Title: How the CLI skill works Description: Understand the workflow and limits of the openpost-cli skill. Canonical: https://docs.openpo.st/mcp/skills/openpost-cli Source: [https://docs.openpo.st/mcp/skills/openpost-cli](https://docs.openpo.st/mcp/skills/openpost-cli) # How the CLI skill works The `openpost-cli` skill gives an assistant OpenPost-specific operating rules. The assistant still runs the installed `openpost` CLI, which uses a saved profile or `OPENPOST_TOKEN` to reach your instance. ## Before an action The skill tells the assistant to inspect the active instance, identity, and available workspaces. It uses `--json` for reads so it can parse IDs and status fields instead of scraping terminal tables. When the saved context is missing or ambiguous, the assistant should pass an explicit `--profile`, `--instance`, or `--workspace` value. It should use the stored keyring token or environment token without printing it. ## Choosing the authoring command The skill distinguishes the main CLI paths: - `openpost post` handles short text, text with media, and editable text drafts. - `openpost thread create` reads two or more posts separated by `---` from a Markdown file. - `openpost publication` handles format-first work such as link shares, image posts, carousels, stories, short video, long video, and provider-specific settings. It also directs the assistant to check provider readiness and capabilities before using a new destination or format. ## Changes and approval The skill treats scheduling, publishing, retries, comment moderation, account disconnection, slot generation, and deletion as changes. The assistant should run them only when your request authorizes that effect. It should default to an unpublished draft when you did not ask to schedule or publish. It should not add `--force` after a revision conflict or bypass an interactive confirmation without your authorization. ## Result checks After a change, the assistant should read the saved object and current server state. For Publication work, that includes the Publication, lifecycle events, related job, and each destination Rendition. The skill does not make the assistant infallible. Read the proposed command before a destructive or public action, and inspect the result in OpenPost.