ai-note-composer

name	ai-note-composer
description	Use this skill whenever the user wants to organize their reading notes, excerpts, and personal insights into a polished short essay (200–500 words). Triggers include: 'help me organize my notes', 'turn my excerpts into an essay', 'polish my reading notes', 'summarize what I read into a short piece', or any request to transform raw note fragments (摘抄、见解、碎片想法) into fluent, human-like writing. Supports four output styles: reflective essay (思考随笔), analytical note (分析笔记), learning diary (学习日记), and critical thinking (批判思考). Do NOT use for summarizing documents directly, writing academic papers, or generating content without user-provided source material.
license	MIT

AI Note Composer — 笔记整理助手

Overview

This skill transforms three raw inputs — article topic, original excerpts (摘抄), and personal insights (见解) — into a fluent, human-voice short essay between 200–500 characters. The output prioritizes the user's own perspective; excerpts serve as supporting evidence, not the main body.

Quick Reference

Input	Required	Notes
Topic / Title	Optional	Provides context for the title and framing
Excerpts (摘抄)	At least one of these two	Raw copy-paste from the source, unedited
Insights (见解)	At least one of these two	Fragments, keywords, half-formed thoughts — all accepted

Style Option	Chinese	Best For
reflective	思考随笔	Personal, layered, natural tone — default
analytical	分析笔记	Logical structure, evidence-based, calm authority
diary	学习日记	Conversational, honest, captures confusion and excitement
critical	批判思考	Opinionated, questioning, takes a clear stance

Length Option	Target Range
short	150–230 字
medium	300–400 字 (default)
long	450–550 字

---

How It Works

Step 1 — Collect Inputs

The user fills three fields (all unstructured input is accepted):

Topic:    关于 RAG 技术的最新进展
Excerpts: [raw copy-paste from article, multiple paragraphs OK]
Insights: [fragments like "感觉 RAG 和 fine-tuning 有点像但又不同 / 工程成本高？"]

Step 2 — API Call

The tool calls claude-sonnet-4-20250514 with a structured system prompt:

const response = await fetch("https://api.anthropic.com/v1/messages", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({
    model: "claude-sonnet-4-20250514",
    max_tokens: 1000,
    system: SYSTEM_PROMPT,   // see System Prompt section below
    messages: [{ role: "user", content: buildPrompt(topic, excerpts, insights) }]
  })
});

Step 3 — Parse Output

The model returns:

• Line 1: Title (≤10 Chinese characters, no brackets or quotes)
• Blank line
• Body: Flowing essay in the requested style and length

const text = data.content.map(i => i.text || "").join("");
const lines = text.trim().split("\n");
const title = lines[0].replace(/^[#\s]+/, "").trim();
const body  = lines.slice(1).join("\n").replace(/^\n+/, "");

---

System Prompt

你是一个帮助学习者整理笔记的写作助手。你的任务是把用户的原文摘抄和碎片化见解，
整理成一篇流畅、自然、有个人风格的中文短文。

要求：
1. [STYLE_INSTRUCTION]            ← injected per style selection
2. 字数控制在 [LENGTH_TARGET]     ← injected per length selection
3. 文章要有标题（10字以内，不要用书名号或引号）
4. 内容以用户的见解为主，摘抄内容作为支撑，而不是简单复述
5. 不要用"首先、其次、总之"等八股套路词
6. 语言要像真人写的，不要有 AI 写作的痕迹
7. 输出格式：第一行是标题，空一行，然后是正文。不需要任何其他标注。

Style Instructions (injected at [STYLE_INSTRUCTION])

Style	Injected Instruction
reflective	以第一人称写一篇有个人温度的思考随笔，语气自然、有层次感，像真实的学习者在记录自己的思路，避免学术腔和AI写作套路
analytical	以第一人称写一篇结构清晰的分析笔记，逻辑层次分明，有观点有论据，语气平稳而有力
diary	以第一人称写一篇轻松真实的学习日记，口语化、有亲切感，像真人在记录今天学到了什么，有困惑、有兴奋、有思考
critical	以第一人称写一篇批判性思考笔记，对摘抄内容提出自己的判断、质疑或认同，语气有张力、有立场

Length Targets (injected at [LENGTH_TARGET])

Option	Injected Value
short	约200字（150–230字之间）
medium	约350字（300–400字之间）
long	约500字（450–550字之间）

---

User Prompt Builder

function buildPrompt(topic, excerpts, insights) {
  const parts = [];
  if (topic)    parts.push(`文章/主题：${topic}`);
  if (excerpts) parts.push(`原文摘抄：\n${excerpts}`);
  if (insights) parts.push(`我的见解（碎片想法）：\n${insights}`);
  return parts.join("\n\n");
}

---

Validation

Input Validation

if (!excerpts.trim() && !insights.trim()) {
  showError("请至少填写摘抄或见解中的一项");
  return;
}

Output Validation

• Title must exist and be ≤ 30 characters (fallback: use topic or "学习笔记")
• Body character count displayed after stripping whitespace: body.replace(/\s/g, "").length
• If data.error is present, surface the message to the user

---

UI Artifact

The full working artifact is saved at:

/mnt/user-data/outputs/ai_note_composer.html

It is a self-contained single-file HTML app (no build step, no dependencies beyond Google Fonts and the Anthropic API). Key UI states:

State	Behavior
Default	All fields empty, button enabled
Loading	Button disabled, spinner shown, output section fades in
Success	Output card appears with title, body, char count, style label
Error	Red error message below button; output unchanged

User actions available after generation:

• 复制全文 — copies title + \n\n + body to clipboard, button text changes to "已复制" for 1.5s
• 重新生成 — re-runs the API call with the same inputs and current style/length settings

---

Design Decisions

• Insights-first principle: The system prompt instructs the model to treat user insights as the main voice and excerpts as supporting material. This prevents the output from becoming a mere paraphrase of the source.
• No structured output: The model returns plain text (title + body) rather than JSON. This produces more natural prose than JSON-constrained generation.
• Fragment-friendly input: Users are explicitly told they do not need to tidy their inputs. Keywords, half-sentences, and rhetorical questions in the insights field all work.
• Style as strategy, not tone: Each style produces structurally different essays (e.g., critical includes explicit disagreement or affirmation; diary includes emotional markers), not just tonal adjustments.

---

Extension Ideas

Feature	Implementation Note
Export to Markdown	Add a download button: blob = new Blob(["# " + title + "\n\n" + body], {type: "text/markdown"})
Tag / topic index	Store generated essays in localStorage with tags for later retrieval
Weekly review mode	Feed 5–7 saved essays as context and ask the model to find connecting themes
Multi-language output	Add a language selector; inject "请用英文输出" or target language into the system prompt