/config.json exists).
• ZOTERO_API_KEY with write scope is set in .env — the local Zotero API is read-only for item modification.

Workflow

Step 1: List candidates

.zotero-context/bin/search_items.py --no-abstract --limit 50 --output summary

This returns items where abstractNote is missing or shorter than 50 characters, as a JSON summary with title, creators, year, DOI, url, abstract, and tags fields.

Step 2: For each candidate (one at a time)

Show the user:

Title: <title>
Authors: <creators>
Year: <year>
DOI/URL: <doi or url>
Current abstract: <abstractNote, "(empty)" if blank>

Search the web for an abstract? [y/n/skip-rest]

If yes, web-search using DOI > URL > "title authors year" as the anchor. Quote a candidate abstract back to the user (clearly attributed to the source).

Proposed abstract (from <source URL>):

<text>

Apply this to the Zotero item? [y/edit/n]

If edit, let the user revise inline. If y, proceed to Step 3.

Step 3: Apply

.zotero-context/bin/enrich_abstract.py \
    --key <key> \
    --abstract "<text>" \
    --lib <library id> \
    --type {user|group}

Watch for the result:

• {"status":"ok",...} — success.
• {"status":"conflict",...} — the item changed in Zotero between scan and PATCH. Refetch and ask the user whether to retry with the latest data.

Step 4: Report

After all candidates, summarize: N attempted, M enriched, C conflicts, S skipped. List the keys that received the claude-enriched-YYYY-MM-DD tag.

Common Mistakes

• Inventing abstracts from training data instead of web-searching for sourced text. Every proposed abstract must be attributable to a URL.
• Running with --dry-run for production usage. Dry-run is for verifying the payload only.
• Bulk-applying without per-item user approval. The user must consent to each enrichment.
• Forgetting that ZOTERO_API_KEY needs write scope; a read-scope key fails on PATCH.

Verification After Run

After enrichment, the user should see in Zotero (web or desktop):

• Updated abstractNote field on each enriched item.
• New tag claude-enriched-YYYY-MM-DD on each enriched item.
• No changes to other fields.