cnki-reference-writer

name

cnki-reference-writer

description

Use the user's logged-in Chrome CNKI account with PDF download access to verify, replace, and add references for Chinese theses, graduation papers, course designs, and technical reports. Use when Codex must search CNKI, download PDF-only candidate papers, obtain GB/T 7714-2025 citation entries, read full PDFs, screen suitable references, clean old citation numbering and DOI residue, rewrite or add body text so citations match PDF evidence, insert verified citations into DOCX files, rebuild bibliographies, and produce one reference verification checklist.

CNKI Reference Writer

Turn a Chinese paper with missing, insufficient, fake, weak, or unverifiable references into a Word document with CNKI PDF-verified citations.

Required Confirmation

Before controlling Chrome or downloading from CNKI, confirm:

1. Chrome is already logged in to CNKI.
2. The account or institution has PDF download permission.
3. The user can manually handle slider verification, permission prompts, blocked downloads, or payment prompts if they appear.
4. The paper DOCX path, output folder, PDF folder, and desired reference count or range are clear.
5. Source-quality preferences are clear, such as Peking University Core, CSCD, EI, master's or doctoral theses, recent-five-year journals, or ordinary journals.
6. The task is PDF-only: papers without downloadable PDFs are skipped and recorded.

Required Chrome Capability

CNKI work must use the @chrome plugin and the Chrome:control-chrome skill because CNKI login state, institution access, download permission, verification prompts, and download behavior live in the user's Chrome profile.

When this skill needs to search CNKI, open CNKI pages, copy CNKI quote entries, or download PDFs:

1. First use the available Chrome:control-chrome skill instructions.
2. Prefer the Chrome plugin capability named @chrome / plugin://chrome@openai-bundled when tool or plugin routing is available.
3. Use the Chrome skill's Node REPL browser-client.mjs bootstrap exactly as documented there when direct Chrome tools are not exposed.
4. Do not use generic web browsing, a fresh browser profile, unauthenticated HTTP requests, or non-Chrome automation as a replacement for CNKI operations.
5. If the Chrome plugin or Codex Chrome Extension cannot be reached after following Chrome:control-chrome, pause and report the Chrome setup issue instead of continuing with unverifiable sources.

Non-Negotiable Rules

• Use the user's Chrome session for CNKI because login state, institution access, and download permission live there.
• Do not implement a separate Chrome discovery or connection flow before checking and following Chrome:control-chrome; this skill only adds CNKI-specific rules on top of that Chrome capability.
• Final citation evidence must come from PDF full text only.
• Do not use CNKI abstracts, keywords, titles, search snippets, HTML reading pages, reference lists, or AI summaries as final evidence.
• Download PDFs only from the CNKI article detail page visible selector a#pdfDown[name="pdfDown"].
• Do not use HTML reading, CAJ download, original reading, recommendation-zone links, advertisement links, or any other non-PDF entry as final evidence.
• Do not count a PDF as downloaded because a browser event fired, a click succeeded, or the detail page has a PDF button. Count a PDF as downloaded only after a new PDF actually appears in the user's default download folder, its size is stable, and it has been copied or moved into the task PDF folder.
• If the detail page has no visible PDF download entry, only offers non-PDF entries, or the click does not create a new PDF or .crdownload in the default download folder, skip the paper and record the reason.
• Use CNKI's quote button li.btn-quote[title="引用"] to obtain GB/T 7714-2025 citation entries when available.
• Before writing CNKI citation entries into the paper, bibliography, verification checklist, or Word document, remove all DOI content. Delete DOI:, DOI：, https://doi.org/..., http://doi.org/..., and any obvious DOI URL or DOI marker with everything after it when it appears at the end of an entry. Final references must not contain DOI, DOI:, DOI：, or DOI URLs.
• Use CNKI search filters in #ModuleGroupFilter when quality, year, source, document type, or subject constraints are required.
• Do not cite weakly related papers just to reach a target count.
• Do not replace user-specified references without explaining why and getting confirmation, unless the user already authorized replacement.
• Do not insert a citation number into an unchanged sentence unless that sentence already directly matches the PDF evidence. Rewrite or add nearby body text so the citation supports the exact claim being made.
• Pause for login, verification, permission, payment, or download problems. Do not bypass access controls.

Standard Workflow

1. Read the DOCX and extract headings, paragraph order, existing citation numbers, existing bibliography entries, and claims that need support.
2. If references already exist, verify them on CNKI before searching for additions or replacements.
3. Plan CNKI searches from the paper topic, modules, methods, and paragraph-level claims.
4. Search CNKI in Chrome, apply filters, open candidate detail pages, copy citation entries, close citation popups, and download PDFs. Download success must be verified by scanning the user's default download folder, not by browser download events.
5. Save PDFs in a task-specific folder using the actual downloaded file object's path. Use traceable archive names such as 01_Title_FirstAuthor.pdf, but never hand-type or shell-concatenate complex CNKI filenames.
6. Read each downloaded PDF and extract at least one complete full-text evidence sentence for every final cited paper.
7. Screen references for authenticity, PDF availability, readability, relevance, source quality, and non-duplication.
8. Insert body citations into the DOCX, defaulting to at most one citation number per paragraph. Rewrite or add concise body text when needed so every citation maps to a real PDF evidence sentence.
9. Clean old citation numbers, then rebuild the bibliography with only PDF-verified references ordered by first appearance.
10. Rescan the final DOCX to confirm body citation numbers and bibliography numbers match, numbering starts at [1] and is continuous, no old numbers remain, no DOI residue remains, and no obvious duplicate punctuation or replacement residue remains.
11. Produce one 参考文献引用核验清单.md file and a final processing summary. Do not produce separate tracking-table and citation-mapping Markdown files unless the user explicitly asks for them.

Reference Files

Load these files only when their details are needed:

File	Use when
references/workflow.md	Need the full process for no-reference, insufficient-reference, or existing-reference scenarios.
references/cnki-chrome.md	Need Chrome/CNKI selectors, Chrome:control-chrome routing, quote dialog behavior, overlay closure, filesystem-based PDF download checks, file naming, or skip conditions.
references/screening.md	Need screening criteria, quality filters, rejection reasons, or PDF evidence sentence rules.
references/docx-citation.md	Need body citation placement, old-number cleanup, evidence-aligned rewriting, numbering, bibliography rebuild, saving, or formatting rules.
references/templates.md	Need the single 参考文献引用核验清单.md template or final summary format.
references/qa-checklist.md	Need final quality checks before delivery.

Script

Use scripts/validate_reference_checklist.py to validate the final 参考文献引用核验清单.md for citation numbering, DOI residue, required fields, and local PDF paths:

python scripts/validate_reference_checklist.py path/to/参考文献引用核验清单.md

Run the script before delivery. If it reports old numbering residue, missing PDFs, DOI residue, or missing evidence fields, fix the Word document and checklist before final response.

Final Response

When completing a real CNKI reference task, include:

• Modified Word file path.
• 参考文献引用核验清单.md path.
• PDF full-text folder path.
• Counts for downloaded-and-read PDFs, no-PDF papers, click-without-PDF-landed papers, failed downloads, permission/verification/login/payment problems, unreadable PDFs, final cited references, and rejected candidates.
• Short citation-change summary covering first-appearance order, removed citation piles, removed old numbering, removed unverified references, DOI cleanup, and body text adjusted to match verified PDF evidence.