AgentSkillsCN

linkedin-stats

提取并展示LinkedIn帖子的各类数据统计,包括曝光量、互动次数、评论数以及转发次数。通过CDP连接Arc浏览器,实时读取用户的近期动态。

SKILL.md
--- frontmatter
name: linkedin-stats
description: Extract and display LinkedIn post statistics including impressions, reactions, comments, and reposts. Connects to Arc browser via CDP to read recent activity.

LinkedIn Stats Extractor

Extracts statistics from LinkedIn recent activity page and displays them in a table format.

IMPORTANT:

  • This skill uses a bash shell script (linkedin_stats.sh), not a TypeScript file
  • When running this skill, only display the markdown table output from the script
  • Do NOT add any summary, analysis, or commentary about the results
  • The output MUST be a properly formatted markdown table with adequate cell padding
  • Simply run: ./linkedin_stats.sh <username> (or just ./linkedin_stats.sh for default)

Usage

Simply invoke the skill and provide your LinkedIn profile username:

code
/linkedin-stats jovezhong

Or use the default (configured for Jove Zhong):

code
/linkedin-stats

What it does

  1. Connects to Arc browser via CDP on port 9222
  2. Navigates to your LinkedIn recent activity page
  3. Extracts post data including:
    • Post date (exact dates for hours/days, relative format like "1w ago" for weeks/months)
    • First few words of post content
    • Impression count
    • Reaction count (likes/emojis)
    • Comment count
    • Repost count
  4. Displays the most recent 10 posts in a markdown table

Requirements

  • Arc browser must be running with remote debugging enabled on port 9222
  • The LinkedIn page must be accessible (user must be logged in)
  • agent-browser command must be available

Example Output

The output is a properly formatted markdown table with adequate cell padding:

code
| # | Date       | Content                                      | Impressions | Reactions | Comments | Reposts |
|---|------------|----------------------------------------------|-------------|-----------|----------|---------|
| 1 | 2026-01-17 | Some of my recent posts about #Claude...     | 2,018       | 27        | 3        | 0       |
| 2 | 2026-01-16 | Langfuse joins ClickHouse! "You can check... | 6,534       | 58        | 7        | 1       |
| 3 | 2026-01-16 | 🎉 Cresta hit 50k followers on LinkedIn...   | 482         | 5         | 0        | 0       |
| 4 | 1w ago     | You probably saw the recent drama: Claud...  | 42,319      | 35        | 11       | 1       |

Important: Only display the table portion, not the progress messages or completion status.

Technical Details

  • Implementation: Bash shell script (linkedin_stats.sh) that orchestrates the extraction
  • Extraction: Python script (scripts/extract_linkedin_posts.py) processes the snapshot
  • Output format: Markdown table with properly padded cells for visual alignment
  • Uses text-based extraction (not screenshots) to minimize token usage
  • Date handling:
    • Exact dates for recent posts: "9 hours ago" → 2026-01-17, "1 day ago" → 2026-01-16
    • Relative format for older posts: "1 week ago" → "1w ago", "2 months ago" → "2mo ago"
    • This approach avoids incorrect date assumptions for posts older than a week
  • Handles LinkedIn's lazy loading by scrolling through the page multiple times
  • Extracts data from accessibility tree snapshot (agent-browser snapshot -c)
  • Captures reaction counts (likes/emoji reactions) in addition to comments and reposts
  • Content field is left-justified with padding to ensure table alignment

Notes

  • LinkedIn may not load all posts due to lazy loading (posts 6-8 are sometimes skipped)
  • Impression counts update in real-time
  • Works best with the "All activity" → "Posts" filter selected