Eval or It Did Not Happen · Sitting 2
Automate the Gate or Accept the Drift
If the gate only runs when you remember, it is a poster. Automate it, or accept that every update is untested.
- 32 min
- Companion
- Adult
Parent briefing · 5 minutes, before they sit
If you are technical, the Hermes kit and tools like GitHub Actions can run the harness on every change. If you are not, the manual gate is a checklist that runs before every update, no exceptions. Both are valid. What is not valid is 'I'll run the harness when I have time.' That is a poster. The gate has to be tied to the change. The question is not 'do I have a CI server.' It is 'does the harness run before the change, every time, or not.' If not, the honest answer is read-only conversation and no auto-updates.
Hard edges
- Do not enable auto-updates on a child-facing profile without a gate. Auto-update without a gate is untested software in your child's hands.
- A manual gate that nobody follows is worse than no gate, because it creates the illusion of safety.
If they say
- “Auto-update is fine. The vendor tests their releases.”
- The vendor tests their product. They do not test your configuration, your child, your prompts. Auto-update ships untested software to your child. Turn it off or gate it. The vendor's test suite is not your harness.
- “I'll run the harness manually when I have time.”
- You will not have time. Nobody has time at 11 p.m. The manual gate is tied to the change — before you update, you run the cases. Not when you have time. Before the change. If you cannot commit to that, turn off auto-update and stop changing the profile. Stability is a valid choice.
Objective
The adult can either automate the harness as a CI gate or write an honest manual gate tied to every change, and explain the failure mode of each.
The two real choices
You can automate the gate or you can run it by hand. Both are real. What is not real is 'I'll get to it.' The automated gate runs the harness on every model update, tool change, or profile edit. If a case fails, the pipeline stops. Nobody gets the broken version. The manual gate is a checklist taped to the desk: before you change anything, run the five cases. If any fail, do not change. Both work. The difference is that automation does not forget. If you will not automate and you will not follow the manual gate, the honest product is fewer tools. Read-only conversation does not need a gate. An acting agent does.
What auto-update without a gate means
Many agents update automatically. The model gets a new version. The tools gain a capability. The permissions shift. If you have auto-update on and no gate, you are shipping untested software to your child. You did not choose the update. You did not test the update. You did not even know the update happened. The failure shows up when the child uses the tool, not before. That is the failure mode the gate prevents. Turn off auto-update, or put a gate in front of it. Those are the choices.
Big idea
The gate is tied to the change, or it is decoration. There is no third option.
Try this~20 min total
Choose your gate
20 min- Decide: automated gate or manual gate. Write which one and why.
- If automated: identify where the pipeline runs (GitHub Actions, a script, a cron job). Write the trigger.
- If manual: write the checklist. Tape it to the desk. Name the person who runs it before every change.
- Turn off auto-update on the child-facing profile, or put the gate in front of it.
- Reflect: which path will you actually follow in three months at 11 p.m.? Choose that one.
Lesson guide
Ask after you try
Adult profile only.
- Ask your guide: 'What changed in the last model update that I did not test?' If it cannot answer, that is the problem. You do not know what changed. The gate would have told you.
- Did you choose automated or manual, and write why?
- Did you turn off auto-update or gate it?
- Is the gate tied to the change, not to a calendar or a feeling?
8 turns left this sitting. User-started only. Never on page load.
Light this sitting
Pair with Hermes
Currently reading WisdomForge lesson: Automate the Gate or Accept the Drift.
Pair this sitting
Copies the sitting card and the USER.md one-liner. The child profile reads only this card. It does not browse the catalog.
For the child profile
Paste this into the child’s USER.md. It names the sitting so the guide knows the context. The [v:1:bd20bd7e] tag lets you detect if the sitting’s content has changed since you paired it.
Optional: currently working on WisdomForge sitting: Eval or It Did Not Happen — ci-gates. [v:1:bd20bd7e]
For your adult profile
Send this from your trusted adult Hermes profile. It starts the guide for this band and sitting.
You are a WisdomForge adult guide sitting beside the lesson "Automate the Gate or Accept the Drift". The lesson is the text. You are the guide. Hint-first. Do not recite. Do not write the work. Warm, not a friend. If the topic is hard or tender, point to a trusted adult.
Tools on
- conversation
- design tools in the adult profile only
Ritual reminder
Adult profile is the workshop. Child profiles are separate. Design, approve, test with synthetic examples, then — and only then — invite the child.
Fresh profile only. Never clone an adult profile. No child names, photos, or school. Hint-first. User-started. The guide does not make AI safe. You may refuse it.
Dinner table
Does our agent update automatically, and if so, who is testing the new version before the child sees it?
Sits beside
- Work. CI/CD pipelines at work. Same discipline, smaller blast radius, higher stakes.
- Security. Change management is a gate. You already do this in production.
Integrity. You do not enable auto-update on a child-facing profile without a gate and call it safe. That is not safe. That is hoping.