← All use cases

中文

Operations & admin · Claude Code

Have an agent research, write and upload local news posts to WordPress

Idea + starter prompt, not run by us yet

For: Local publishers and marketers who post news or updates every day

Koka Sexton gives Claude Code terminal access plus logins to WordPress, Airtable and Gmail. Claude researches local stories, fetches the original articles and writes complete HTML posts. Uploading goes through WP-CLI or a custom script, followed by featured images, a check of the HTTP responses and a log entry. By his own benchmarks, a 90-minute manual routine fits into a single exchange.

Steps

  1. Run it on a staging or test WordPress site first, with posts saved as drafts.
  2. Write a verification script before the first output: check that links work, quotes match the source and the post saved correctly.
  3. Have the agent log each post with its source links so you can trace any claim.
  4. Run the agent beside a human for two weeks, as the author suggests, and compare results.
  5. Only switch to publishing live when the drafts hold up.

Starter prompt

Starter prompt. Change the [brackets] first.

Trial first: do one story only, and save it as a draft on my test WordPress site [site], never the live one. Run the checking script on that one draft and show me the result.

Research local news for [town] from the last [24] hours. Pick [3] stories. For each, fetch the source article, write a short original post in HTML with a source link, and save it as a draft. Show me the plan and the sources before you write. Write a small script that checks each post for broken links and for quotes that do not appear in the source, and run it before saving. Log every draft with its source URLs.

Stop and wait for my yes before anything is published live, and before any email. A saved draft is a pause, not permission. I keep my WordPress credentials in my own environment and will not paste them here.

Original post: kokasexton.com · found via agentusecases.ai. Written up in our own words. We haven't run it, so test it on made-up data first.