User research direction: configurable in the body (for example "diversity of a given taxon in a given ecosystem"). The "relevance to my own research" section is written against that direction; when it is not configured, use a generic "implications for the work" instead.

Literature Reader

Two output modes, chosen automatically from the user's trigger words or set manually.

Mode selection

The trigger-word table lives in references/output-templates-zh.md. In short: a reading-note request produces a structured HTML file; a reading-report request produces a plain-text three-part report; an unqualified request defaults to HTML and the user can switch it by asking for a reading report.

Mode 1: Detailed HTML report

Output

The HTML file is saved to the folder that holds the PDF. File naming is specified in references/output-templates-zh.md.

Structure

The section tree for the single-paper note — and, when two or more PDFs are supplied, the additional comparison report — is in references/output-templates-zh.md. The headings themselves are written in Chinese.

Cover, in order: basic information (title, authors, year, journal, DOI), research background, scientific question, methods summary (design, techniques, statistics, sample size), key quantitative results, conclusions and novelty, limitations, and finally the relevance-to-your-own-research block.

The relevance block must cover at least three of these five angles: method transfer, citable data, mechanistic insight, filling a stated gap, and framework crossover.

HTML styling

• Card layout, CSS inlined, no external resources
• Soft professional palette, Chinese fonts prioritised
• Key terms bold or highlighted, alternating row shading in tables
• Responsive (desktop and mobile)

Mode 2: Three-part reading report

Output

Plain text, ready to paste into Word or hand to a supervisor.

Paragraph structure

The three-part, two-part and review-paper variants, the required opening phrasings for each paragraph, and the style rules (paragraph count, citation format, fonts, sentence length) are all in references/output-templates-zh.md.

Key constraints:

• Results must be quantified (numbers, percentages, trend words)
• Short sentences separated by full stops
• No subheadings or numbering inside the report
• The second paragraph must carry both a transferable method and an explicit link to the user's own research
• Priority in the final paragraph: transfer > implication > limitation; limitations are optional

Selecting method highlights

Pick two or three, in priority order: direct relevance to the user's own topic; novel experimental design or statistics; data acquisition route; novel quantitative index or calculation; novel angle on the scientific question.

Workflow

1. Confirm the PDF path and decide the mode (HTML or reading report)
2. Extract the PDF content
3. Pull the information into the template for the chosen mode
4. Generate the output (HTML file or plain text)
5. The relevance section must be written against the user's own research direction, or as a generic "implications for the work" when none is configured

Efficiency rules (highest priority)

1. Extract the PDF in one pass: pull the full text or the key results pages (5 pages or fewer) at once; do not work through it page by page over several rounds
2. Do not trace every number back: once you have enough from the PDF, write the report; do not then re-search the PDF to verify each value. A source annotation is enough — no per-number grep
3. Write the DOCX in one pass: write the content first, then write the DOCX once; do not repeatedly clear and rebuild it
4. Keep the whole job to 3–5 tool calls: extract PDF (1) → write DOCX (1) → final check (1, optional)

Notes

• Scientific accuracy first: stay strictly with the PDF text; never add content that is not in the source
• Mark anything that cannot be extracted as not stated in the source, or as insufficient information
• Keep ecological terminology accurate; retain the English term where that is the correct usage
• The relevance section must not be a generic template — cover at least three of the five angles for every paper
• Review papers may focus on field progress and the gaps that remain