Quickstart
Assuming you've installed AgentHub, this page walks through your first real command so you know what to expect.
1. Check it's loaded
Run the capability index — it's LIGHT, so no gate:
/hub:helpOn Codex: @hub help. You should see 17 commands, 20 agents, and 42 skills listed.
2. Run something MEDIUM
Pick a small real task. A bug, a feature idea, anything. Try:
/hub:debug login returns 500 after the reset-password flowBecause debugis MEDIUM, you'll see a one-line preview and a prompt — something like:
MEDIUM: dispatch hub:debugger with systematic-debugging + testing-patterns
estimate: ~18k tokens
Proceed? [y/n/tweak]Type y to run, n to cancel, or type tweak followed by a short instruction to adjust scope before dispatching.
3. Run something HEAVY
The HEAVY tier is where AgentHub really earns its name. Try:
/hub:create minimal Next.js SaaS landing page with Stripe checkoutThe gate now shows planned agents, skills, a token estimate, a MoSCoW scope, and at least two lighter alternatives. Nothing fans out until you approve.
4. Bypass the gate when you know what you want
Pass --yes (or -y) as the first argument to skip the prompt. Usage still logs to .hub/usage.json.
/hub:debug -y login returns 500 after reset-password5. Review what you spent
/hub:ledger weekly (or @hub ledger weekly) shows ISO-week totals by tier, top agents, and top skills — pulled from the local .hub/usage.json file. It never leaves your machine.
Next steps
- Read the recommended workflow for multi-step work.
- Browse the commands reference for the full list.
- Learn the approval-gate tiers in depth.