On completion
A task finishes — Operator decides whether to verify, ship, or queue the next step.
Operator → Automations
Wakes on events, not just prompts.
The Operator does not need you to type for it to do something. It reacts to completions, failures, schedules, and incoming messages — with explicit policies for what it's allowed to do.
Screenshot needed
Ideally a 5s loop: a tool fails (red toast) → an automation triggers → Operator picks it up, retries with a different specialist → result resolves green. Static fallback: list of triggers (on completion / on failure / on schedule / on message) with status dots, plus a side panel showing the most recently-fired automation.
/shots/operator-automations.pngTriggers
Four ways the Operator wakes up.
On failure
A tool errors out — Operator catches it, retries with a different model or specialist, and reports back.
On schedule
Cron-style or one-shot. The Space wakes itself at 9:00 to run the morning brief.
On your message
You write a line. Operator picks it up, decides if it can act, and either runs or asks a clarifying question.
Policies
Three levels of autonomy.
Pick per Space. Tighten when you don't trust it yet, loosen when it earns it.
Observe
Watch only. Useful for shadowing a workflow before granting execute rights.
Local Execute
Run actions inside the Space. No outbound writes to other systems without approval.
Execute & Commit
Full autonomy within declared boundaries — write to the Graph, ship the change, close the loop.