literature-search

Literature Search — Unified WOS + CNKI Entry Point

This is the only literature search entry point in the Nature Paper Suite. All literature queries, whether English or Chinese, are routed through this skill.

Two internal engines, never called directly from outside:

literature-search/
├── SKILL.md                    ← this file (you are here)
├── wos-monitor/                ← WOS engine (8 sub-skills, 86 journals)
│   ├── workflow.md              ← WOS engine overview + sub-skill index
│   ├── wos-journals.txt
│   ├── agents/wos-researcher.md
│   ├── skills/                 (wos-search, parse-results, navigate-pages,
│   │                            paper-detail, download, export, excel-export, dom)
│   ├── scripts/                (excel_manager.py, journal_monitor.py)
│   └── references/             (wos-api.md, wos-selectors.md, publisher-selectors.md)
└── cnki-skills/                ← CNKI engine (11 sub-skills, 124 journals)
    ├── cnki-journals.txt
    ├── agents/cnki-researcher.md
    ├── skills/                 (cnki-search, advanced-search, parse-results,
    │                            navigate-pages, paper-detail, download,
    │                            journal-search, journal-index, journal-toc,
    │                            export, excel-export)
    ├── workflow.md              ← CNKI engine overview + sub-skill index
    └── scripts/excel_manager.py

Requires: institutional WOS and CNKI access via IP. Chrome browser with manual IP-based login completed.

---

Execution Routing (MANDATORY)

This SKILL.md is the only auto-discovered skill file in literature-search. All execution instructions live in workflow.md files that are not auto-discovered by Codex. Before executing any operation, you MUST read exactly the workflow file(s) listed in the routing table below. Do NOT read more files than needed, and do NOT improvise execution steps from memory — the workflow files contain the exact JS code, API endpoints, DOM selectors, and Chrome CDP calls required.

WOS Routing Table

All paths relative to literature-search/.

User intent	Read this file
WOS keyword search, topic search	wos-monitor/skills/wos-search/workflow.md
Browse / next page / previous page	wos-monitor/skills/wos-navigate-pages/workflow.md
View paper full record (abstract, refs)	wos-monitor/skills/wos-paper-detail/workflow.md
Download PDF	wos-monitor/skills/wos-download/workflow.md
Export to Zotero / RIS / BibTeX	wos-monitor/skills/wos-export/workflow.md
Export / append to Excel	wos-monitor/skills/wos-excel-export/workflow.md
Parse current results page (auto after search)	wos-monitor/skills/wos-parse-results/workflow.md
WOS anti-detection rules, API patterns	wos-monitor/agents/wos-researcher.md
WOS page CSS selectors reference	wos-monitor/references/wos-selectors.md
Publisher page selectors (Elsevier, Springer...)	wos-monitor/references/publisher-selectors.md

For any WOS operation, the routing table above is the only lookup you need. Do NOT read wos-monitor/workflow.md unless you need the full engine overview.

CNKI Routing Table

All paths relative to literature-search/.

User intent	Read this file
CNKI keyword search	cnki-skills/skills/cnki-search/workflow.md
Advanced search (author / journal / date / source filter)	cnki-skills/skills/cnki-advanced-search/workflow.md
Browse / next page / sort results	cnki-skills/skills/cnki-navigate-pages/workflow.md
View paper details (abstract, keywords, fund)	cnki-skills/skills/cnki-paper-detail/workflow.md
Download PDF / CAJ	cnki-skills/skills/cnki-download/workflow.md
Export to Zotero / RIS / GB/T 7714	cnki-skills/skills/cnki-export/workflow.md
Export / append to Excel	cnki-skills/skills/cnki-excel-export/workflow.md
Parse current results page (auto after search)	cnki-skills/skills/cnki-parse-results/workflow.md
Find journal by name / ISSN / CN	cnki-skills/skills/cnki-journal-search/workflow.md
Query journal indexing & impact factor	cnki-skills/skills/cnki-journal-index/workflow.md
Browse journal table of contents	cnki-skills/skills/cnki-journal-toc/workflow.md
CNKI anti-bot rules, page management	cnki-skills/agents/cnki-researcher.md

For any CNKI operation, the routing table above is the only lookup you need. Do NOT read cnki-skills/workflow.md unless you need the full engine overview.

Rules

1. One intent → one file. Read only the file(s) that match the current operation.
2. Read before execute. Always read the workflow file before running any Chrome CDP call.
3. Never improvise. If you haven't read the workflow file, you don't know the correct selectors or API calls.
4. Compound operations. A search-then-export workflow reads two files sequentially, not at once.

---

Pre-Step: Clarify Your Research Direction First

This skill should NOT be the first step of your workflow.

Before searching for literature, use deep-research (socratic mode) to clarify your research question. A well-defined RQ makes literature search dramatically more efficient and focused.

Once your direction is clear, return here to execute the search.

---

Quick Start

Just say what you want to find. The skill auto-selects the engine:

You say	Engine used
"Search for urban resilience papers since 2024"	WOS (English keywords)
"帮我搜城市韧性的最新论文"	CNKI (Chinese keywords)
"Search WOS for climate adaptation"	WOS (explicit)
"在知网搜水资源管理"	CNKI (explicit)
"Search both WOS and CNKI for ecosystem services"	Both (explicit)
"Search arXiv for graph neural network papers"	arXiv (explicit opt-in)
"Search Google Scholar for urban governance"	Google Scholar (explicit opt-in)

---

Source Policy

WOS and CNKI are the default literature sources. No automatic fallback to any other database. arXiv and Google Scholar are available only when explicitly requested by the user.

Source	Engine	Journals	Access
Web of Science Core Collection	wos-monitor/	86 journals (wos-journals.txt)	Chrome CDP, IP login
CNKI (中国知网)	cnki-skills/	124 journals (cnki-journals.txt)	Chrome CDP, IP login
arXiv (opt-in only)	Codex browsing	Preprints	Explicit: "search arXiv for ..."
Google Scholar (opt-in only)	Codex browsing	Broad discovery	Explicit: "search Google Scholar for ..."

If WOS or CNKI is unavailable (IP blocked, session expired), stop and report rather than falling back to another source.

---

Language-Based Auto-Selection

When no source is specified, the engine is chosen by keyword language:

Keyword Language	Auto-Selected Engine	Reason
English (ASCII only)	wos-monitor/ (WOS)	WOS indexes English journals
Chinese (any CJK characters)	cnki-skills/ (CNKI)	CNKI is the primary index for Chinese journals
Mixed English + Chinese	Ask user, or search both engines
Any	arXiv / Google Scholar	Never auto-selected; only when user explicitly says "arXiv" or "Google Scholar"

---

Keyword Logic

User says	Operator	Query
"A B" (space only, no connector)	OR	A OR B
"A and B"	AND	A AND B
"A + B"	AND	A AND B
"A;B" / "A,B"	OR	A OR B

Default rule: Space-separated keywords = OR (broader search). Use explicit connector (and/+) for AND.

---

Keyword Expansion for WOS

Universal rule: before every WOS search, expand each concept into multiple English terms — regardless of whether the input was Chinese or English. A single keyword is a single point of failure.

This applies to ALL WOS searches:

• Chinese keywords → translate to English + add synonyms
• English keywords → add near-synonyms (no translation needed)
• Abbreviations → verify meaning + add full form + exclude false matches

Step 1: Translate (if input is Chinese)

Convert each Chinese concept to its primary English academic term. Do NOT translate word-for-word — use established terminology in the field.

Step 2: Expand synonyms (always, for every concept)

For each English concept, generate 2-4 search terms:

• the primary term (from translation or user input)
• a narrower or more specific synonym
• a broader or adjacent synonym
• an alternative phrasing commonly used in the target journals

Combine synonyms for the same concept with OR.

Step 3: Handle abbreviations (when applicable)

If any English term has a known academic abbreviation:

1. Include the abbreviation in the OR group: "Lorenz asymmetry coefficient" OR "LAC"
2. Check whether this abbreviation is ambiguous (shared with unrelated concepts). For example, LAC can mean:
   • Lorenz Asymmetry Coefficient
   • Latin America and the Caribbean
   • Lactate
   • Local Area Coverage
3. If ambiguous, add a NOT exclusion to filter out the false meaning:

   TS=("Lorenz asymmetry coefficient" OR ("LAC" NOT ("Latin America" OR Caribbean OR lactate)))
   

4. If the abbreviation is unambiguous in the target literature, no NOT is needed.

Abbreviation disambiguation table (check before search):

Abbreviation	Possible meanings	When searching for...	Add NOT for...
LAC	Lorenz asymmetry coefficient / Latin America and Caribbean / Lactate	Lorenz asymmetry	"Latin America", Caribbean, lactate
ML	Machine learning / Maximum likelihood / Milliliter	Machine learning	"maximum likelihood" (if relevant)
AI	Artificial intelligence / Artificial insemination	Artificial intelligence	Usually unambiguous in CS journals

When in doubt, check the abbreviation on the target journal's typical usage or add the NOT exclusion. A narrower result is better than a noisy one.

Example 1: Chinese input → translation + synonyms + abbreviation

Input: 在WOS中搜索洛伦兹不对称系数评估生态系统

Auto-expansion:

Chinese	English synonyms + abbreviation	Exclusion
洛伦兹不对称系数	"Lorenz asymmetry coefficient", "Lorenz asymmetry", "LAC"	NOT ("Latin America" OR Caribbean)
生态系统	"ecosystem", "ecological system", "ecosystem service*"	—

Final WOS query:

TS=("Lorenz asymmetry coefficient" OR "Lorenz asymmetry" 
    OR ("LAC" NOT ("Latin America" OR Caribbean OR lactate)))
AND TS=(ecosystem* OR "ecological system" OR "ecosystem service*")
AND TS=(evaluat* OR assess* OR effect* OR measur*)

User confirmation:

Keyword expansion:
  洛伦兹不对称系数 → "Lorenz asymmetry coefficient" / "Lorenz asymmetry" / "LAC"
    ⚠ "LAC" is ambiguous — excluding "Latin America", Caribbean, lactate
  生态系统 → ecosystem* / "ecological system" / "ecosystem service*"
  评估/效果 → evaluat* / assess* / effect* / measur*

Search now?

Example 2: English input → synonyms only (no translation)

Input: Search WOS for water governance resilience

Auto-expansion:

English	Synonyms	Exclusion
water governance	"water governance", "water management", "integrated water resource management"	—
resilience	"resilience", "adaptive capacity", "robustness"	—

Final WOS query:

TS=("water governance" OR "water management" OR "integrated water resource management")
AND TS=("resilience" OR "adaptive capacity" OR "robustness")

Reasoning

WOS topic search matches only what you type. Authors use different terms for the same concept — "urban resilience" in one paper, "city resilience" in another, "urban adaptive capacity" in a third. Multiple synonyms cast a wider net without sacrificing precision because the journal filter already narrows the domain. Abbreviation disambiguation prevents noise from homographic abbreviations.

Always report the expansion to the user before executing.

---

WOS Engine (wos-monitor/)

Internal Structure

wos-monitor/
├── workflow.md                  ← WOS engine overview + sub-skill index
├── wos-journals.txt             ← 86-journal list (external, editable)
├── agents/
│   └── wos-researcher.md        ← Agent orchestrator
├── skills/
│   ├── wos-search/workflow.md      ← Search + journal filter (reads wos-journals.txt)
│   ├── wos-parse-results/workflow.md
│   ├── wos-navigate-pages/workflow.md
│   ├── wos-paper-detail/workflow.md
│   ├── wos-download/workflow.md
│   ├── wos-export/workflow.md
│   └── wos-excel-export/workflow.md
├── scripts/
│   ├── excel_manager.py         ← Excel read/append/dedup
│   └── journal_monitor.py       ← Batch WOS API search + Excel
└── references/
    ├── wos-api.md               ← WOS internal API reference
    ├── wos-selectors.md         ← WOS page CSS selectors
    └── publisher-selectors.md   ← Publisher page selectors

Sub-skill routing is defined in the WOS Routing Table above — always consult that table before any WOS operation.

Journal-List Filtering

All WOS searches are filtered against the journal list in wos-journals.txt. Two filtering methods are available:

• "Search within results": After performing a topic search on WOS, use the "Search within results" sidebar to add journal-name constraints.
• Advanced query: Build a single query with SO=(...) clause containing the journal names from wos-journals.txt.

Workflow (7 phases)

1. Setup: Load WOS URL, parse keywords and time range, load existing Excel
2. Refine: Add keywords via WOS sidebar "Search within results"
3. Parse: Extract title, authors, date, journal, DOI, WOS URL
4. Deduplicate: Against existing Excel by DOI or (title, journal, year)
5. Detail: Click through to full record for abstracts (new papers only)
6. Translate: Optional Chinese translation of titles/abstracts
7. Export: Append to Excel, report totals

---

CNKI Engine (cnki-skills/)

Internal Structure

cnki-skills/
├── cnki-journals.txt            ← 124-journal list (FMS Tier 1 & 2, external)
├── agents/
│   └── cnki-researcher.md       ← Unified agent
├── skills/
│   ├── cnki-search/             ← Keyword search
│   ├── cnki-advanced-search/    ← Field-filtered search (北大核心+CSSCI fixed)
│   ├── cnki-parse-results/      ← Parse results page into structured data
│   ├── cnki-navigate-pages/     ← Pagination and sort
│   ├── cnki-paper-detail/       ← Full paper details (abstract, keywords)
│   ├── cnki-download/           ← PDF/CAJ download
│   ├── cnki-journal-search/     ← Journal lookup by name/ISSN/CN
│   ├── cnki-journal-index/      ← Journal indexing status and impact factor
│   ├── cnki-journal-toc/        ← Browse journal table of contents
│   ├── cnki-export/             ← Export to Zotero or RIS
│   └── cnki-excel-export/       ← Export to Excel with dedup
└── scripts/
    └── excel_manager.py         ← Excel read/append/dedup

Sub-skill routing is defined in the CNKI Routing Table above — always consult that table before any CNKI operation.

Workflow

1. Determine intent → look up the CNKI Routing Table → read the matched workflow file
2. Execute search via Chrome CDP (following the workflow file instructions)
3. Parse results into structured data
4. Deduplicate against existing Excel
5. Export and report

---

Opt-In Sources (arXiv / Google Scholar)

arXiv and Google Scholar are available but never auto-selected. The user must explicitly name the source:

User says	Action
"Search arXiv for climate adaptation preprints"	Search arXiv only
"Use Google Scholar to find urban governance papers"	Search Google Scholar only
"Search WOS and also check arXiv"	WOS primary, arXiv supplementary

When arXiv or Google Scholar is requested:

• Results are marked as unverified (no institutional indexing guarantee)
• No journal-list filtering is applied
• WOS/CNKI deduplication rules still apply (mark duplicates, don't silently drop)
• Report source explicitly: "Source: arXiv (unverified)"

---

Dual-Engine Search

When the user asks to search both WOS and CNKI:

1. Execute WOS search and CNKI search in parallel
2. Deduplicate within each engine independently
3. Report results separately (WOS section + CNKI section)
4. Do NOT cross-deduplicate between engines (different indexing systems)

---

Integration with deep-research

When the bibliography_agent in deep-research needs to search for literature:

1. bibliography_agent formulates search terms and journal lists
2. Invokes literature-search with the terms
3. literature-search routes to WOS or CNKI based on keyword language
4. Returns annotated bibliography to synthesis_agent

---

Safety Rules

1. Never enter credentials — WOS and CNKI login is IP-based
2. Do not bypass rate limits or CAPTCHAs
3. Keep request rates human-like: 1-2 seconds between page loads
4. Do not download PDFs or full text from WOS, CNKI, or linked publisher sites
5. If session expires, stop and report to the user
6. If Chrome is not available or WOS/CNKI tab is not open, stop and report

---

Output Format

Search report
- Engine: [WOS / CNKI / Both]
- Keywords: [...]
- Time range: [...]
- Total found: N
- Duplicates skipped: N
- New papers saved: N
- Output file: [absolute path to Excel]

Top results:
1. [Author, Year] Title. Journal. DOI.
2. ...