Instagram and TikTok posts can’t be taken down through the API, so a wrong post there stays up, and LinkedIn rejects near-duplicate text for weeks, so the corrected version of a bad post may get refused as a repeat. My social-publishing skill is built around those two facts, and nothing gets created until the person has approved the exact text, accounts, time and media in the conversation.
The skill publishes or schedules posts and videos to LinkedIn, Instagram, YouTube, TikTok and Facebook through Zernio, a service that connects to those accounts and handles the posting. The point is to finish a video and get it out on a sensible schedule without opening five apps.
How a post gets out
A run starts with a read-only check of the setup that proves the Zernio key works and lists the connected accounts. Then the skill asks which accounts this post is for, every time, because handles change and a brand account may not match the person. It agrees the mode with them and spells out what each one costs: a public post or a scheduled one is real and uses up that content, while the only fully private end-to-end test is an unlisted or private YouTube upload.
The skill uploads the media, builds one post with per-platform adjustments, validates it as a dry run and shows a preview as a page. That preview holds every platform’s full text, the account, the time and the media exactly as they’ll go out, plus every change it made per platform, like a link moved into the first comment on LinkedIn or a “Part 3 of” footer dropped where the earlier parts haven’t been posted yet. It also says which platforms won’t let a post be taken down later. When the video already has its own page from my video-director skill, the preview goes at the top of that page’s Posts tab instead of a new page.
The fingerprint
The preview prints a short fingerprint of the exact post, a hash of its contents, and the create command refuses to run without that fingerprint. Change one character after the preview and the fingerprints stop matching, so the script refuses until the person sees a fresh preview and approves again. The skill’s instructions forbid recomputing the fingerprint to get past a refusal, which closes the obvious shortcut an eager assistant might take.
Two smaller protections sit next to it: the script saves a request id beside the post file before it sends anything, so re-running the same create after a dropped connection can’t publish twice, and a timeout is treated as an unknown outcome that gets checked before anything is re-sent. A test-only switch makes the script refuse every write and allow only reads and dry-run validation, so testing can’t post anything by accident.
The key never enters the chat
The person creates the Zernio API key themselves, copies it, and tells Claude “copied”. The script reads it from the clipboard, checks it with Zernio, stores it in the Mac’s Keychain, prints only its last four characters and clears the clipboard. Pasting the key into the chat still works, and in that case the skill tells the person once, accurately, that the key now sits in plain text in that session’s log on their Mac and went to Anthropic with the conversation, and that they should replace it if that log could ever be seen.
Series and scheduling
For a series the skill posts once per platform per day and stays inside Zernio’s limits, which cap an account at 25 posts an hour, Instagram at 100 a day and TikTok at 15 videos a day. It also knows that Zernio’s best-time analytics count days from Monday in UTC while its queue slots count from Sunday in the queue’s timezone, so it converts between the two deliberately instead of copying a number across. After creating a post it reports each platform’s status, and for a publish worth confirming it watches the post every 60 seconds until it’s live, partial or failed.
What’s been tested live
The skill is at version 1.2.2, first written on September 13, 2026. Upload, validation, a scheduled post to three platforms, status checks and deleting a post before its publish time have all been verified against the live API. A post actually going live hasn’t been observed through the skill yet, so it treats the first real publish as a test and watches it closely, and the Facebook and TikTok paths follow Zernio’s documentation without having been exercised.