Post to social media from OpenClaw

An agent on your own machine, reached from chat

OpenClaw is installed on your own computer or server. A single Gateway process connects it to chat apps (its docs list Discord, iMessage, Signal, Slack, Telegram, WhatsApp and more), so you message your agent from your phone.[6]

That changes what connecting means. There is no vendor settings page to open: you install a skill, or save a server definition, on your own Gateway. Postnavo is launching with both a skill and a server for that.

Skills are directories, and OpenClaw says to read them

OpenClaw calls skills “markdown instruction files that teach the agent how and when to use tools”. Each lives in a directory with a SKILL.md file, and OpenClaw follows the AgentSkills spec.[1]

openclaw skills install puts a skill into the active workspace’s skills/ directory. It takes entries from ClawHub, the public registry, as well as Git repositories and local directories. Workspace skills win over shared and bundled ones of the same name.[1]

The docs are blunt about trust: “Treat third-party skills as untrusted code. Read them before enabling.”[1]

That goes for Postnavo’s skill too. It is launching as a small instruction package, short enough to read before you enable it.

The MCP route on a Gateway

Server definitions live under mcp.servers in OpenClaw’s config. You add one in the Control UI under Settings, MCP, where you pick Streamable HTTP, SSE or Stdio, or from the command line.[2]

Tools from a server pass through the same tool policy as everything else: “connecting a server does not bypass your policy”. For a server that uses OAuth, openclaw mcp login <name> starts the sign-in.[2]

Running around the clock, and what stops it

Automations are OpenClaw’s built-in scheduler. It persists jobs, wakes the agent at the right time, and can deliver the output to a chat channel, a webhook, or nowhere.[3]

They run inside the Gateway process, not inside the model. The docs put it in one line: “The Gateway must be running for schedules to fire.” Restarts do not lose schedules.[4]

An OpenClaw job therefore depends on your machine being up. A post already handed over does not: Postnavo is launching as the calendar that holds it. A sensible split is to let OpenClaw draft on its schedule and let Postnavo keep the publish times.

Approvals for commands on a real host

When a sandboxed agent wants to run a command on a real host, exec approvals decide. Commands run “only when policy + allowlist + (optional) user approval all agree”.[5]

OpenClaw is clear about the limits: approvals reduce the risk of accidental execution, and are not a per-user auth boundary.[5]

On the CLI route each Postnavo command is one of those commands. Postnavo is launching with its own approval setting as a second check, so an agent that works all night still cannot publish without you while that setting is on.