Documentation / Working with Genie
Runbooks
Named, versioned procedures Genie applies at the right moment — proposed from repeated work, confirmed by a person, and retired when they stop earning their place.
What a runbook is
A named, versioned, human-readable procedure: when it applies, what to do in what order, what to check afterwards, and when to stop and ask.
Text a technician can read and correct — not a learned policy or an opaque artifact. That property is why it works: a procedure whose correctness can be argued about is a procedure you can trust.
Where they come from
- Shipped with the platform for common work
- Written by you
- Proposed by Genie from work he has seen repeated, with the evidence — how many times, with what outcomes, over what period
Proposed, never adopted
A proposal is inert. It influences nothing until a person approves it. A procedure Genie wrote for himself and started following is indistinguishable from a confident mistake with persistence.
Declared triggers, not similarity
A runbook declares the condition under which it applies — this alert class on this asset class, this ticket type, this stage of onboarding.
Semantic search may rank candidates among triggered runbooks. It never selects one on its own — a declared trigger can be tested adversarially and a similarity score cannot.
They decay
| Status | Meaning |
|---|---|
| Experimental | New, unproven, proposes only |
| Test | Running with outcomes being recorded |
| Stable | Proven by its outcome record |
| Deprecated | Decayed, superseded, or changing no outcome |
A runbook nobody has confirmed decays toward deprecated on its own. It does not sit in production looking authoritative — which is exactly how every vendor's script library became four hundred files nobody trusts.
A runbook grants nothing
Every action a runbook invokes passes the same chokepoint at its own tier. It changes what Genie knows to do; it never changes what he is allowed to do.
Not running it yet? Start a thirty-day evaluation — read-only, on your own estate, no card.
Thirty days · read-only · no card
Run it beside what you already have, against your real clients. It tells you what your tools are reporting that is not true.