Add a plugin

A plugin is how the agent gains a new capability — a tool it can call, a channel it can speak on, a scheduled job, a skill. And like everything else in TypeClaw, you don't write it; the agent does. You describe the capability you want, and the agent writes the plugin, wires it in, and restarts itself to load it.

Just ask

From the terminal or any channel where you're the owner, describe the capability:

Build me a plugin with a tool that posts a message to this webhook: https://hooks.example.com/notify

The agent writes the plugin file, registers it, and restarts to pick it up. After that, the new tool is part of its repertoire — it'll use it whenever a task calls for it.

What you can ask for

A plugin can contribute any of these — phrase your request around the outcome, not the machinery:

• A tool — a new action the agent can take ("post to this webhook", "query our metrics API")
• A channel — a new platform it can send and receive on
• A scheduled job — recurring work, on a cron schedule
• A skill — a procedure it loads on demand for a certain kind of task
• A subagent — a focused helper with its own brief, for heavy or specialized work

You can also ask it to give the plugin its own settings (an API endpoint, a threshold, a schedule) that you can tweak later in typeclaw.json without touching the code.

Tweaking and removing

Because the plugin and its settings live in the agent folder, changing them is conversational too: "point that webhook plugin at a different URL", or "remove the notify plugin." The agent edits the file or config and restarts as needed.

Going deeper

The rest of this page is for when you want to understand — or hand-write — the contract the agent is using.

---

Next: Lock down a public channel — once the agent is somewhere strangers can reach it, decide who can do what.