LinkedIn Feed Brief (Playwright MCP)

Goal

Collect up to N LinkedIn feed posts reliably (collection is primary), filter out ads/modules, and output a brief in the user’s format: <Overall Title> then (Category(Keywords + (Title + Author + Posted Date + Summary + Link)*N))*M.

Title guidance:

• Create a witty, feed-wide umbrella title, or pick one hooky theme that would make people stop scrolling on SNS.
• Keep it short (ideally 1 line) and specific (hint at a tension/tradeoff/trend).
• Avoid the word AI in the title (too obvious); use concrete nouns (e.g., agents, automation, governance, workflow) instead.

Workflow

1) Navigate

• Use Playwright MCP to navigate to https://www.linkedin.com/feed/.
• If redirected to login, stop and ask the user to log in first.

2) Validate you are on the feed

• Confirm URL contains /feed/ and the page shows the main nav + feed layout.
• If unsure, take a Playwright snapshot and verify the center column is the infinite scroll feed.

3) Collect posts (primary)

• Run the JS snippet in scripts/extract_feed_posts.browser_run_code.js via Playwright browser_run_code.
• Default: collect N=20 unique posts by data-urn.
• Prefer DOM permalink (/feed/update/...); fallback to URN link if missing.

If collection returns 0 or times out:

• Avoid reading full innerText for large containers.
• Increase waits between scrolls (LinkedIn lazy-loads heavily).
• Click “New posts” if present, then retry.

4) Filter (required)

• Exclude “Promoted” posts.
• Exclude non-post modules like job cards and in-app promotions.
• Exclude Employment / job-change posts and do not count them toward N.
• Apply rules in references/api_reference.md.

5) Normalize links (required)

• Preferred: DOM permalink extracted from the post.
• Fallback: https://www.linkedin.com/feed/update/{URN}/

6) Write the brief (secondary)

• Categorize posts using the fixed taxonomy in references/api_reference.md.
• Choose the output language based on the dominant language of the collected posts:
  • If the majority of posts are written in Korean, write the brief in Korean.
  • Otherwise, write the brief in English.
• Save the final brief to a .txt file (in addition to printing it in chat).
• For each category:
  • Keywords: 3–5 items, plain text only (no emojis inside the keyword list). Use · as the separator.
  • Emoji rule: Put one emoji before the Keywords: label (e.g. 🧠 Keywords:) and do not reuse that emoji across categories within the same brief.
    • Example: 🧠 Keywords: k1 · k2 · k3, 🧰 Keywords: k1 · k2 · k3, 📈 Keywords: k1 · k2 · k3
  • Category naming rule:
    • If the category is ETC, include a best-guess candidate label in parentheses.
  • Post numbering rule:
    • Number posts within each category as 1), 2), 3)... and restart numbering at 1) for each new category.
  • For each post: Title + Author + Posted Date + 1~2 sentence summary + Link
    • Posted Date rule:
      • If the feed only provides relative times (e.g. 18h, 4 days ago), convert it to an absolute local timestamp by comparing against the script execution time, using the system timezone by default (unless the user specifies otherwise).
      • Output Posted Date as absolute when possible (e.g. 2026-01-15 09:42); otherwise fall back to relative (e.g. 4 days ago).
  • Post formatting rules:
    • Write the Title (Author, Posted Date) line, then put the summary on the next line (no extra blank line between title and summary).
    • Put one blank line between posts within the same category.
• Use assets/example_asset.txt as the output shape.

Output format (must match)

<Overall Title>

🧩 Category Name

🧠 Keywords: k1 · k2 · k3 · ...
1) Title (Author, Posted Date)
- 1~2 sentence summary
(link)

2) Title (Author, Posted Date)
- 1~2 sentence summary
(link)
...

Notes / Guardrails

• Only collect what is visible on the page.
• If LinkedIn UI changes, update selectors and the extraction snippet first.