WeChat Article Summarize

Read one or more WeChat public account article links from mp.weixin.qq.com, extract cleaned full text and optional image links, summarize each article in Chi...

Security Scan

The skill's code and instructions match its stated purpose (fetch WeChat article pages, extract/clean text, call a summarization CLI, and write structured markdown) with no obvious mismatches or unexplained access requests.

Purpose & Capability

The name/description match the included scripts: reading mp.weixin.qq.com pages, extracting body/title/images, cleaning text, invoking a 'summarize' CLI to produce Chinese summaries, and writing markdown. Required capabilities (HTTP fetch, file I/O, calling an external summarizer) are consistent with the stated purpose.

Instruction Scope

SKILL.md instructions and the orchestrator script (run_wechat_mindmap_workflow.py) limit actions to fetching specified WeChat URLs, repairing HTML, invoking the summarizer, normalizing text, and writing files to a user-chosen directory. The code does not read unrelated system paths or attempt to exfiltrate data to hidden endpoints; it only extracts image URLs (but does not download them).

Install Mechanism

There is no install spec; the skill is instruction+script based and relies on local Python and an external 'summarize' CLI. No network download/install of arbitrary code is embedded in the skill files.

Credentials

The skill declares no required env vars. It supports passing an --env-file to load environment variables into the process (summarize_cn.py implements load_env_file). This is reasonable to supply an API key for the external summarizer, but users should avoid passing env files containing unrelated/privileged secrets because the workflow will import those values into child processes.

Persistence & Privilege

always is false and the skill does not attempt to modify other skills or system-wide agent settings. It writes output files only to user-specified or local working directories.

Assessment

This skill appears coherent and does what it says: it fetches pages from mp.weixin.qq.com, cleans text, calls a local/external 'summarize' CLI to produce Chinese summaries, and writes markdown to a directory you choose. Before installing or running: 1) Confirm you trust the 'summarize' CLI referenced by the scripts (inspect that binary or package — it will receive your article text and any env vars you provide). 2) Do not pass an env-file containing unrelated secrets (AWS keys, tokens) — the skill will load any variables in the provided env-file into the subprocess environment. 3) Expect the skill to make outbound HTTP requests to the WeChat article URLs (it sets a browser-like User-Agent); it extracts image URLs but does not download image files by default. 4) Run the workflow in a directory you control and review the generated files before sharing. If you want extra assurance, run the scripts locally on a test article and inspect the called subprocess (summarize) behavior and environment variables it receives.

把一个或多个微信公众号文章链接整理成结构化 markdown，支持单篇整理和多篇日报汇总。

功能简介

• 读取一个或多个 mp.weixin.qq.com 文章链接
• 抽取文章正文、标题、发布时间，以及可选的图片链接
• 自动修复常见的微信正文乱码问题
• 调用 summarize 用中文总结全文内容
• 生成结构化 markdown 文件
  • 单篇文章整理
  • 多篇文章汇总 / 日报
• 支持按日期 + 标题，或日期 + 篇数 + 汇总说明命名
• 支持把文件保存到用户指定目录

使用前需要确认

在真正开始抓取文章之前，需要先确认：

1. summarize 已经配置好 API key，并且可正常使用
2. 是否需要在最终 markdown 中保留图片链接
3. 最终文件保存到哪个目录

适用场景

• 总结单篇微信文章
• 把多篇微信文章汇总成一份日报
• 输出适合继续阅读、归档或二次整理的 markdown 文件

Workflow

Step 0: Confirm prerequisites before fetching anything

Do not fetch article content until all three items are clear:

1. summarize is ready

   • Ask the user to configure summarize API access first if needed.
   • Verify summarize by running a tiny Chinese test.
   • Proceed only if summarize returns a usable summary.

2. Image preference

   • Ask whether the final markdown should include image links.
   • Map user intent to include_images=true|false.

3. Output directory

   • Ask where to save the final markdown file.
   • If the user says “下载文件夹”, use ~/Downloads.
   • Create the target directory if it does not exist.

If any of the three items is missing, stop and ask before continuing.

Step 1: Extract each WeChat article

For each mp.weixin.qq.com URL, run:

python3 scripts/read_wechat_article.py '<wechat_url>' --out '<temp_dir>'

This produces structured metadata, raw HTML, and a first-pass markdown export.

Step 2: Clean the body text

Do not trust the first-pass article markdown blindly.

If the body contains mojibake or obvious encoding corruption, repair it from raw.html by running:

python3 scripts/fix_wechat_body.py '<raw.html>' --out '<body-fixed.txt>'

Use the cleaned body text as the canonical input for summarization.

Step 3: Summarize in Chinese

Always summarize the cleaned local text, not the original WeChat URL.

Run:

python3 scripts/summarize_cn.py '<body-fixed.txt>' --out '<summary.json>' --length short

or for a combined report:

python3 scripts/summarize_cn.py '<combined-input.md>' --out '<summary.json>' --length medium

The script enforces Chinese output and fails if the returned summary is not sufficiently Chinese.

Step 4: Normalize summary text before writing markdown

Never write summarize output directly into the final file.

Normalize paragraph breaks and spacing with:

python3 scripts/normalize_markdown_text.py '<input.txt>' --out '<normalized.txt>'

Use this for:

• each single-article summary
• the combined daily-report overview

This prevents ugly line wrapping and mixed-language formatting artifacts.

Step 5: Build the final markdown

Single article

Run:

python3 scripts/build_mindmap_markdown.py \
  --result '<result.json>' \
  --body '<body-fixed.txt>' \
  --summary '<summary.json>' \
  --output-dir '<chosen-dir>' \
  --include-images true

Multiple articles / daily report

Run:

python3 scripts/build_batch_report.py \
  --inputs '<dir1>' '<dir2>' '<dir3>' \
  --output-dir '<chosen-dir>' \
  --include-images true \
  --report-label '微信文章日报'

The batch report must:

1. summarize all articles individually
2. summarize the full set as one combined overview
3. place the combined overview first
4. then append each single article section

Output rules

Naming

Single article

YYYYMMDD-文章标题.md

Multiple articles

YYYYMMDD-<总文章数量>篇-<汇总说明>.md

Content rules

Single article output should contain

• title
• source URL
• publish time
• summarize-generated Chinese summary
• mindmap-style structure
• optional image section

Batch report output should contain

• combined daily overview at the top
• combined mindmap
• per-article title, URL, date, summary, and mindmap
• optional image overview

Non-negotiable quality gates

Before writing the final markdown:

1. Summary language check

   • If the summary is not mainly Chinese, retry or fail.

2. Paragraph normalization

   • Collapse unnatural line breaks inside prose.
   • Keep markdown headings and bullet lists intact.

3. Clean body source

   • Prefer repaired text from raw.html when the extracted body is corrupted.

Bundled scripts

• scripts/read_wechat_article.py — fetch WeChat article metadata, body, raw HTML, and image links
• scripts/fix_wechat_body.py — repair mojibake and extract clean text from raw HTML
• scripts/summarize_cn.py — run summarize in Chinese and enforce a language check
• scripts/normalize_markdown_text.py — normalize prose paragraphs and line breaks
• scripts/build_mindmap_markdown.py — generate single-article markdown files
• scripts/build_batch_report.py — generate multi-article combined reports
• scripts/run_wechat_mindmap_workflow.py — orchestrate the full workflow end to end after the required user confirmations