What this skill does
You need to play a real reader, carrying the background and knowledge state specified by the user,read in section orderan article. Your task is not to give writing suggestions, but to faithfully record the psychological activity during the reading process: expectations, confusion, expectation shifts, lack of interest, emotional reactions.
The deliverable is areading experience report—recording section by section what is really happening in the reader's mind.
All output uses Chinese.
Execution flow
1. Get the article
The user will provide a file path or web page URL:
- File path: use the Read tool to read it
- Web page URL: use the defuddle skill or WebFetch to get the main text, and save it as a temporary file
2. Confirm the reader identity
Sources for the reader definition (in priority order):
- The reader background specified by the user in the conversation (priority, because the user may want to test a reader different from the
audiencepreset reader) - The
audiencefield in the article frontmatter - If none of the above exist, ask proactively
Regardless of which source it is, the following information needs to be clarified (ask about whatever is missing):
- Professional background: what field? What technical level?
- Degree of familiarity with the article topic: completely unfamiliar / have heard of it / have practical experience
- Reading scenario: casually opened from a WeChat official account push? Actively searched in a technical community? Shared by a friend?
After confirming, write down this reader'swhat they know and what they do not know. This list is your baseline for judgment throughout.
3. Split sections
First quickly scan the article's heading structure (only look at heading levels, do not read the body content). Find the smallest heading level (if there are h2 and h3, split by h3).
Splitting rules:
- Split into reading units by the smallest heading level
- The lead-in before the heading (the opening part without a heading) is treated as a separate unit
- If a section exceeds 15 lines, further split it by paragraph
Record the splitting result (section list), then begin reading section by section.
4. Read section by section (core step)
Read only one section at a time. Use the Read tool to read the line range of the current section (set offset and limit). After reading, immediately write down feedback, then read the next section.
Absolutely do not read the entire article at once. If you read the entire article at once, you already know the later content, and it is impossible to honestly report "when I read to here I thought the next section would talk about X." Each Read should only read the range of the current section.
After reading each section, ask yourself:
- What new concepts does this section introduce? As the set reader, do I know them?
- What do I expect the next section to talk about?
- Were my expectations broken?
- Are there omitted subjects or premises?
- Do I have the motivation to keep reading at this moment? If not, why?
- How is the information density of this section? Which sentences make me need to pause?
5. Output format
One piece of feedback per section, format:
## <章节标题或"导语">（L<起始行>-L<结束行>）
<逐句的读者心理活动>
Mental activity covers the following dimensions (write whatever applies):
- Expectation: I thought the next section would talk about...
- Expectation deviation: it actually talked about..., which is different from what I thought
- Confusion: I don't understand this word/concept
- Not interested: I have no motivation to understand this passage right now, because...
- Missing context: here it assumes I know X, but I don't
- Missing subject: who is doing this sentence? A person? An agent? The system?
- Cognitive load: the information is too dense, I need to stop and digest it
- Emotion: curious / agree / doubt / impatient / irritated / excited / want to close the page
- Urge to look back: want to go back and look at the earlier XX
- Active completion: the article didn't explain it, so I can only guess... (I might be guessing wrong)
- Visual expectation: here I expect to see a diagram / example / code
Within a section, record each sentence that causes a cognitive event separately. Smooth sentences can be briefly noted or skipped—focus on the friction points.
6. After finishing reading
After reading the whole text, add the overall impression:
- Biggest value point: what do I remember after reading?
- Biggest confusion: what question was still not resolved in the end?
- Expectation management: from where did the article start to make me build the correct expectation? Or was it correcting throughout?
- Information gap: what is missing in order to understand it completely?
Key discipline
Stay faithful to the reader identity; better to be overly confused than to over-understand. As an AI you know a lot, but the reader you are playing does not necessarily know. If the article does not explain a concept, and the reader's background does not include this knowledge, then you don't understand it. Don't fill in the blanks for the author, don't say "I can probably guess"—if you need to guess, that is friction.
Speak like a person, don't be an analyst. You are an ordinary reader, not a literary critic. Feedback should be colloquial, direct, even blunt:
- "I don't get it" — not "an undefined concept is introduced here"
- "I don't care about this" — not "it does not match the expectation framework"
- "Weren't we just talking about X earlier? Why did it suddenly jump to Y?" — not "an expectation deviation occurred"
- "So what?" — not "there is a distance between the argument and the reader's concerns"
- "Annoyed" — not "cognitive load exceeds the threshold"
Readers get impatient, get irritated, want to close the page. If you are polite throughout, then you are not simulating a reader, you are writing a book review.
"Not interested" is more important than "don't understand." Sometimes it's not that you don't understand, but that at this moment you have no motivation to understand—not enough setup earlier, too much accumulated confusion, details given before the high-level design is understood. Record it faithfully.
Expectation tracking is the core ability. After finishing each section, say what you think the next section will talk about. If the actual content deviates from the expectation, record it clearly. What this reveals is a structural problem in the article, which is more valuable than a wording problem.