home / skills / bahayonghang / my-claude-code-settings / zoterosynth

zoterosynth skill

safe

/content/skills/academic-skills/zoterosynth

This skill searches, analyzes, and synthesizes Zotero libraries via zotero-mcp to generate literature reviews and export BibTeX.

npx playbooks add skill bahayonghang/my-claude-code-settings --skill zoterosynth

Review the files below or copy the command above to add this skill to your agents.

Files (10)

SKILL.md

3.5 KB

---
name: zotero-synth
description: Search, browse, and analyze Zotero libraries via zotero-mcp to summarize papers, generate topic or collection literature reviews, synthesize evidence with backlinks, and export BibTeX. Use when tasks involve Zotero collection browsing, collection-based review writing, paper summary, or evidence synthesis.
category: Research
tags: [zotero, synthesis, research, academic, mcp]
---

# ZoteroSynth

## 1) Environment Gate
!`zotero-mcp version 2>/dev/null || echo '{"status":"error","message":"zotero-mcp is NOT installed!","install":"uv tool install \"git+https://github.com/54yyyu/zotero-mcp.git\" && zotero-mcp setup","hint":"Install zotero-mcp first, then retry this skill. DO NOT proceed without it."}'`
If the check returns an error, stop and reply:
```bash
zotero-mcp 未安装。请先运行：
uv tool install "git+https://github.com/54yyyu/zotero-mcp.git"
zotero-mcp setup
```

## 2) Entry Workflow

Arguments: `<task> [query-or-item]`

- `check` → verify MCP and Zotero connectivity. Follow `references/ERRORS.md` on failure.
- `summarize <query|key>` → single-paper flow. Read `references/WORKFLOWS.md` § 1, output per `assets/prompts/summarize.md`.
- `review <topic|collection_key|collection_name>` → multi-paper review (Map-Reduce). For collection-based review, follow `references/WORKFLOWS.md` § 2 and output strictly per `assets/prompts/review.md`.
- `synthesize <question>` → evidence synthesis. Read `references/WORKFLOWS.md` § 3, output per `assets/prompts/synthesize.md`.
- `extract <query|collection>` → Extract papers to `papers.json` using `scripts/extract_papers.py`.
- `bibtex <query|keys>` → export citations via `zotero-mcp:zotero_get_item_metadata` with `format="bibtex"`.

## 3) Output Rules
- **文献综述 (AI & Control 交叉学科核心约束)**:
    - 强制应用 `assets/prompts/review.md` 为规范输出契约。
    - **严厉禁止单篇流水账罗列**：必须按"AI算法/控制策略"或"被控对象特征"进行主题聚合综合（Thematic Synthesis）。
    - **交叉聚类引用**：每段论述必须涵盖 3-8 篇交叉文献以验证控制策略的有效性，使用聚类格式（如 `[Author1, Year1, key1; Author2, Year2, key2]`）。
    - 综述不仅提供信息抽取，必须在最后推导出明确的**方法论空白（Methodological Gap）与新架构构想**。
    - 至少覆盖 50 篇（指定主题）或 80 篇（广泛综述）文献（当数据可用时）。
    - 必须包含元素：`itemType`, `year`, `itemLink` (不可改写), `dataSource`, `dataSource url`。
    - 必须输出一张专门针对控制领域的结构化综合对比总表。
- **引用回链**：所有涉及机制、性能结论的主张必须有清晰可追溯的文献回链：`[Author, Year, item_key]`。
- 标注信息缺失点：用 `[需确认]` 标记，不可凭空推导控制参数或环境设定。
- 对于超过 10 篇的综述任务，必须采用 Map-Reduce 架构，在 Reduce 阶段执行交叉审查。

## 4) Error Handling
- Follow `references/ERRORS.md` for user-facing error responses.
- Before semantic retrieval, check index status with `zotero-mcp:zotero_get_search_database_status`.
- If full text is unavailable, fallback to metadata + annotations and state limitations.

## 5) Resource Index
- Tool matrix: `references/TOOLS.md`
- Workflow specs: `references/WORKFLOWS.md`
- Output prompts: `assets/prompts/summarize.md`, `assets/prompts/review.md`, `assets/prompts/synthesize.md`
- Advanced setup: `references/ADVANCED.md`

Overview

This skill connects to a Zotero library via zotero-mcp to search, browse, summarize, and synthesize literature. It supports single-paper summaries, collection-based literature reviews, evidence synthesis with claim backlinks, and BibTeX export. The skill enforces traceability and explicit handling of missing full text or evidence.

How this skill works

The skill calls zotero-mcp endpoints to verify connectivity, inspect collections, and fetch item metadata or full text when available. For multi-paper tasks it uses a Map-Reduce workflow to analyze many items while preserving per-paper traceability and backlinks. Outputs follow strict templates for summaries, reviews, and syntheses and flag missing evidence explicitly.

When to use it

Summarize a single paper by key or query.
Produce a collection-based literature review or topic review across many items.
Synthesize evidence across a set of papers to answer a focused research question.
Export citation metadata or BibTeX for items or collections.
Extract a batch of papers into a JSON file for downstream processing.

Best practices

Ensure zotero-mcp is installed and configured and Zotero sync is accessible before running tasks.
Provide precise collection keys or clear topic queries to target relevant papers and meet minimum coverage requirements.
For large reviews, accept the Map-Reduce flow so the skill can preserve per-paper analysis and backlinks.
State whether full text is essential; if unavailable the skill will rely on metadata and annotations and will mark limitations.
Request BibTeX export only for specific keys or narrow queries to avoid oversized outputs.

Example use cases

Generate a concise summary of a single article to prepare an annotated bibliography.
Compile a literature review for a defined topic or Zotero collection, with per-paper analysis and a final synthesis table.
Synthesize evidence on a research question and produce claim-level backlinks to supporting items.
Bulk-extract collection items into papers.json for machine processing or sharing with collaborators.
Export BibTeX entries for a subset of items to populate a manuscript reference list.

FAQ

What prerequisites are required to run this skill?

You must have zotero-mcp installed and configured with access to your Zotero library. The skill verifies connectivity before proceeding and will report errors if the tool or index is missing.

How many papers will a review cover?

The review flow targets at least 50 papers when a topic is specified and 80 papers for broad reviews, when that many relevant items exist in the library. If fewer items are available the output will note gaps and mark missing evidence.

How does the skill handle missing full text?

If full text is unavailable, the skill falls back to metadata and annotations, states the limitation explicitly, and marks claims that lack direct evidence so you can follow up.