linkedin-browser-automation

LinkedIn Browser Automation

Live LinkedIn networking — alumni-first, relationship-first, intel-only. Uses your authenticated browser session to access alumni filters, connection-degree markings, and mutual-connection signals (data you cannot get from third-party APIs).

For bulk enrichment from a list of profile URLs without browser/session, use the sibling skill linkedin-datasearch (Bright Data API).

Reference Documents

• references/golden-goose-referral-tactics.md — full tactics guide (why referrals, objection handling, timing)
• references/connection-note-strategies.md — creative openers, timing, tracking, coffee-chat asks
• references/50-proven-email-scripts.md — Ramit Sethi email principles applied to LinkedIn

⚠️ Non-Negotiable Rules

🚫 NEVER send connection requests. This skill finds prospects, drafts messages, and logs them. The user sends every message manually on LinkedIn. You are an intel/drafting assistant — not a spam cannon.

🛡️ Browser automation ONLY for LinkedIn fetching. Never web_search or web_fetch LinkedIn URLs — LinkedIn blocks scrapers and flags/bans accounts. Use a browser tool (Playwright MCP, or the user's configured browser profile) with 10+ second delays and max 20-30 profile views per session.

👤 Human in the loop for every send. Draft, log, review — the user clicks "Send" on LinkedIn themselves.

Configuration

This skill reads per-user config from ~/.linkedin-automation.config.yml. Config is loaded at skill activation:

!cat ~/.linkedin-automation.config.yml 2>/dev/null || echo "NOT_CONFIGURED"

First-Run Flow

If the config above reads NOT_CONFIGURED, you must onboard the user before doing any LinkedIn work. Ask conversationally (do NOT dump a form):

1. School — "What school or program do you want to lead with for alumni connections? (e.g., Saddleback College, MIT, General Assembly)"
2. Major / program — "What are you studying or what program did you go through?"
3. Year / stage — "Year in school or career stage? (e.g., rising sophomore, career switcher, new grad)"
4. Target roles — "What roles are you hunting? (e.g., SWE intern, data analyst, PM)"
5. Target companies (optional) — "Any specific companies at the top of your list?"
6. Location — "What region? Used for local-bond connection notes."
7. Tracker backend — "Where do you want prospects logged? Options: (a) Supabase project, (b) local SQLite file, (c) Google Sheet, (d) plain Markdown file. Pick one."
8. Tracker details — based on (7), ask for project ID + env-var name, or file path, or Sheet ID.
9. Browser profile name — "Which browser profile is logged into LinkedIn? (default: default)"

Then write the answers to ~/.linkedin-automation.config.yml using the schema in ${CLAUDE_SKILL_DIR}/config.example.yml. Confirm the file was written. Only proceed with LinkedIn work after config exists.

Secrets

Config values that are secrets (API keys, database passwords) MUST live in env vars — never in the config file. The config file stores only the env-var name to reference (e.g., api_key_env: TRACKER_API_KEY). At runtime, the skill reads the named env var via printenv.

---

Workflows

Find Alumni at a Company

browser action:navigate targetUrl:https://www.linkedin.com/school/{school_slug}/people/?keywords={company} profile:{browser_profile}
# wait 10+ seconds
browser action:snapshot profile:{browser_profile}

{school_slug} = the LinkedIn school URL slug (e.g., saddleback-college, mit). If the user's school isn't found, try the generic LinkedIn search with alumni filters.

Extract Prospect From Profile

Visit the profile, snapshot, and extract:

• Name, title, company
• Location
• About-section hooks (career path, side projects, shared interests)
• Connection degree (2nd = 3-5× higher accept rate)
• Mutual connections

Then generate 5 personalized connection-note variations (see Connection Notes section below) and log to the tracker.

Check for Accepted Connections

browser action:navigate targetUrl:https://www.linkedin.com/mynetwork/invitation-manager/sent/ profile:{browser_profile}
# wait 5 seconds
browser action:snapshot profile:{browser_profile}

Update tracker: flip status from Pending → Connected for each accepted request.

After Each Session

1. Insert/update prospects in the configured tracker backend.
2. Verify writes succeeded by querying the tracker. File logs alone are not sufficient — data must be in the system of record.
3. Log a summary: X prospects found, Y status updates, Z ready for follow-up.
4. Share the tracker link if the user has a URL configured.

---

Connection Notes — The 5-Variation Rule

ALWAYS generate 5 variations per prospect. The user picks the best one. Store all 5 in the tracker so they're not lost.

Critical Rules

1. Note #1 = default template (using the user's school/major/role). Notes 2-5 = variations based on the prospect's specific profile.
2. Pre-fill names/companies — no [name] placeholders. Ready to copy-paste.
3. Casual tone — "I'd love to get to know you" energy. Not "I'm fascinated by your exact role."
4. Not pushy, not fake — don't feign deep interest in their specific job unless genuine.
5. No AI tells — no em-dashes, no "I hope this message finds you well," no corporate-speak. Read it out loud; if it sounds like a bot, rewrite.
6. Under 300 characters — LinkedIn's connection-note limit.

Default Template (always Note #1)

Hi {first_name}, I want to connect — I'm a {year} at {school} studying {major} and I'd love to intern at {company} someday. Would love to hear about your experience if you have a few minutes. Thanks!

Template Categories (rotate across notes 2-5)

1. Alumni Bond (highest-priority when {school} matches)

Hi {first_name}, fellow {school} alum here! I'm a {year} studying {major} hoping to intern at {company}. Would love to hear about your path there if you have time. Thanks!

2. Relatable / Casual (similar age or background)

Going from {their_previous} to {their_current} is crazy — I'm on a similar path at {school}. Would love to connect!

3. Respect the Journey (nontraditional paths — career switchers, bootcamp → FAANG, etc.)

Hi {first_name}, I really admired your journey from {previous_role} to {current_role}. Takes real grit — would love to connect and hear what helped you get there.

4. Hook-Based (specific thing from their profile — project, post, award)

Hi {first_name}, noticed you're also into {hook}. That's awesome — I'm a {major} student at {school} exploring {company}. Would love to connect.

5. Local / Shared Context ({location} match)

Hi {first_name}, fellow {location} local here! Always cool to see {school} people at {company}. Would love to connect.

6. Recruiter / Post Response (they posted about hiring)

Hi {first_name}, saw your post about the {role} opening at {company}. I'm a {major} student at {school} and already applied — would love to connect and learn more.

7. Mutual Connection Leverage

Hi {first_name}, I see we're both connected to {mutual}! I'm a {major} student at {school} interested in {company}. Would love to hear about your experience there.

Second-Opinion Review (optional)

Before inserting notes, you MAY pipe them through a second model (Gemini, another Claude instance, etc.) for tone review. Useful when the user says "make sure these don't sound like AI." Prompt template:

"Review these LinkedIn connection notes for a {year} student. Check for: casual genuine tone, not pushy/fake, no AI jargon, sounds human. Brief feedback only."

---

Priority Order

1. Alumni — Same school as the user (highest signal)
2. 2nd Degree — Mutual connections (3-5× higher accept rate)
3. Location — Same region / metro
4. Role — Similar career path

Tracker Backends

The skill supports multiple backends, chosen in config. All backends must support this minimum schema:

Field	Type	Purpose
name	string (required)	Prospect full name
company	string	Current employer
position	string	Current title
profile_url	string	LinkedIn URL
location	string	City / region
connection_status	enum	To Review / Pending / Sent / Connected / New Message
connection_degree	int	1, 2, or 3
is_alumni	bool	School match with user
profile_hook	string	Specific hook used in notes
connection_notes	jsonb (array of strings)	5 note variations
target_roles	text[]	Role tags
follow_up_date	date (YYYY-MM-DD)	Next touchpoint
follow_up_count	int	Touchpoints sent so far
rating	decimal (0-10, 0.5 step)	User-set priority
tags	text[]	Custom tags
archived	bool	Soft-delete flag
source	string	manual / extension_save / agent / etc.
user_id	uuid	Required (RLS). FK to auth.users.id.
created_at	timestamptz	auto
updated_at	timestamptz	auto

Supabase backend (linkedintracker.app)

Project: <your-tracker-project-id> · Table: public.prospects · RLS: enabled

Prefer the Supabase MCP (plugin:supabase:supabase) for all reads and writes — it uses the authenticated session, so auth.uid() resolves correctly for RLS. Do NOT use the REST API with a hard-coded anon key; that bypasses RLS policy enforcement.

Insert flow:

1. Resolve user_id once per session — query public.user_profiles WHERE email = '<user's email>' and cache the UUID.
2. Build the row with user_id populated. connection_notes must be a JSON array (not a bare string).
3. INSERT INTO public.prospects (...) VALUES (...) RETURNING id; via supabase.execute_sql.
4. Verify: SELECT id, name, connection_status FROM prospects WHERE id = '<returned_id>'; — confirm the row is there.
5. Status update: UPDATE prospects SET connection_status = ..., updated_at = now() WHERE id = ... AND user_id = ...;

Local SQLite backend

Store at the path in config (default ~/.linkedin-automation.db). Use sqlite3 CLI. Generate schema on first use if missing.

Google Sheets backend

Use gspread via Python, or the user's configured Sheets MCP if available. The Sheet must have tabs: Prospects, Job Listings.

Markdown backend

Append to {config.tracker.path} — one Markdown table per Sheet. Useful for single-user setups who don't want a database.

---

Timing Intel (Optional)

If the user has an internship tracker configured, cross-reference fresh postings against the prospect list. When a fresh posting matches a target company:

1. Flag it for immediate application.
2. Search LinkedIn for alumni at that company.
3. Add prospects with target_app_date: "OPEN NOW".
4. Prioritize their outreach.

---

Target Metrics

Metric	Target
Prospects found / session	15-20
Acceptance rate	10-20%
Response rate	5-10%
Calls scheduled / week	2-3

If acceptance drops below 10%, re-read references/connection-note-strategies.md — the notes are probably too templated.

---

Tips

• Never send generic requests — always personalize.
• Max 10-20 people per company per session.
• Profile hooks > generic compliments. "Your post about X" > "your impressive background."
• Recruiter posts are gold — filter LinkedIn search to Posts with role keywords.
• Sound human — if it reads robotic out loud, rewrite.
• Ratings (0-10) help the user prioritize when the list gets long.
• Use the 1-2 punch: connect first, ask for referral only after relationship exists.