Linkedin Message Writer

LinkedIn Message Writer

The agent writes each personalized message; the bundled scripts deterministically lint character limits and export the tool CSV. Research uses the linkedin-profile-post-scraper capability (Apify, or its keyless Playwright cookie degrade).

When to use

• "Write LinkedIn messages" / "personalized connection requests / InMails / DMs for these leads."
• Given LinkedIn URLs + a goal/angle, produce per-lead messages.
• The drafting half of linkedin-outreach; downstream of linkedin-commenter-extractor, linkedin-influencer-discovery, kol-discovery.

How it works (agent steps)

1. Intake leads (LinkedIn URLs), message_type, account_tier (free/premium — needed for connection requests), goal, angle, tone, export tool.
2. Research each lead — run linkedin-profile-post-scraper on each URL for profile + recent posts (Apify if APIFY_API_TOKEN set, else web-automation + cookie, else write from user-supplied context only). Skip post-scrape for very large lists; personalize on headline alone.
3. Write the message for the chosen type, leading with the signal/relevance, within the per-type limit:
   • connection_request: 200 (free) / 300 (premium)
   • inmail: subject 200 + body 1900
   • dm / message_request: 8000
   • post_comment / comment_reply: 1250 (must add value — never "Great post!")
4. Lint the drafts (see below); rewrite — never truncate — anything over limit.
5. (sequence) Order CR -> value-first DM -> social-proof DM -> breakup with timing.
6. Export the tool CSV.

How to run

Lint drafted messages (Python 3 stdlib, no install):

python3 ${SKILL_DIR}/scripts/linkedin_lint.py --input ${WORKSPACE}/messages.json

Message JSON — one object or a list:

{"linkedin_url":"https://www.linkedin.com/in/x","name":"Ada","message_type":"connection_request",
 "account_tier":"free","message":"..."}

Export to a tool-ready CSV:

python3 ${SKILL_DIR}/scripts/export_csv.py --input ${WORKSPACE}/messages.json \
  --tool phantombuster --output ${WORKSPACE}/linkedin_outreach.csv

--tool: generic | dripify | expandi | botdog | phantombuster. The linter exits 1 on any over-limit message (rewrite, don't truncate).

Outputs

• Per-lead messages {linkedin_url, name, company, title, message(s), char_count, personalization_notes}, validated against type limits.
• A tool-ready CSV (per-tool column mapping) to workspace / channel attachment.

Credentials / env

• env.required: none. The writing is done by the host agent; the linter + exporter are stdlib Python. No paid service is applicable to the drafting/lint/export steps themselves.
• env.optional: APIFY_API_TOKEN — only for the delegated research step via linkedin-profile-post-scraper: if set → that skill's cookieless Apify actor; else → its keyless Playwright li_at-cookie degrade, or write from user-supplied context only (no fresh research). The fallback lives in the delegated skill, not here.

Notes & edge cases

• The connection-request limit is the gatekeeper — strictly 200 (free) / 300 (premium); rewrite to fit. Comments must add value, never filler.
• This skill produces the CSV — it does not send. Sending is linkedin-outreach / phantombuster.
• Research depth scales cost (Apify ~credits per profile/post); for huge lists, skip the post scrape and personalize on headline only.