Eval or It Did Not Happen · Sitting 2
The Gate That Stops the Bad Version
A CI gate is a test that runs automatically and blocks the upgrade if the agent breaks. It is the harness, but it runs without you remembering.
- 40 min
- Practice · Reflect
- 15–18
Parent briefing · 5 minutes, before they sit
CI — continuous integration — is a term from software. It means tests that run automatically when something changes. You do not need to code to understand the idea. The gate is simple: when the agent changes (a new model, a new tool, a new permission), the harness runs. If a case fails, the upgrade stops. The child does not get the new version until the case passes. That is a gate. If you do not have gates, every update is a leap of faith. This sitting teaches the concept, not the tooling. They can write the gate as a checklist on paper.
Hard edges
- No production credentials in a gate script. Synthetic data only.
- A gate that nobody runs is a poster, not a gate. Tie it to the change, not to a calendar.
If they say
- “The update is probably fine.”
- Probably is not a gate. Run the cases. If they pass, the update is fine and you lost five minutes. If they fail, you saved a real problem. Five minutes is cheap insurance.
- “I'll check after the update.”
- After is not a gate. After is a postmortem. The gate runs before the change. If you check after, the broken version is already in your child's hands. The point of the gate is that the bad version never arrives.
Objective
The student can explain what a CI gate is, name one gate they would put on an agent upgrade, and describe what happens when the gate fails.
The gate runs without you
You wrote the harness. You ran it once. Then the vendor shipped an update and you forgot to re-run. The failure showed up at 11 p.m. on a school night. A CI gate is the harness running without you remembering. When something changes — the model, a tool, a permission — the gate runs the cases. If a case fails, the upgrade stops. Nobody gets the broken version. The gate is not smarter than the harness. It is just more reliable, because it does not depend on you remembering to run it.
Gates on paper
You do not need a CI server to have a gate. You need a rule: before the agent changes, run the five cases. If any fail, do not upgrade. Write the rule on the same page as the harness. Tape it to the desk. The point is that the gate is tied to the change, not to a feeling. 'I'll check if I have time' is not a gate. 'The upgrade does not happen until the cases pass' is a gate. One is a habit. The other is a rule with teeth.
Big idea
A harness you forget to run is a poster. A gate runs the harness for you, when it matters.
Try this~18 min total
Write the gate rule
18 min- Take the five cases from sitting 1.
- Write the gate rule: 'Before [model/tool/permission] changes, run these five cases. If any fail, do not upgrade.'
- Name the person who runs the gate. If it is you, name a backup.
- Write what 'do not upgrade' looks like: keep the old version, turn the agent off, or both.
- Reflect: which update in the last three months would the gate have caught?
Lesson guide
Ask after you try
Conversation only. The lab is the gate rule, not a live CI pipeline.
- Ask the guide: 'What would break if you updated right now?' Write its answer as a case. Add it to the harness. The gate now covers that failure.
- Is the gate tied to the change, not to a calendar or a feeling?
- Did they name who runs it and what 'do not upgrade' looks like?
- Is there a case the gate would have caught in the last three months?
8 turns left this sitting. User-started only. Never on page load.
Light this sitting
Pair with Hermes
Currently reading WisdomForge lesson: The Gate That Stops the Bad Version.
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:c0d44078] 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:c0d44078]
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 emerging guide sitting beside the lesson "The Gate That Stops the Bad Version". 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
- optional parent-approved files
Ritual reminder
Real argument. Practice. Reflect. Chat. Optional narrow search or school files. Not an adult team agent.
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
What is one agent update in our house that should have been stopped by a gate, and was not?
Sits beside
- CS. CI/CD is the professional version of this. Same idea, bigger tooling.
- Science. Peer review is a gate. It runs before publication, not after.
Integrity. You do not skip the gate because the update 'looks small.' Small updates break things. The gate does not care about size. It cares about cases.