GitHub links from Hacker News posted to X and LinkedIn

GitHub links from Hacker News posted to X and LinkedIn

An n8n workflow that runs every six hours. It reads the Hacker News front page, picks the posts that link to GitHub and skips
any it has handled before. For each new one, an OpenAI model writes a post for X and a longer one for LinkedIn. You get both
drafts on Telegram, and five minutes later the workflow publishes them.

What is inside

• workflow.json: the workflow as n8n exports it, twenty-six nodes. Nineteen of them run and seven are sticky notes. It holds no
  credential.
• README.md: these notes.

What we changed from the original copy

• Extract Meta was a Python Code node that installed two packages with micropip and read the page with BeautifulSoup. In
  n8n's current source, read on 12 November 2026, Python runs in a sandbox where imports work only if the instance allows them,
  so the node would stop on its first import. The copy does the same job in JavaScript, with no packages. It keeps the same
  output fields. We ran its code in Node against the front page's markup that day, outside n8n: it found the four
  GitHub posts among the thirty rows, with their scores, ages and comment counts.
• The Airtable base and table, the Telegram chat and the LinkedIn person belonged to the file's author. They are placeholders
  now: YOUR_BASE_ID, YOUR_TABLE_ID, <YOUR_TELEGRAM_CHAT_ID> and <YOUR_LINKEDIN_PERSON_ID>. Every credential link is gone.
• The Wait node carried a webhook ID from the author's n8n. It is dropped, and n8n makes a new one.
• Nothing else. The prompts, the filters and the other Code nodes are as the file had them.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• An Airtable account with a personal access token, and a base with one table (see below).
• A Telegram bot, and the ID of the chat that should get the drafts.
• An X developer app that can post, and a LinkedIn app that can post for your own profile.

The Airtable table needs these fields: Post (single line text), Title, Url, Tweet, LinkedIn, Date (a created time
field) and the checkboxes TDone and LDone.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Make the Airtable table. Copy the base ID, which starts with app, and the table ID, which starts with tbl, from its address.
3. In Search Item, Create Item, Update X Status and Update L Status, replace YOUR_BASE_ID and YOUR_TABLE_ID, and select
   an Airtable credential.
4. Add an OpenAI credential and select it in Generate Content.
5. In Ping Me, replace <YOUR_TELEGRAM_CHAT_ID> with your chat's ID, and select a Telegram credential. Brackets go.
6. Select an X credential in X. In LinkedIn, select a LinkedIn credential and replace <YOUR_LINKEDIN_PERSON_ID> with your
   person ID. Brackets go here too.
7. Run the workflow once by hand with Test workflow. Open each node's output, then check the Telegram message.
8. For a first test, disable X and LinkedIn so nothing is posted by accident. Enable them again to go live, then press Publish.

How a run goes

1. Schedule Trigger fires every six hours.
2. Crawl HN Home loads https://news.ycombinator.com/ once.
3. Extract Meta reads the page and keeps each post whose link contains github.com. Each item has Post (Hacker News's ID),
   title, url, site, score, author, age, comments and hn_url.
4. Search Item looks each Post up in the Airtable table. Merge puts those records and the fresh posts in one list, and
   Filter Unposted Items drops every post whose ID is already in the table.
5. Visit GH Page opens each remaining project's address, and Convert HTML To Markdown turns the page into text.
6. Generate Content calls gpt-4o-mini with a system message and a prompt that ask for JSON with two fields, twitter and
   linkedin. The X post must stay under 280 characters with the link, and neither post may use emoji.
7. Validate Generate Content checks that both fields exist, and Filter Errored lets only items with no error through.
8. Create Item stores the post ID, title, address and both drafts in Airtable.
9. Ping Me sends both drafts to Telegram, beginning with Hi There, here is your readymade tweet.
10. Wait for 5 mins before posting is a Wait node, version 1.1, with its unit set to minutes. Its amount is not in the file, so it
    takes that version's default, 5.
11. X posts the tweet and LinkedIn posts the longer text. Update X Status ticks TDone and Update L Status ticks LDone.

Change the look

• How often it runs: the interval in Schedule Trigger.
• Which links it takes: the github.com test in Extract Meta. Change it to any other site to follow its links.
• The voice, length and rules of the posts: the system message and the prompt in Generate Content.
• The model: Generate Content.
• How long you have to stop a post: the amount in Wait for 5 mins before posting.
• Only draft, never post: delete X, LinkedIn and the two Update nodes, and read the drafts on Telegram or in the table.

Good to know

• Telegram is a notice, not an approval. Nothing waits for your answer, and the posts go out when the five minutes end. To stop
  one, stop the waiting run under Executions before then.
• A run can hold several projects, since the front page often links more than one GitHub repository. The Wait covers them all, and
  X and LinkedIn then post each one in turn, so expect a short burst of posts.
• X is set to continue when it fails, so a post that X refuses still ticks TDone. LinkedIn has no such setting.
• If a reply ever arrives as text and not as JSON, the fallback in Validate Generate Content cannot repair it, because it tests
  the original text again. With Generate Content's JSON output on, the first test normally passes.
• Extract Meta reads Hacker News's page markup. A redesign of the page can leave it with no items.
• OpenAI's deprecations page, read on 12 November 2026, gives 14 January 2027 as the shutdown of gpt-4o-mini, with gpt-6-luna
  as the replacement. Pick another model in Generate Content before then.
• X and LinkedIn both set rules for posts made by software. Read them for your own account, and check what your X plan allows.
• The prompt tells the model not to imply you own a project, and to write no emoji. Read each draft anyway: it works from a page's
  text and can get a detail wrong.
• Where the data goes: the project pages' text goes to OpenAI, the drafts go to Airtable and Telegram, and the posts go to X and
  LinkedIn. Hacker News and GitHub are loaded straight from n8n.