Skip to main content
Her command line was shaped for something that reads flags and parses output, not for a person with a mouse. That happens to be what your assistant does best.

​
No prompts, no screen-scraping

Your agent writes the whole call up front, positional arguments and flags on one line, and sends it. Nothing pauses halfway to wait for a keystroke. A run that fires at four in the morning, with nobody awake to answer a prompt, still gets to the end.
An agent composes the whole postqueen command up front, flags and all, then reads the JSON that comes back
Your agent gets JSON back, behind a single human-readable header line, so it reads a field by name instead of guessing at wording on a screen. Drop the header with tail -n +2 before piping into jq:
Failures are just as easy to detect, so a script or an agent loop can branch on the result without parsing prose.
auth:status is the one exception. It is a human-readable report rather than a JSON command, and it exits 0 whether or not the key is valid. Read its output for the ✅ Credentials are valid line rather than branching on the exit code.
Discovery closes the loop. An agent that has met a channel for the first time can still post to it correctly, because it asks her instead of assuming.

​
👑 One command teaches your agent

If your agent supports the skills registry, a single command hands it her whole playbook:
What lands is her SKILL.md: Nobody has to point at it. Your agent picks the skill up on its own when the task looks like posting work, and arrives already knowing the discovery flow, the flags on posts:create and the media rule it has to respect. You do not paste command syntax into a chat window to get started.
The skill is plain Markdown over a plain CLI, so no model has an advantage. Claude and GPT read the same file as Gemini or an open-source model on your own machine, and they all type the same commands. If a thing can run a shell command, it can run your social media.
Prefer tool calls to shell commands? Connect over MCP and you get the same capabilities as 11 tools, with no local install. Setup: MCP setup.

​
Two hard rules

Her skill file leads with these two, because almost every failed agent run traces back to one of them.
Rule 1: authenticate first. Commands fail without valid credentials. Export your key before anything else, then confirm with postqueen auth:status.
Your key lives in the app at app.postqueen.ai: open Settings > Developers > Public API and click Reveal. Treat it like a password, since it grants full access to your account.
Rule 2: upload media before attaching it. Every value passed to -m, or to an image field in JSON mode, has to be a URL returned by postqueen upload. She rejects raw local paths like photo.jpg and external links like https://example.com/photo.jpg, and providers such as TikTok, Instagram and YouTube only accept a PostQueen-verified URL.
postqueen auth:login runs an OAuth2 device flow against cli-auth.postqueen.ai unless you point it elsewhere with --auth-server or POSTQUEEN_AUTH_SERVER. That host is not serving yet, so on the hosted service the API key is the route to take today, and self-hosters can run the auth server themselves in the meantime.

​
The core workflow

Every agent session follows the same seven beats, whether it is a coding assistant on your laptop or a job that wakes up at four in the morning.
1

Authenticate

Set the key, then verify it before spending a single API call on a real post.
✅ Credentials are valid in that output means the key works and the rest of the loop can run.
2

Discover

List the connected channels and read what each one accepts.
3

Fetch dynamic data

Pull the values that only the platform knows, such as flairs, playlists or company pages.
4

Prepare media

Send each file through upload and keep the returned path. The tail -n +2 drops the header line, which is not JSON.
5

Post

Schedule it. The -s date is required and takes ISO 8601, and -t draft saves the post without scheduling it instead.
6

Analyze

Ask how it landed, per channel or per post. -d is the number of days to look back.
7

Resolve

If postqueen analytics:post returns {"missing": true}, the post published but the platform withheld its ID. List the provider’s recent content and link the right one.
That is the whole loop. An agent that follows it runs your calendar on its own, and every step it takes stays a normal command with a result you can read.

​
Next steps

Agents overview

The three ways in and a setup guide for each agent she works with.

Install the skill

The API key, the CLI, self-hosted setup, and what goes wrong with each fix.

Command reference

Every command and flag, with the exact syntax an agent needs.

MCP setup

Connect an MCP client to her 11 hosted tools in one line.