Skip to content

The Sharek agent skill

An AI coding agent can run the CLI without any help from us. It will also get two things wrong on the first attempt, every time: it will forget to authenticate, and it will pass a local file path where the post needs the address an upload returns.

The Sharek skill teaches it both, plus the shape of a working session.

Terminal window
npx skills add sharekhq/sharek-agent

The skill installs under the name sharek. Your agent picks it up when you ask for something Sharek-shaped, without you naming it.

It expects the CLI to be present, so install that too if it is not already:

Terminal window
npm install -g sharek-cli

Authenticate before anything. The skill makes sharek auth:status the first move and stops there if the answer is no, rather than running a command that was always going to fail. Sign in yourself with sharek auth:login, or put SHAREK_API_KEY in the environment for an agent working on a server.

Upload before attaching. Every file has to go through sharek upload first, and the returned path is what a post attaches. Local paths are never uploaded for you: the file does not reach the platform, and the post can publish without it. Left to itself an agent will try -m ./image.png and wonder where the picture went.

The order of a working session. Authenticate, list the channels, read the settings schema for the ones you are posting to, fetch any ids those settings need, upload media, create the post, then check analytics later. Missing the schema step is what produces posts that fail validation on one platform and succeed on another.

Ask in plain language. The agent finds the channels, reads what each platform needs, and runs the commands:

Schedule a post for tomorrow at 9am on my LinkedIn and X channels announcing the new pricing page, with the image at ./launch.png

Behind that: auth:status, integrations:list, integrations:settings for each channel, upload ./launch.png, then one posts:create carrying both channel ids and the uploaded path.

If you want an assistant scheduling posts and you do not need it in a terminal, connect over MCP instead. Same account and much of the same ground, no CLI to install. The two are not identical: the MCP tools have no analytics, and they add image and video generation. The skill is the better fit when the agent is already working in your codebase.