An agent's playbook
A hired role isn't just something you dispatch to. It has standing instructions — what to watch, what it's driving at, and what it may do without asking.
Where it lives
Agents → open a hired role → Manage playbook.
The trust stage
Four stages: shadow, supervised, monitored, trusted, with Promote and Demote. A shadow hire is fully usable for anything reversible — only irreversible outside actions need your nod until it's trusted.
Setup
A checklist of what the role needs connected before its work can run — the systems it reads from, and a computer if any of its workflows run a script. Each item says what it unlocks.
Sensors
- Triggers — things that make it act when something happens.
- Objectives — things it works towards on its own schedule.
Either can be paused.
Standing policies
Your manager instructions: what it may do, and what needs a signature. Some are platform-level and can't be removed — external messages always wait for you until the agent is trusted.
What it's waiting on
Anything it stopped to ask about, with Approve and Reject, plus a log of what it's actually been doing.