Trigger Words
- Analyze my WeRead
- Reading persona / reading profile
- My reading report / reading retrospective
- Guess my MBTI based on my reading
- Reading interest analysis
Prerequisites
weread-skillsInstalled- Environment variable
WEREAD_API_KEYConfigured
Workflow
Step 1: Data Collection
Execute scripts/fetch.py, automatically fetch:
/user/notebooksPaginated full data- For each book's
/book/bookmarklist+/review/list/mine /readdata/detailmode=annually (current year + previous year, covering the last 12 months)/shelf/sync(full bookshelf overview)
python3 ~/.claude/skills/read-weread-analyzer/scripts/fetch.py /tmp/read-weread-analyzer/data.json
Data range note:Prioritize data from the last 12 months. If the user's first valid reading to now is less than 12 months, use all available data and note the range on the report cover.
Step 2: In-Depth Analysis (all 10 dimensions must be completed)
Produce one by one,skipping any item is prohibited. The output requirements for each dimension are as follows:
1. Reading Persona
- Main topics of interest (3-5, ranked by weight)
- Reading preferences (depth/breadth/tool-oriented/reflective...)
- Which books were "truly absorbed" (high completion rate + many highlights + has thoughts)
- Which books were "collected but unread" or "barely sampled" (on the bookshelf but low completion rate / no highlights)
- MBTI inference: Give an inference based on reading subject preferences, note content, and reflective density, mark the confidence level (low/medium/high), and provide 2-3 supporting reasons
2. Interest Drift
- Segment by "earliest 1/3, middle 1/3, most recent 1/3" or slice by quarter
- List the topics of interest for each segment
- Long-term stable interests vs short-term fluctuations
- Summarize in one sentence:what issue do you truly care about recently
3. Knowledge Structure & Blind Spots Map & Blind Spots
- Areas with considerable accumulation (3-5)
- Clearly weak areas (3-5)
- How weaknesses may affect: work judgment / expression ability / product thinking / personal growth (all four perspectives must be given)
4. Reading Quality Diagnosis
- Comprehensively judge the reading mode: systematic learning / interest-based browsing / problem-driven / anxiety-driven collecting
- Key metrics: average completion rate, average reading duration, highlight density (highlights/hour or highlights/book), rereading situation
- Give an honest diagnosis, do not gloss over
5. Core Insights
- Extract 5-8 core viewpoints from all highlights and thoughts that best represent long-term concerns and thinking habits
- Must paraphrase, do not quote at length; if quoting is necessary, each quote must not exceed 20 characters
- Each item links to 1-2 source books
6. Book Network
- Identify among the books you've read: common questions, conflicting viewpoints, and mutual complements
- Form 2-3 "theme clusters"
- Summarize the currenthidden worldview(most important)
7. Actionable Advice
Divide into four categories, with 2-3 specific suggestions per category:
- Work decisions
- Personal growth
- Expression and writing
- Product thinking
8. 10 Topic Ideas
Based on reading notes and highlights, distill 10topics suitable for writing into articles or Xiaohongshu notes. Each includes:
- Title direction (one sentence)
- Core viewpoint (one sentence)
- Suitable books or sources of viewpoints to cite (1-2 books)
9. Knowledge Gaps
Based on the existing reading structure and real needs (the user is an AI productivity + content creator, refer to CLAUDE.md), divide into three tiers:
- Knowledge that must be supplemented(2-3 items)
- Knowledge worth supplementing(3-5 items)
- Knowledge not needed for now(2-3 items, give reasons)
10. Filling the Gaps
For the above knowledge gaps, give specific recommendations, categorized:
- Books (5-8)
- Classic papers or research reports (2-4)
- High-quality courses (2-3)
- Data sources or databases (2-3)
- Authors / institutions / columns suitable for long-term following (5-8)
Step 3: Ask for the save location (must ask)
Before generating the HTML, first ask the user:
报告要保存到哪个目录?
- 直接回车 = 当前工作目录
- 或者输入一个绝对路径 / 相对路径(如 ~/Documents/阅读分析)
If the path does not exist, automatically mkdir -p create it. The filename is fixed as 阅读分析-YYYY-MM-DD.html。
Step 4: Generate the HTML visualization report
Call the template templates/report.html copy it into a new file, thenReplace the 10 <!-- PAGE_N_CONTENT --> marker blocks one by onewith your analysis content, and save to the directory specified in step 3.
After generation, open it with the system default browser (macOS:open <path>;Linux:xdg-open <path>;Windows:start <path>)。
Step 5: Report the results
Output:
- HTML file path
- Data range (e.g., "based on the most recent 12 months, 87 books / 436 highlights / 48 thoughts")
- One-sentence core conclusion (MBTI + dominant reading pattern)
Visual specifications
Strictly follow the design of templates/report.html:
- Each page is fixed at 750×1334px (9:16, corresponding to an iPhone screen)
- Pure white background (
#fafaf7) - Single warm-toned accent color (
#c9a96ebeige /#1a1a1amain text) - Lots of white space
- Large page numbers (48px semi-transparent), title 56px Bold, body text 22px
- Structure of each page: page number → large Chinese title → small English subtitle → content area → footer
- The content area must not exceed the page boundaries; long content should be streamlined or split
- External fonts or JS libraries are not allowed
- The entire HTML is self-contained
Content writing specifications
- Text should be clean and direct,Do not use —— (em dash), use periods/colons/parentheses instead
- Do not use the parallel sentence pattern "not X, but Y", state the conclusion directly
- Paraphrasing takes priority over quoting
- Keep the word count of each page's content between 200-400 characters (title + body); if too long, split or condense it
- Highlight data with bold or a larger font size
- Use
•or numbering for lists, avoid piling them up
Notes
- Do not omit dimensions: The analysis of 10 dimensions must be completed, none can be missing
- Data range: On page 1, clearly indicate "based on the most recent 12 months" or "based on all available data"
- MBTI must be given: Infer based on real data, indicate the confidence level
- Honest diagnosis: The 4th dimension is allowed to give judgments like "anxious hoarding", without sugarcoating
- HTML self-contained: All CSS is inlined in the
<style>tag, no external resources are referenced - Screenshot-friendly: Each page is an independent 9:16 block, and users can take screenshots as Xiaohongshu material