Automated LinkedIn Pipeline

Showing up consistently on LinkedIn is a content problem disguised as a discipline problem. Manual posting is the first thing to slip during a busy sprint, and SaaS schedulers charge a subscription to do something a cron job can do for free — while holding your account credentials on their servers.

The Automated LinkedIn Pipeline is a self-hosted agent that lives entirely in a GitHub repository. On a schedule, a GitHub Actions runner spins up, asks the Gemini API to draft a technical post, publishes it, and records what it shipped. No server, no subscription, and credentials never leave GitHub's encrypted secrets store.

Who it's for: developers and technical founders who want a reliable, hands-off publishing cadence without paying for — or trusting — a third-party scheduler.

The cron scheduler: 0 10 * * 2,4

The whole pipeline is triggered by a single workflow schedule. The five-field expression reads field-by-field as: minute 0, hour 10, any day-of-month, any month, on days-of-week 2,4 — Tuesday and Thursday. GitHub evaluates cron in UTC, so this is a dependable twice-weekly 10:00 UTC run with a manual workflow_dispatch escape hatch for ad-hoc posts.

on:
  schedule:
    - cron: "0 10 * * 2,4"   # Tue & Thu, 10:00 UTC
  workflow_dispatch: {}        # manual run from the Actions tab

Python + Gemini integration

The agent selects the next topic, assembles a structured prompt, and calls Gemini for a draft constrained to a house style and length. The model call is the only network dependency for content generation; everything else is local file work on the runner.

State tracking via published.json

Idempotency is the hard part of any scheduler — a retried or duplicated run must never double-post. The agent treats published.json as an append-only ledger: before publishing it checks the ledger, and after a successful post it appends the entry and commits the file back to the repo. Because the state lives in the repository, history is auditable in git log and there is no external database.

published = json.loads(LEDGER.read_text())
topic = next(t for t in topics if t["id"] not in published)

post = gemini.generate(prompt_for(topic))
publish_to_linkedin(post)               # post first...

published[topic["id"]] = {"date": today}  # ...then record
LEDGER.write_text(json.dumps(published, indent=2))

Secure repository secrets

The Gemini key and LinkedIn credentials are stored as GitHub Actions secrets, injected into the job's environment only at run time and masked in logs. They are never committed, never printed, and never reach a third party — the runner is destroyed when the job ends.

- name: Publish
  env:
    GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
    LINKEDIN_TOKEN: ${{ secrets.LINKEDIN_TOKEN }}
  run: python scripts/publisher.py