Publish

One post, one or more channels, now or at a set time. For YouTube uploads use yt-upload; for Naver Blog posts use naver-post; for a week of posts use schedule-week.

Procedure

1. Channels. Call select_channels. With no scope it returns the candidates: show them and let the person choose. With what they named (platform names, handles or account ids, or all), it returns accountIds to pass to publish unchanged, plus skipped (expired or not-yet-live channels, with a sentence to read out and a reconnect link) and limits, the tightest rules across the chosen channels. If it answers workspace_required, show the workspaces it lists, ask which one, and pass workspaceId on this and every later call.
2. Rules. Write to limits from select_channels. When you need one channel's rules or what it can do (media required, chains, comments, edits after publishing), read list_platforms and its features table. Never write a limit from memory.
3. Channel settings. Open references/<channel>.md only for the channels in this post (threads, instagram, facebook, youtube, tiktok, telegram, discord, pinterest, x, bluesky, naver_blog, ghost, wordpress). They say which options.<channel>.* fields exist and which ones you must ask about.
4. Text. If a voice file exists at ~/.uplika/<workspace-slug>/voice.md, read it first; if not, write plainly and offer the voice-profile skill afterwards. A channel can get its own text with options.<channel>.content. A long post on a channel that supports chains goes in threadItems (first item is the root).
5. Media.
   • With a terminal (Claude Code, Codex, OpenClaw, Hermes) and a local file: media_presign with the file name, type and size, PUT the bytes to the returned uploadUrl with curl, then media_complete.
   • Without a terminal (claude.ai, ChatGPT): a public https file goes through media_from_url; a file on the person's device goes through media_upload_link (give them the link, then poll media_upload_status until it is ready). In ChatGPT, files in the conversation go through media_from_file: attached files come through as they are, and images ChatGPT made or files made by code come through by their /mnt/data path. A file id alone is refused.
   • Write altText when you can see the image. Pass aiGenerated when the media was made with AI.
6. Show and confirm. Before anything goes out, show the person the exact text per channel, the media, the channels and handles, and the time (now, or scheduledAt with its time zone). Publish only after they approve. A post they want to finish themselves can be saved with draft: true.
7. Publish. Call publish with accountIds, content, mediaIds, options, and wait: true so the response tells you whether it is really out. For a later time pass scheduledAt (ISO 8601 with offset). TikTok needs get_publish_options first (see references/tiktok.md).
8. Report. Call get_post with the returned id and give the person each channel's status and permalink. A target that failed has a reason: read it out, call get_help with its code, and offer retry_post once the cause is fixed. A wait that times out is not a failure: the post is usually still going out, so check with get_post or list_posts instead of publishing again.

Things to know