Skip to content

Examples

You do not write any of this yourself. You describe what you want and your assistant works it out. These examples show the calls it makes underneath, so that when something comes out wrong you can see which step went wrong.

Every example assumes you have already connected your tool.

You say: “Post ‘Our new pricing is live’ to X tomorrow at 9am.”

Your assistant finds the channel with integrationList, which gives it the id and confirms the account is X. Then it reads the rules with integrationSchema, which comes back with that channel’s character limit and the settings X requires. Then it schedules:

{
"socialPost": [
{
"integrationId": "channel-id",
"isPremium": false,
"date": "2027-09-01T09:00:00.000Z",
"shortLink": false,
"type": "schedule",
"postsAndComments": [
{ "content": "<p>Our new pricing is live</p>", "attachments": [] }
],
"settings": [{ "key": "who_can_reply_post", "value": "everyone" }]
}
]
}

The date is UTC. If you say “9am” your assistant converts from your timezone, so it is worth confirming the time it reports back.

You say: “Announce the release in our Discord.”

Discord will not take a channel name. integrationSchema tells the assistant that this provider has a helper for listing channels, and it calls that helper through triggerTool with the channel id and the helper’s name. The helper returns the server’s channels with their ids, the assistant picks the one you meant, or asks you if it is ambiguous, and passes the id in settings:

"settings": [{ "key": "channel", "value": "channel-id" }]

Reddit, Pinterest and several others work the same way. When your assistant pauses to ask “which board?” or “which subreddit?”, this is the step it is on.

You say: “Put this clip up as a Reel with some trending audio.”

Instagram exposes an audio search helper. Your assistant calls it through triggerTool with an empty query, which returns trending tracks rather than search results, picks one, and attaches it:

"settings": [
{ "key": "post_type", "value": "post" },
{ "key": "audio", "value": { "id": "audio-track-id" } }
]

There is no reel type: post_type takes post or story, and a single video posted as post goes out as a Reel. Audio applies only to Reels, never to stories.

Ask for a specific song instead and the same helper searches for it by name.

You say: “Post about the launch on LinkedIn, and make an image for it.”

The assistant calls generateImageTool with a prompt drawn from your post. That returns media already stored in your library:

{ "id": "media-id", "path": "https://media.example.com/generated.png" }

It then puts that path in the post’s attachments:

"postsAndComments": [
{
"content": "<p>The launch is here</p>",
"attachments": ["https://media.example.com/generated.png"]
}
]

The image lands in your media library too, so you can reuse it. If your image credits are exhausted the tool returns an error and your assistant should tell you, rather than posting without the image.

You say: “Post this image” and paste a URL from somewhere else.

That URL cannot go straight on a post. Attachments have to be hosted by Sharek, so the assistant calls uploadFromUrlTool first, gets back the stored copy, and attaches that instead. The URL you gave has to be publicly reachable, or the upload fails.

You say: “Make it a three part thread on X.”

Threads and comments are the same mechanism: postsAndComments is a list where the first item is the post and everything after it follows on.

"postsAndComments": [
{ "content": "<p>Part one</p>", "attachments": [] },
{ "content": "<p>Part two</p>", "attachments": [] },
{ "content": "<p>Part three</p>", "attachments": [] }
]

On LinkedIn the same shape gives you a post with a first comment underneath. What the platform does with the extra items is the only thing that changes.

You say: “Post this to X, LinkedIn and Threads at 9am.”

One call, three entries in socialPost, one per channel, each with its own id and its own settings. Your assistant reads the schema for each platform first, because what LinkedIn requires and what X requires are not the same.

The same shape handles bulk scheduling across dates: twenty posts to one channel across twenty days is twenty entries in a single call, not twenty separate requests.

You say: “Draft these, do not publish yet.”

Same call with type set to draft. The posts appear in your calendar unqueued, and you publish them from there when you are ready. Use schedule to queue for a date, and now to publish immediately.

Drafts skip the platform checks: a draft with a missing setting or text over the length limit is still saved, and only a draft with no text and no image is refused. Fill in what is missing before you schedule it.