LinkedIn Outreach

Use this skill for personal-account LinkedIn workflows that run on the user's own machine/IP. It is intentionally separate from any VPS, rented account, proxy, database, or systemd setup.

Operating Rules

1. Run everything on the user's local machine with their normal internet connection. Do not use proxies, rented accounts, VPS sessions, CAPTCHA solvers, stealth patches, or credential-based auto-login.
2. Keep the browser visible for first login. The user signs in manually in the Chromium/Chrome window opened by the script.
3. Treat discovery and sending as separate phases. Scout first, review/queue leads second, send only queued leads third.
4. Dry-run is the default for connection requests and DMs. Use --execute only when the user explicitly asks to send.
5. Keep daily limits low. Prefer 3-5 connection requests and 3-10 DMs per run unless the user intentionally changes the cap.
6. Stop if LinkedIn shows a checkpoint, CAPTCHA, restricted-account warning, or any unusual security prompt. Ask the user to handle it manually.
7. Log every material event to the local JSON database so the next agent can resume without guessing.
8. If LinkedIn shows CAPTCHA, checkpoint, identity verification, restriction, or unusual-activity text, the script records a 24-hour safety cooldown in ~/.linkedin-outreach/db/safety_state.json. Do not bypass this unless the user has manually resolved LinkedIn in their normal browser and explicitly accepts the risk.

For the detailed safety and database rules, read references/operating_rules.md and references/database_schema.md when changing the workflow or interpreting stored records.

Script