linkedin

name	linkedin
model	sonnet
description	LinkedIn automation and data analysis via playwright-cli. Use when extracting profiles, reading feeds, writing recommendations, or searching people. Triggers on LinkedIn, profile extraction, connections, feed, LinkedIn search, recommendation.
disableModelInvocation	true
requires_standards	["english-only"]

LinkedIn Skill

Browser automation skill for LinkedIn access via playwright-cli, plus offline analysis of GDPR data exports.

When to Use

Use this skill when you need to:

• Extract LinkedIn profile information (contacts, connections)
• Search for people or companies on LinkedIn
• Read LinkedIn feed, posts, or notifications
• Write recommendations for connections
• Analyze LinkedIn GDPR data exports (connections, messages, activity)
• Format extracted data for structured output

Mode Routing

Request Type	Mode	Reference
Browse profiles, feed, search, messages	Browser automation	references/browser-automation.md
Write recommendations	Browser automation	See "Writing Recommendations" section below
Analyze GDPR CSV exports	Offline analysis	references/gdpr-analysis.md

Prerequisites

playwright-cli (token-efficient, headless, persistent profile):

• Profile: ~/.local/share/playwright-cli/profiles/linkedin/ (already authenticated)
• Usage: playwright-cli -s=linkedin open --profile=~/.local/share/playwright-cli/profiles/linkedin <url>
• For visual verification: add --headed flag

Browser Automation (Summary)

Extraction scripts live in scripts/. Each targets a specific page type. WHY: JavaScript extractors return structured JSON, avoiding large page snapshots in context.

Core workflow:

playwright-cli -s=linkedin open --profile=~/.local/share/playwright-cli/profiles/linkedin <url>
sleep 3
playwright-cli -s=linkedin eval "$(cat scripts/extract_profile.js)"

Key scripts: extract_profile.js, extract_feed.js, extract_search_results.js, extract_company.js, extract_connections.js, extract_messages.js, extract_notifications.js

Formatting scripts: format_profile_for_vault.js, format_post_for_vault.js, format_company_for_vault.js -- generate structured markdown with YAML frontmatter.

For full script table, URL patterns, extraction workflows, and formatting details, see references/browser-automation.md.

Navigation Quick Reference

Action	URL Pattern
Profile	linkedin.com/in/{username}/
Experience details	linkedin.com/in/{username}/details/experience/
People search	linkedin.com/search/results/people/?keywords={query}
Company search	linkedin.com/search/results/companies/?keywords={query}
Feed	linkedin.com/feed/
Connections	linkedin.com/mynetwork/invite-connect/connections/
Messages	linkedin.com/messaging/
Notifications	linkedin.com/notifications/
GDPR export	linkedin.com/mypreferences/d/download-my-data
Write recommendation	Profile → More → Recommend (see workflow below)

Wait & Scroll Rules

LinkedIn uses heavy lazy-loading. Always wait after actions. WHY: SPA transitions and IntersectionObserver-based loaders need time to render.

Action	Wait
After navigation	3s
After scroll	2s
After search/filter	3s

Scrolling: Use JS scrollTo() for reliable infinite scroll triggering:

playwright-cli -s=linkedin eval "window.scrollTo(0, document.body.scrollHeight)"

WHY: Ensures scrolling reaches the absolute bottom to trigger LinkedIn's loaders.

Always take a screenshot/snapshot after waiting to verify content loaded.

GDPR Data Export Analysis (Summary)

LinkedIn GDPR exports contain CSV files (Connections, Messages, Invitations, etc.).

Analyze with Python tools in gdpr/:

uv run python gdpr/gdpr_analyzer.py /path/to/linkedin-export/
uv run python gdpr/gdpr_analyzer.py /path/to/export/ --section connections

Key gotchas:

• Export is async (minutes to hours), not instant
• Two-part delivery -- Part 2 has full data, wait for it before analysis
• playwright-cli has no download command -- use --headed mode and click the download link manually
• Store exports in ~/code/second-brain/data/linkedin/exports/

For full CSV listing, analysis options, and download procedures, see references/gdpr-analysis.md.

Content Pipeline Integration

Extracted LinkedIn data (browser-based and GDPR exports) can feed into downstream tools:

• Profile/company data formatted as structured markdown for knowledge bases
• Connection exports analyzed for networking patterns
• Feed posts captured for content curation
• GDPR exports provide bulk historical data for analytics

Writing Recommendations

Standard form elements (combobox selects + textbox) — fully automatable, no contenteditable issues.

Workflow:

1. Navigate to profile, scroll down, click in-page More → Recommend
   • Or via direct URL: linkedin.com/in/{username}/edit/forms/recommendation/write/?profileFormEntryPoint=TopLevel&profileUrn=...
   • If coming from profile page, LinkedIn skips the person search (step 1) and goes straight to the form
2. Step 1 of 2 (if shown): Search for person → select from typeahead → click Continue
3. Step 2 of 2: Fill three fields:
   • select Relationship (e.g. "X reported directly to you", "You worked with X in the same group")
   • select Position at the time (populated from their experience history)
   • fill Add recommendation textbox (max 3,000 chars)
4. Click Send

Tips:

• The sticky toolbar "More" button may be intercepted by the nav bar — scroll down ~400px and use the in-page "More" button instead
• Multiple recommendations for the same person are allowed (e.g. one per role/company)
• Draft the recommendation text first, review with user, then fill and send
• LinkedIn typeahead (person search): first click may not register — if selection doesn't stick, clear the field, retype the name, wait 2s for the dropdown, then click the option
• Profile URLs are not always /in/firstname-lastname — they may include a hash suffix (e.g. /in/name-7806a667). If a guessed URL returns 404, search by name via /search/results/people/?keywords=...

Do NOT

• Automate posting -- LinkedIn's contenteditable editor is unreliable for automation. Draft content, copy to clipboard with pbcopy, let user paste manually. WHY: React contenteditable doesn't respond to programmatic text insertion.
• Download GDPR archives in headless mode -- playwright-cli has no download command, and headless click discards downloads. Use --headed mode instead. WHY: Headless Chromium discards downloads triggered by click events.
• Use simple click on radio buttons/checkboxes -- use JS dispatchEvent with change+input events. WHY: React synthetic events bypass native DOM clicks.
• Skip wait times -- LinkedIn rate-limits and lazy-loads aggressively.
• Return full page content -- extract metadata only, write to file, return path. WHY: Full pages waste context tokens.
• Start GDPR analysis on Part 1 only -- wait for Part 2 with complete data.

DOM Changes (Feb 2026)

LinkedIn's DOM was completely overhauled in Feb 2026:

• CSS classes are fully obfuscated (e.g. _92e46e58) -- never match on class names
• h2 is used instead of h1 for profile names
• Section IDs (#about, #experience, #education, #skills) are gone
• .pvs-entity no longer exists
• aria-hidden spans no longer used for text
• .text-body-medium, .text-body-small, .t-bold, .t-normal classes no longer exist

Extraction strategy: Find sections by h2 text content → walk up to parent <section> → parse innerText. The extract_profile.js script implements this approach with bilingual support (EN/DE headings like "Experience" / "Berufserfahrung", "Skills" / "Kenntnisse").

Limitations

• Rate limiting -- LinkedIn may throttle rapid navigation
• Login required -- user must be authenticated (playwright-cli profile)
• Dynamic content -- requires scrolling to load all sections
• Anti-automation -- avoid rapid consecutive actions

Resources

• references/browser-automation.md -- Full extraction scripts, URL patterns, formatting, gotchas
• references/gdpr-analysis.md -- GDPR export contents, analysis tools, download procedures
• scripts/ -- JavaScript extraction and formatting helpers
• gdpr/ -- Python GDPR analysis tools