인터페이스는 현지화되어 있으며, Skill 소개와 설명 문서는 선택한 언어로 AI가 자동 번역합니다. 코드·명령어·경로·전문 용어는 원문을 유지하며 원작자 문서가 우선합니다.
Skills 둘러보기
개발 및 엔지니어링 · SKILL.md

ernie-image-monitor

ernie-image-monitor

This skill should be used when the user wants to monitor, search, or analyze social media and web content about ERNIE-Image across Chinese platforms including Zhihu, Xiaohongshu, WeChat Official Accounts, and Baidu Search.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill 문서

GitHub

name: ernie-image-monitor description: "This skill should be used when the user wants to monitor, search, or analyze social media and web content about ERNIE-Image across Chinese platforms including Zhihu, Xiaohongshu, WeChat Official Accounts, and Baidu Search."

ERNIE-Image 舆情监测

Monitor mentions of ERNIE-Image across Zhihu, Xiaohongshu, WeChat Official Accounts, and Baidu Search. Collect structured data including account name, content, sentiment judgment, and source URL.

Prerequisites

  • OpenCLI installed at ~/.local/bin/opencli (v1.7.4+)
  • Chrome browser extension installed for cookie-based platforms (Zhihu, XHS)
  • Run ~/.local/bin/opencli doctor to verify browser bridge connectivity

If OpenCLI is not installed, run:

npm install -g @jackwener/opencli --prefix ~/.local

For Chrome extension: load unpacked from ~/.local/lib/node_modules/@jackwener/opencli/extension/.

Quick Start

Run a full multi-platform scan (with XHS detail fetch):

python3 scripts/monitor.py --limit 20 --save

Run without XHS detail fetch (faster):

python3 scripts/monitor.py --limit 20 --save --no-detail

Run on specific platforms:

python3 scripts/monitor.py --platforms zhihu xhs --limit 20
python3 scripts/monitor.py --platforms weixin baidu --limit 10

Output Fields

Each collected item includes:

FieldDescription
platform知乎 / 小红书 / 微信公众号 / 百度搜索
account发布者账号或公众号名称
title标题
content_snippet内容摘要(前 200 字)
sentiment正面 ✅ / 负面 ❌ / 中性 ➖
url来源链接
fetched_at采集时间

Sentiment Analysis (Semantic — by Claude)

The script collects raw data with sentiment field left empty. Claude must fill in sentiment judgment semantically after receiving the results — do NOT use keyword matching.

For each item, read title + content_snippet and judge:

  • 正面 ✅: Content praises, recommends, expresses satisfaction, or is objectively positive toward ERNIE-Image
  • 负面 ❌: Content criticizes, complains, expresses disappointment, compares unfavorably, or raises concerns
  • 中性 ➖: Factual introduction, news report, tutorial, or neutral discussion without clear positive/negative stance

When presenting results to the user, include the sentiment label for each item and provide a summary count at the top (e.g., "共 N 条 | 正面 X ✅ 负面 Y ❌ 中性 Z ➖").

Published Time & Sort Order

Each item includes a published_at field:

  • Zhihu: from created_time or published_at field in API response
  • 小红书: from published_at field in API response
  • 微信公众号 / 百度搜索: extracted via regex from Google snippet text

Results are sorted by published_at descending (newest first). Items without a date appear at the end.

Feishu Upload (Required Every Run)

Every monitoring run must upload results to Feishu as a NEW document — never overwrite existing documents.

After generating the Markdown report:

  1. Write report to ~/feishu_report.md
  2. Upload using lark-cli (must run from ~ directory with relative path):
    cd ~ && ~/.local/bin/lark-cli docs +create --title "ERNIE-Image 舆情监测报告 <crawl_time>" --markdown @feishu_report.md
    
  3. The document title must include the crawl timestamp (format: YYYY-MM-DD HH:MM)
  4. Share the resulting Feishu document URL with the user

Detailed Investigation

For deeper analysis of individual posts found during monitoring:

知乎问题详情:

python3 scripts/zhihu_detail.py <question-id-or-url> --limit 10

小红书笔记详情 + 评论:

python3 scripts/xhs_detail.py <note-id-or-url> --comments --limit 20

Platform Coverage

PlatformMethodRequires Login
知乎opencli zhihu searchYes (cookie)
小红书opencli xiaohongshu searchYes (cookie)
微信公众号Google site:mp.weixin.qq.comNo
百度搜索Google site:baidu.com 系列No

For full OpenCLI command reference, load references/opencli_commands.md.

Search Keywords Used

  • ERNIE-Image
  • ERNIE-Image-Turbo
  • ERNIE_Image
  • 文心图像
  • 百度文心图像生成

To add more keywords, edit the SEARCH_KEYWORDS list in scripts/monitor.py.

Date Filtering

Only content published on or after 2026-04-14 (ERNIE-Image official release date) is kept. Items with unknown dates are retained for manual review.

This is enforced via DATE_FILTER = date(2026, 4, 14) in monitor.py.

Relevance Filtering

Each item must contain at least one of these keywords in title + body + tags: ernie-image, ernie_image, ernie image, ernie-image-turbo, 文心图像, 百度文心图像

Xiaohongshu Detail Fetching

For each XHS search result, the script calls opencli xiaohongshu note <note-id> to fetch:

  • Full post body (content / desc field, up to 200 chars)
  • Tag list (used for relevance filtering and report display)

Use --no-detail flag to skip this step and speed up execution.

Report Fields

Each collected item in the final report includes:

FieldDescription
platform知乎 / 小红书 / 微信公众号 / 百度搜索
account发布者账号或公众号名称
title标题
content_snippet内容大意(正文前 200 字或摘要)
tags标签(小红书)
published_at发布时间
sentiment正面 ✅ / 负面 ❌ / 中性 ➖(Claude 语义分析)
url来源链接
저장한 Skills