LinkedIn Export Skill

Parse LinkedIn GDPR data exports into structured JSON, then search messages, analyze connections, export to Markdown, and ingest into RLAMA for semantic search.

Prerequisites

• Python 3.10+ via uv
• LinkedIn GDPR export ZIP — Request at: LinkedIn → Settings → Data Privacy → Get a copy of your data
• RLAMA + Ollama (optional, for semantic search ingestion)

Quick Start

# 1. Parse the export ZIP (run once)
uv run ~/.claude/skills/linkedin-export/scripts/li_parse.py ~/Downloads/Basic_LinkedInDataExport_*.zip

# 2. Search, analyze, export, or ingest
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --list-partners
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py summary
uv run ~/.claude/skills/linkedin-export/scripts/li_export.py all --output ~/linkedin-archive/
uv run ~/.claude/skills/linkedin-export/scripts/li_ingest.py

All scripts read from ~/.claude/skills/linkedin-export/data/parsed.json. Parse once, query many times.

---

Parse — li_parse.py

Unzip and parse all CSVs from the LinkedIn GDPR export into structured JSON.

uv run ~/.claude/skills/linkedin-export/scripts/li_parse.py <linkedin-export.zip>
uv run ~/.claude/skills/linkedin-export/scripts/li_parse.py <zip> --output /custom/path.json

Output: ~/.claude/skills/linkedin-export/data/parsed.json

Parses 23 CSV types:

Core: messages, connections, profile, positions, education, skills, endorsements, invitations, recommendations, shares, reactions, certifications

Extended: comments (548), projects (3), honors (2), organizations (3), volunteering (1), languages (9), events (12), member_follows (828), job_applications (443, merged from multiple files), recommendations_given (3), inferences (4)

Auto-detects CSV column names (case-insensitive), handles LinkedIn's preamble format (Connections.csv), and merges split files (Job Applications).

---

Search Messages — li_search.py

Search messages by person, keyword, date range, or combination.

# Search by person
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --person "Jane Doe"

# Search by keyword
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --keyword "project proposal"

# Date range
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --after 2025-01-01 --before 2025-06-01

# Combined filters
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --person "Jane" --keyword "meeting" --after 2025-06-01

# Full conversation by ID
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --conversation "CONVERSATION_ID"

# List all conversation partners (sorted by message count)
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --list-partners

# Show context around matches
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --keyword "AI" --context 3

# Full message content + JSON output
uv run ~/.claude/skills/linkedin-export/scripts/li_search.py --keyword "proposal" --full --json

Flags: --person, --keyword, --after, --before, --conversation, --list-partners, --context N, --full, --limit N, --json

---

Network Analysis — li_network.py

Analyze the connection graph — companies, roles, timeline.

# Summary stats
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py summary

# Top companies by connection count
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py companies --top 20

# Connection timeline
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py timeline --by year
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py timeline --by month

# Role/title distribution
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py roles --top 20

# Search connections
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py search "Anthropic"

# Export connections to CSV or JSON
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py export --format csv
uv run ~/.claude/skills/linkedin-export/scripts/li_network.py export --format json

Subcommands: summary, companies, timeline, roles, search, export

---

Export to Markdown — li_export.py

Convert parsed data to clean Markdown files.

# Export messages (one file per conversation)
uv run ~/.claude/skills/linkedin-export/scripts/li_export.py messages --output ~/linkedin-archive/messages/

# Export connections as Markdown table
uv run ~/.claude/skills/linkedin-export/scripts/li_export.py connections --output ~/linkedin-archive/connections.md

# Export everything
uv run ~/.claude/skills/linkedin-export/scripts/li_export.py all --output ~/linkedin-archive/

# Export RLAMA-optimized documents
uv run ~/.claude/skills/linkedin-export/scripts/li_export.py rlama --output ~/linkedin-archive/rlama/

Subcommands: messages, connections, all, rlama

---

RLAMA Ingestion — li_ingest.py

Prepare RLAMA-optimized documents and create a semantic search collection.

# Full pipeline: prepare docs + create RLAMA collection
uv run ~/.claude/skills/linkedin-export/scripts/li_ingest.py

# Prepare docs only (no RLAMA required)
uv run ~/.claude/skills/linkedin-export/scripts/li_ingest.py --prepare-only

# Rebuild existing collection
uv run ~/.claude/skills/linkedin-export/scripts/li_ingest.py --rebuild

Collection: linkedin-tdimino (fixed/600/100 chunking, reranker enabled, 13 docs, 2.14 MB)

Query (default: retrieve-only, Claude synthesizes):

# Retrieve raw chunks — Claude reads and synthesizes (best quality)
python3 ~/.claude/skills/rlama/scripts/rlama_retrieve.py linkedin-tdimino "What projects has Tom built?" -k 10

# Fallback: local LLM answers (only without Claude)
rlama run linkedin-tdimino --query "Who works at Google?"

RLAMA document structure (13 files):

• messages-conversations-{a-f,g-l,m-r,s-z}.md — Conversations grouped alphabetically
• connections-companies.md — Connections by company
• connections-timeline.md — Connections by year
• profile-positions-education.md — Resume data
• endorsements-skills.md — Skills and endorsements
• shares-reactions.md — Posts and activity
• comments-activity.md — 548 comments with dates and links
• projects-honors-volunteering.md — Projects, honors, volunteering, organizations
• metadata-languages-events-follows.md — Languages, events, follows, job applications, recommendations given, inferences
• INDEX.md — Collection metadata and counts

---

Data Format Reference

See references/linkedin-export-format.md for complete CSV column documentation.

Key files in the LinkedIn export ZIP (23 parsed):

CSV	Contents
messages.csv	All messages and InMail
Connections.csv	1st-degree connections (preamble format)
Profile.csv	Profile data
Positions.csv	Work history
Education.csv	Education
Skills.csv	Listed skills
Endorsement_Received_Info.csv	Endorsements received
Invitations.csv	Connection requests
Recommendations_Received.csv	Recommendations received
Shares.csv	Posts and shares
Reactions.csv	Post reactions
Certifications.csv	Certifications
Comments.csv	Comments on posts
Projects.csv	Projects (Bazaar, Dream Daimon, etc.)
Honors.csv	Awards and hackathon wins
Organizations.csv	Clubs and groups
Volunteering.csv	Volunteer roles
Languages.csv	Language proficiencies
Events.csv	LinkedIn events
Member_Follows.csv	People/companies followed
Jobs/Job Applications*.csv	Job applications (split across multiple files)
Recommendations_Given.csv	Recommendations written
Inferences_about_you.csv	LinkedIn's inferences

---

Script Selection Guide

Task	Script	Example
First-time setup	li_parse.py	Parse the ZIP
Find a conversation	li_search.py --person	Search by person name
Find a topic	li_search.py --keyword	Search by keyword
Who do I talk to most?	li_search.py --list-partners	Sorted partner list
Company breakdown	li_network.py companies	Top companies
Network growth	li_network.py timeline	Connections over time
Archive messages	li_export.py messages	Markdown per conversation
Semantic search	li_ingest.py	RLAMA collection