Skills — Reusable Playbooks

Skills are reusable playbooks the agent follows: a name, a description of when to use it, and step-by-step instructions. The app ships with built-in skills (browser, social, Jev); you can add your own for anything you do repeatedly.

How skills work

When you give the agent a task, it scans skill trigger descriptions and loads the matching skill's instructions into context. The skill then guides the whole task — exact endpoints, exact field names, exact steps, learned lessons.

Think of skills as distilled experience: instead of re-explaining your workflow every time, you write it once and the agent applies it forever.

Add a skill

1. Open Settings → Skills.
2. Create a skill with three parts:
   • Name — short, e.g. deploy-fxembed.
   • Trigger description — "use when…", written like a rule the agent matches against your request.
   • Instructions — the concrete steps, commands, and gotchas.
3. Skills activate automatically when a task matches the trigger — no manual invocation needed.

Writing good skills

• One job per skill. "Deploy the fxembed worker" beats "handle everything about social".
• The trigger description is everything — it's the only thing the agent sees when deciding. Write it as a matching rule: "Use when the user asks to deploy/update the fxembed worker, or mentions Cloudflare social embeds."
• Keep instructions concrete — exact endpoints, exact field names, exact commands. Vague advice ("be careful with auth") doesn't survive contact with a real task.
• Record gotchas — every time something fails once, add the fix to the skill so it never fails the same way twice.
• Test it — ask the agent something that should trigger the skill and check it actually loads.

Built-in skills

Skill	Trigger
Browser	Web automation, CDP, tunnel setup
Social	X/Twitter embeds, fxembed worker
Jev	Browser decision-engine setup, pruning context

What's next

• Agents & Subagents — delegating work to background agents
• Shell — the execution environment skills often drive