Skill runtime — load, execute, unload
When a run needs a hard stop for send, money, or deploy. The runtime loads one skill at a time, gives it only the tools it needs, and unloads it when the job is done. Approval is required before anything irreversible.
How it runs
- Pick by description. Skills are routed by what they say they do, not by guesswork.
- Execute with scoped tools. The skill sees only the tools its task needs, nothing else.
- Unload. Approval required. The skill exits when finished, and any send, spend, or deploy waits for your yes.
Guardrails
Scoped tools, hard stops. Send, money, and deploy always require approval.