Skip to content
Back to search

paper

SkillWorks with Claude CodeGitHub

Academic paper workflow — find, read/annotate (HTML), daily digest, cite. Use when user invokes /paper, needs paper annotation, literature search, citation generation, or daily paper discovery.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/sjqsgg--paperwise/badge.svg)](/listings/sjqsgg--paperwise)

HTML

<a href="/listings/sjqsgg--paperwise"><img src="/api/listings/sjqsgg--paperwise/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Copy `SKILL.md` into your project's skills directory:

your-project/
└── .agents/
    └── skills/
        └── paper/
            ├── SKILL.md
            └── references/
                └── template.css   ← customizable CSS for all HTML output

Then run `/paper find "your topic"` in Claude Code. That's it.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 377 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: paper
  3. description: Academic paper workflow — find, read/annotate (HTML), daily digest, cite. Use when user invokes /paper, needs paper annotation, literature search, citation generation, or daily paper discovery.
  4. ---
  5. # /paper — Academic Paper Workflow
  6. ## Default Config
  7. Edit this section to customize for your project.
  8. ```yaml
  9. output_dir: "./papers"
  10. # Obsidian vault users: set paper_output_dir in your project's CLAUDE.md
  11. # Drive/OneDrive: use the appropriate MCP, then set path (e.g. "gdrive://My Drive/Papers")
  12. venues: [CHI, ACL, EMNLP, NAACL, NeurIPS, ICML, ICLR, EDM, LAK, AIED, ITS, CSCW, SIGIR, CIKM]
  13. keywords:
  14. - cognitive load
  15. - adaptive learning
  16. - intelligent tutoring
  17. - conversational learning
  18. - LLM tutoring
  19. - learning analytics
  20. - behavioral signals
  21. - educational dialogue
  22. min_citations: 5 # /paper find: filter out papers with fewer citations
  23. daily_max: 10 # /paper digest: max papers per run
  24. annotation_lang: zh # zh = Chinese annotations | en = English; override with --lang
  25. openalex_email: "" # Optional. Add email to join OpenAlex Polite Pool (higher rate limits).
  26. semantic_scholar_api_key: "" # Optional. Free key: semanticscholar.org/product/api
  27. ```
  28. Override any config value in your project's `CLAUDE.md` using keys: `paper_output_dir`, `paper_venues`, `paper_keywords`, `paper_annotation_lang`.
  29. ---
  30. ## Subcommands
  31. | Command | Usage | Description |
  32. |---------|-------|-------------|
  33. | `find` | `/paper find "cognitive load LLM"` | Search papers → Digest HTML with top N cards |
  34. | `read` | `/paper read /path/to/file.pdf` | Annotate a single PDF → full dual-column HTML |
  35. | `digest` | `/paper digest` | Daily new papers from arxiv (used by cron) |
  36. | `cite` | `/paper cite [[note-name]]` | Generate APA + BibTeX citation from existing annotation |
  37. **Flags:**
  38. - `--context "ML课第3周"` — override project context
  39. - `--questions "Q1:... Q2:..."` — switch to Question mode (Mode A)
  40. - `--questions` — interactive question mode (Claude asks you)
  41. - `--output /path/` — override output directory for this run
  42. - `--top N` — return N results (default: 5 for find, daily_max for digest); also accepts bare number after query
  43. - `--lang en` — override annotation language for this run
  44. - `--source arxiv` — search only arxiv (latest preprints, no citation filter)
  45. - `--source venues` — search only papers from config `venues` list (citation-weighted ranking)
  46. - `--local /path/` — read local PDFs from folder (no API calls); or `--local a.pdf b.pdf` for specific files
  47. > *Flags can be written with or without `--`. Claude accepts natural language equivalents (e.g. `top 5`, `source venues`, `local /path/`, `questions "..."`).*
  48. ---
  49. ## Context Resolution (in priority order)
  50. 1. `--context` flag in command
  51. 2. Current directory's `CLAUDE.md` (auto-loaded by Claude Code)
  52. 3. No context → use Mode B directly (zero-config, logic analysis mode)
  53. ---
  54. ## Data Sources
  55. ### Primary: OpenAlex
  56. Free, no API key required, 10 req/s rate limit.
  57. ```
  58. https://api.openalex.org/works?search={query}
  59. &select=title,authorships,publication_year,cited_by_count,primary_location,doi
  60. &per-page=20
  61. [&mailto={openalex_email} ← add if configured]
  62. ```
  63. **After fetching:** immediately extract and keep only — title, first author, year, cited_by_count, venue name (`primary_location.source.display_name`), DOI. Discard all other fields from the response before further processing.
  64. ### Secondary: Semantic Scholar (only if `semantic_scholar_api_key` is set)
  65. ```
  66. https://api.semanticscholar.org/graph/v1/paper/search?query={topic}
  67. &fields=title,authors,year,citationCount,venue,externalIds&limit=20
  68. ```
  69. Header: `x-api-key: {semantic_scholar_api_key}`
  70. ### Tertiary: arxiv (fallback + digest source)
  71. ```
  72. https://export.arxiv.org/api/query?search_query=(cat:cs.CL+OR+cat:cs.HC+OR+cat:cs.AI+OR+cat:cs.LG)
  73. +AND+({keywords})&sortBy=submittedDate&sortOrder=descending&max_results=30
  74. ```
  75. Note: arxiv papers have no citation count. Label as `preprint`.
  76. ### Rate Limit Handling
  77. If a source returns 429: check `Retry-After` header → wait that many seconds. If no header: wait 2s → 5s → 10s (3 retries). After 3 failures → skip source, move to next in priority. Note which source was skipped in output.
  78. ---
  79. ## Subcommand: find
  80. 1. **Parse flags:** topic query, `--top N` (default: 5), `--lang`, `--source` (default: mixed), `--local` (path or file list)
  81. - `--top N` also accepts a bare number immediately after the query string (e.g. `/paper find "query" 10` → top 10)
  82. 2. **Resolve source strategy:**
  83. | `--source` / flag | API calls | Ranking formula |
  84. |-------------------|-----------|----------------|
  85. | *(default, mixed)* | Single OpenAlex call `sort=cited_by_count:desc`; arxiv fallback if <3 results | `0.5 × relevance_rank + 0.3 × log(cited_by_count+1) + 0.2 × recency_score` |
  86. | `arxiv` | arxiv API only (`sortBy=submittedDate`); no `min_citations` filter | `0.5 × relevance_rank + 0.5 × recency_score` |
  87. | `venues` | Single OpenAlex call; keep only papers where venue name matches any entry in config `venues` list (case-insensitive partial match on `primary_location.source.display_name`) | `0.5 × relevance_rank + 0.5 × log(cited_by_count+1)` |
  88. | `--local /path/` or `--local a.pdf b.pdf` | No API calls — read local PDFs only | Take first `--top N` files (alphabetical order) |
  89. **`--local` and `--source` are mutually exclusive.** If both are given, return an error.
  90. 3. **If `--local` flag present — execute local PDF steps instead of steps 3–6:**
  91. 1. Scan the given folder for all `.pdf` files, or use the explicitly listed files directly
  92. 2. If file count > `--top N`, take first N (alphabetical order)
  93. 3. For each PDF, extract: title, authors, year, abstract (first 300 words), conclusion paragraph (last 300 words)
  94. 4. Generate a Digest Card per PDF (same 4-sentence format: 研究问题 / 核心方法 / 关键发现 / 相关性)
  95. 5. Citation badge: fixed as `📄 Local PDF` (no OpenAlex lookup)
  96. 6. Card bottom link: `/paper read /absolute/path/to/file.pdf 精读` (use absolute path)
  97. 7. Save to: `{output_dir}/FindResults/find-local-YYYY-MM-DD-HHmm-{folder-slug}.html`
  98. 8. Print summary and exit — skip steps 4–8 below
  99. 3b. **Fetch papers (online mode):**
  100. - Add `&filter=cited_by_count:>{min_citations}` for default/venues mode (skip for arxiv)
  101. - After fetch: immediately discard all fields except title, first author, year, cited_by_count, venue name, DOI
  102. - On any source failure → follow Rate Limit Handling
  103. 4. **Deduplicate against existing files:**
  104. - Check `{output_dir}/FindResults/` and `{output_dir}/` root for `[FirstAuthor-YYYY-slug].html`
  105. - Mark existing papers as `[已有]`, include in summary with their path, skip card generation
  106. 5. **Rank top N** using the formula for the active source strategy
  107. 6. **Generate Digest HTML** — a single self-contained HTML file containing N paper cards.
  108. **Digest Card format** (one card per paper):
  109. ```
  110. [Title] — [First Author] et al., [Year] — [Venue or arXiv]
  111. [Citation badge] [Source tag]
  112. 研究问题: [1 sentence — what problem does this paper solve?]
  113. 核心方法: [1 sentence — what approach do they use?]
  114. 关键发现: [1 sentence — what is the main result?]
  115. 相关性: [1 sentence — why this matters for your research, based on CLAUDE.md context]
  116. [DOI link if available] · → /paper read <doi_or_path> 精读
  117. ```
  118. - Citation badge uses tier logic from Paper Quality Bar section
  119. - Annotation language for cards follows `annotation_lang` config (or `--lang` flag)
  120. - HTML structure: simple cards layout, embed `references/template.css` in `<style>`
  121. 7. **Save** Digest HTML to: `{output_dir}/FindResults/find-YYYY-MM-DD-HHmm-{query-slug}.html`
  122. - Create subdirectory automatically if it doesn't exist
  123. 8. **Print summary:**
  124. ```
  125. Found N papers via {source}. Digest saved:
  126. → 30_Research/FindResults/find-2026-03-14-1430-cognitive-load-llm.html
  127. Papers included:
  128. - Jin-2025-llm-teachable-agent [⭐ 47 citations · CHI]
  129. - Cai-2025-intrinsic-load [📄 preprint · arXiv]
  130. - Klepsch-2017-two-types-icl [已有 → 30_Research/FindResults/...]
  131. ```
  132. ---
  133. ## Subcommand: read
  134. 1. Read the PDF at provided path
  135. 2. Extract: title, authors, year, venue/journal, abstract, full text
  136. 3. Resolve mode:
  137. - No `--questions` → **Mode B** (logic analysis)
  138. - `--questions "..."` → **Mode A (inline)**. Parse questions using these rules:
  139. - Strip optional `Q1:` / `Q2:` etc. prefixes (they are decorative, not required)
  140. - Split on: comma `,` / semicolon `;` / newline / `Q\d+` pattern boundaries
  141. - Extract up to 6 questions; if more are given, take first 6 and warn the user
  142. - Accept any of these formats:
  143. ```
  144. "Q1: 方法? Q2: 与 baseline 比较?" ← Q-label + colon
  145. "方法?, 与 baseline 比较?" ← comma-separated
  146. "方法?; 与 baseline 比较?" ← semicolon-separated
  147. ```
  148. - `--questions` (no argument) → **Mode A (interactive)**:
  149. Output exactly: `请输入你的阅读问题(每行一个,或逗号分隔,最多6个):`
  150. Wait for user input, then parse using the same rules above, then proceed with Mode A
  151. 4. Resolve annotation language: `--lang` flag > `annotation_lang` config
  152. 5. Generate full dual-column HTML annotation (see HTML Template section)
  153. - Paper Quality Bar: omit (no citation data from local PDF unless found in text)
  154. 6. Save to `{output_dir}/[FirstAuthorLastName-YYYY-keyword].html`
  155. - Write directly to path; do not ls the directory beforehand
  156. 7. Output: file path + brief summary of key arguments found
  157. ---
  158. ## Subcommand: digest
  159. 1. **Check cache:** if `{output_dir}/PaperDigests/YYYY-MM-DD-digest.html` exists today → skip fetch, output cached path
  160. 2. **Fetch arxiv:** same URL as Data Sources section above
  161. 3. **Filter:** keep top `daily_max` most relevant to config keywords (semantic match on title+abstract)
  162. 4. **Deduplicate:** skip arxiv IDs seen in previous 7 days' digests
  163. 5. **Annotate:** for each paper, run Mode B annotation (brief version); annotation language follows config
  164. 6. **Combine:** all annotations → single HTML digest file; each paper has a compact Paper Quality Bar
  165. 7. **Save:** `{output_dir}/PaperDigests/YYYY-MM-DD-digest.html`
  166. 8. **Update daily note:** append to `10_Daily/YYYY-MM-DD.md`:
  167. ```
  168. ## Paper Digest
  169. → [[30_Research/PaperDigests/YYYY-MM-DD-digest|Today's Paper Digest]] (N papers)
  170. ```
  171. ---
  172. ## Subcommand: cite
  173. 1. Parse note name from user input (e.g., `[[Klepsch-2017-annotation]]` or file path)
  174. 2. Find HTML or MD file in `{output_dir}/` or vault root
  175. 3. Extract metadata: title, authors, year, venue, DOI/URL
  176. 4. Output directly (no file saved):
  177. - **APA 7th:** `Author, A., & Author, B. (Year). Title. *Venue*, *vol*(issue), pages. https://doi.org/...`
  178. - **BibTeX:**
  179. ```bibtex
  180. @article{key,
  181. author = {...}, title = {...}, journal = {...}, year = {...}, doi = {...}
  182. }
  183. ```
  184. ---
  185. ## HTML Annotation Template
  186. ### Paper Quality Bar (find/digest only — between navbar and section-nav)
  187. Shows: `{source_icon} {source}` | `{venue}` | `{year}` | `{citation_badge}` | `{venue_type_tag}`
  188. **Citation tiers:**
  189. | Condition | Badge | CSS class |
  190. |-----------|-------|-----------|
  191. | cited_by_count ≥ 100 | 🔥 高引 N citations | `badge-high` |
  192. | 20–99 | ⭐ N citations | `badge-important` |
  193. | 5–19 | ✓ N citations | `badge-valid` |
  194. | < 5 or no data | 📄 Preprint | `badge-preprint` |
  195. **Venue type tag:** `A* 会议` (top venues: CHI/NeurIPS/ACL/EMNLP/ICML/ICLR/SIGIR/CSCW) | `会议/期刊` (other config venues) | `Workshop` | `Preprint`
  196. **Source icon:** 🔍 OpenAlex | 📚 Semantic Scholar | 📄 arXiv
  197. CSS: see `references/template.css` (`.quality-bar`, `.badge-*`, `.qb-*` classes).
  198. ---
  199. ### Mode B: Logic Analysis (Default)
  200. Generate a complete, self-contained HTML file. CSS: embed `references/template.css` as `<style>` in `<head>`. Google Fonts: import Lora + IBM Plex Sans.
  201. **5-Color Highlight System:**
  202. | Color | Class | Represents |
  203. |-------|-------|-----------|
  204. | 🟡 Yellow `#fef08a` | `thesis` | Core thesis / main claim |
  205. | 🔴 Red `#fecaca` | `concept` | Key concepts / terminology |
  206. | 🔵 Blue `#bfdbfe` | `evidence` | Empirical evidence / data |
  207. | 🟢 Green `#bbf7d0` | `concession` | Concessions / counterargument handling |
  208. | 🟣 Purple `#e9d5ff` | `methodology` | Methodology description |
  209. **Document structure:**
  210. ```
  211. 1. TOP NAVBAR
  212. - Paper title
  213. - Context label (or "General Reading" if none)
  214. - Color legend: 5 colored chips with dimension labels
  215. 2. PAPER QUALITY BAR (find/digest only — omit for /paper read)
  216. 3. STICKY SECTION NAV
  217. - Links: Abstract | Introduction | Related Work | Methods | Results | Discussion | Conclusion
  218. - Highlight active section on scroll
  219. 4. DUAL-COLUMN BODY — one <div class="paragraph-group"> per paragraph
  220. Left column — original text (ALWAYS IN ORIGINAL LANGUAGE — never translate):
  221. - Copy verbatim from PDF; annotation_lang ONLY controls the right column
  222. - Highlight key phrases: <mark class="thesis">...</mark> etc.
  223. Right column — annotation cards (language follows annotation_lang):
  224. [Colored left border matching paragraph's dominant highlight]
  225. ① 段落功能 / Paragraph Function:...
  226. ② 逻辑角色 / Logical Role:...
  227. ③ 论证技巧或潜在漏洞 / Rhetorical Technique or Logical Gap:...
  228. 5. BACK-TO-TOP BUTTON
  229. <button id="back-to-top" title="返回顶部">↑</button>
  230. <script>
  231. const btn = document.getElementById('back-to-top');
  232. window.addEventListener('scroll', () => { btn.style.display = window.scrollY > 300 ? 'flex' : 'none'; });
  233. btn.addEventListener('click', () => window.scrollTo({ top: 0, behavior: 'smooth' }));
  234. </script>
  235. 6. BOTTOM — Argument Structure Overview (language follows annotation_lang)
  236. zh labels: 问题 / 论点 / 证据 / 反驳处理 / 结论
  237. en labels: Problem / Argument / Evidence / Concession / Conclusion
  238. - Author's core claim (1 sentence)
  239. - 最强论证 / Strongest argument
  240. - 最弱论证 / Weakest argument
  241. - APA Citation (auto-formatted) + copy button:
  242. <button class="copy-btn" data-label="复制 APA" onclick="copyText(this, '{apa_string}')">复制 APA</button>
  243. - BibTeX block + copy button:
  244. <button class="copy-btn" data-label="复制 BibTeX" onclick="copyText(this, '{bibtex_string}')">复制 BibTeX</button>
  245. Add this script alongside the back-to-top script:
  246. <script>
  247. function copyText(btn, text) {
  248. navigator.clipboard.writeText(text);
  249. btn.textContent = '✓ 已复制'; btn.classList.add('copied');
  250. setTimeout(() => { btn.textContent = btn.dataset.label; btn.classList.remove('copied'); }, 1500);
  251. }
  252. </script>
  253. ```
  254. ---
  255. ### Mode A: Question-Driven (triggered by --questions)
  256. Same structure as Mode B, with these differences:
  257. - Color legend: per-question colors Q1–Q6 (cycling through a distinct palette)
  258. - Left highlights: `<mark class="q1">`, `<mark class="q2">`, etc.
  259. - Right annotation cards: labeled `【Q2 核心论点】` / `[Q2 Core Argument]`:
  260. ① Paragraph Function ② Argument Logic ③ Which question/layer this paragraph answers
  261. - Bottom: **Q&A Worksheet** (replaces Argument Structure Overview):
  262. One card per question — Core argument + Key evidence + Potential counter/limitation
  263. ---
  264. ### CSS
  265. All CSS is in `references/template.css` (adjacent to this SKILL.md). When generating any HTML output, embed the file contents as a `<style>` block in `<head>`. Only load when generating HTML — `/paper cite` does not need it.
  266. ---
  267. ## Error Handling
  268. - **PDF unreadable:** output error with path, suggest checking permissions
  269. - **No results from any source:** output "No papers found — try broader keywords"
  270. - **OpenAlex 429:** retry with backoff (2s→5s→10s); after 3 failures fall back to arxiv
  271. - **Semantic Scholar 429:** retry with backoff; fall back to OpenAlex/arxiv
  272. - **Missing output_dir:** create directory automatically before saving
  273. - **Malformed --questions:** re-prompt user for correct format
  274. - **Source fallback notice:** always state which source(s) were used, e.g. "⚠️ OpenAlex unavailable — results from arXiv only (no citation counts)"
  275. ---
  276. ## Notes for Open-Source Use
  277. 1. Set `output_dir` in your project's `CLAUDE.md`:
  278. ```
  279. paper_output_dir: 30_Research
  280. ```
  281. 2. Override keywords/venues/language:
  282. ```
  283. paper_venues: [CHI, AIED, EDM, LAK]
  284. paper_keywords: [your, topic, keywords]
  285. paper_annotation_lang: en
  286. ```
  287. 3. Optional — OpenAlex Polite Pool (higher rate limits, no account needed):
  288. ```
  289. openalex_email: yourname@example.com
  290. ```
  291. 4. Optional — Semantic Scholar secondary source (free key):
  292. ```
  293. semantic_scholar_api_key: your_key_here
  294. ```
README.md · README · 176 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. [English](README.md) | [中文](README.zh.md)
  2. # Paperwise — Academic Paper Skill for Claude Code
  3. > Turn any paper into a **browser-readable annotated webpage** — dual-column layout, color-coded logic highlights, and per-paragraph annotation cards. Powered by OpenAlex. Zero API keys required.
  4. ![Demo](Demo-cn.png)
  5. ## What it does
  6. | Command | Output |
  7. |---------|--------|
  8. | `/paper find "cognitive load LLM"` | Searches OpenAlex → generates Digest HTML with top N overview cards |
  9. | `/paper read /path/to/file.pdf` | Fully annotates a single local PDF |
  10. | `/paper digest` | Daily digest of new arxiv papers matching your keywords |
  11. | `/paper cite [[Author-YYYY-note]]` | Generates APA 7th + BibTeX from an existing annotation |
  12. Each annotation is a **self-contained HTML file** with:
  13. - **Paper Quality Bar** — source, venue, year, citation tier badge (🔥 high · ⭐ important · ✓ valid · 📄 preprint)
  14. - **5-color logical highlights** — thesis · concept · evidence · concession · methodology
  15. - **Dual-column layout** — original text on the left, annotation cards on the right
  16. - **Argument structure overview** at the bottom — full logical skeleton of the paper
  17. - **One-click copy buttons** for APA and BibTeX citations
  18. ## Installation
  19. Copy `SKILL.md` into your project's skills directory:
  20. ```
  21. your-project/
  22. └── .agents/
  23. └── skills/
  24. └── paper/
  25. ├── SKILL.md
  26. └── references/
  27. └── template.css ← customizable CSS for all HTML output
  28. ```
  29. Then run `/paper find "your topic"` in Claude Code. That's it.
  30. ## Configuration
  31. Edit the `## Default Config` block at the top of `SKILL.md`. All fields are optional.
  32. ```yaml
  33. output_dir: "./papers" # where to save HTML files
  34. venues: [CHI, ACL, NeurIPS, ...] # venues for badge tagging
  35. keywords: [...] # default keywords for /paper digest
  36. min_citations: 5 # filter threshold for /paper find
  37. daily_max: 10 # max papers per /paper digest run
  38. annotation_lang: zh # zh = Chinese | en = English
  39. # override per-command: --lang en
  40. openalex_email: "" # optional: join OpenAlex Polite Pool (higher rate limits)
  41. semantic_scholar_api_key: "" # optional secondary source
  42. ```
  43. **Per-project override** — add to your project's `CLAUDE.md`:
  44. ```
  45. paper_output_dir: 30_Research
  46. paper_annotation_lang: en
  47. ```
  48. ## Usage
  49. ### Find papers
  50. ```
  51. /paper find "retrieval augmented generation"
  52. /paper find "transformer attention" --top 10
  53. /paper find "transformer attention" 10 # bare number = --top 10
  54. /paper find "diffusion models" --source venues # peer-reviewed only (config venues list)
  55. /paper find "LLM agents 2025" --source arxiv # latest preprints only
  56. /paper find "X" --lang en # English annotations
  57. # Local PDF digest — no API calls needed:
  58. /paper find --local ~/Downloads/papers/ # all PDFs in folder
  59. /paper find --local a.pdf b.pdf c.pdf # specific files
  60. /paper find --local ~/Downloads/ --top 3 # limit to 3 files
  61. ```
  62. Output is organized automatically:
  63. ```
  64. papers/
  65. ├── {Author-YYYY-keyword}.html ← /paper read output (flat)
  66. ├── FindResults/
  67. │ ├── find-2026-03-14-1430-cognitive-load-llm.html ← /paper find
  68. │ └── find-local-2026-03-14-1445-downloads.html ← /paper find --local
  69. └── PaperDigests/
  70. └── 2026-03-14-digest.html ← /paper digest
  71. ```
  72. Re-running the same query skips existing files and marks them `[already exists]` in the summary.
  73. ### Read a local PDF
  74. ```
  75. /paper read ~/Downloads/paper.pdf
  76. /paper read paper.pdf --context "Week 3 of NLP course"
  77. /paper read paper.pdf --questions "Q1: What is the method? Q2: How does it compare?"
  78. ```
  79. ### Daily digest
  80. ```
  81. /paper digest
  82. ```
  83. Fetches today's new arxiv papers matching your configured keywords. Saves to `{output_dir}/PaperDigests/YYYY-MM-DD-digest.html`.
  84. ### Generate citation
  85. ```
  86. /paper cite [[Jin-2025-llm-teachable-agent]]
  87. ```
  88. Outputs APA 7th + BibTeX directly. Nothing saved — just copy from the terminal.
  89. ## `--source` flag
  90. | Flag | Strategy | Ranking | Best for |
  91. |------|----------|---------|----------|
  92. | *(default)* | 2× OpenAlex (by citations + by year), merged | `0.5 relevance + 0.3 log(citations) + 0.2 recency` | Balanced: classics + recent |
  93. | `--source venues` | OpenAlex only, only papers matching config `venues` list | `0.5 relevance + 0.5 log(citations)` | Peer-reviewed only |
  94. | `--source arxiv` | arxiv only | `0.5 relevance + 0.5 recency` | Latest preprints |
  95. | `--local /path/` | Local PDFs only — no API calls | Alphabetical, truncated to `--top N` | Offline / existing downloads |
  96. The default mode makes **two** API calls (sorted by citations, then by year) so both highly-cited classics and recent publications appear in results.
  97. > **Flag syntax is flexible** — flags can be written with or without `--`. Claude accepts natural language equivalents: `top 5`, `source venues`, `local /path/`, `questions "..."`.
  98. ## Annotation modes
  99. **Mode B (default):** Logic analysis. Right-column cards break down each paragraph's function, logical role, and rhetorical technique. Works with zero configuration.
  100. **Mode A (`--questions`):** Question-driven annotation. Highlights and cards are organized around your specific research questions.
  101. ```
  102. # Inline questions:
  103. /paper read paper.pdf --questions "Q1: What method? Q2: How does it compare to baselines?"
  104. # Interactive (Claude asks first):
  105. /paper read paper.pdf --questions
  106. # → Claude: "请输入你的阅读问题(每行一个,或逗号分隔,最多6个):"
  107. # Labels are optional — all formats accepted:
  108. /paper read paper.pdf --questions "method, dataset size, limitations"
  109. /paper read paper.pdf --questions "method; dataset size; limitations"
  110. ```
  111. **Mode A output:**
  112. - Colors assigned per-question (Q1 = color 1, Q2 = color 2, …)
  113. - Annotation cards explain which question each paragraph addresses
  114. - Bottom section = Q&A Worksheet (one card per question: core argument + key evidence + limitation)
  115. ## Data sources
  116. | Source | Role | API key? |
  117. |--------|------|----------|
  118. | **OpenAlex** | Primary | None required |
  119. | **arxiv** | Digest + fallback | None required |
  120. | **Semantic Scholar** | Optional secondary | Free key |
  121. ## Requirements
  122. - Claude Code
  123. - Network access (OpenAlex / arxiv APIs) — not needed for `--local` or `read`
  124. - No API keys required for basic use
  125. ## License
  126. MIT © [sjqsgg](https://github.com/sjqsgg)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.