AI Product-Facing Guide

Purpose

Use this guide after setup is already done.

It is the canonical public post-setup guide for product-facing teammates such as:

• product owners
• founders
• QA
• operators
• anyone who needs product answers from the repo without reading code directly

If you still need to choose a client, clone the repo, or decide whether setup should bootstrap anything, start here first:

• Setup/AI_EASY_SETUP.md

What To Open

Best:

• open the repo root

Good enough:

• open one project root such as <ProjectName>/

Do not open:

• Assets/
• Assets/Scripts/
• a random nested subfolder

Best Prompt Shape

Use this shape in a fresh chat:

We are working in this Unity repo.
Read Agents.md first.
Use XUUnity product protocols.
Answer for a product-facing teammate.
Verify current behavior against source code before answering.
If project memory and code disagree, code wins.
Use verification status in the answer.
For startup, SDK, manifest, plist, entitlement, privacy, or compliance-sensitive questions, use code-first verification.

Then ask one focused prompt such as:

xuunity product explain this feature
Project: <ProjectName>
Feature: rewarded continue after fail
Goal: explain current user-visible behavior

Good rule:

• one project per chat
• one question per prompt
• say what decision you need

The Main Commands

Explain current behavior

xuunity product explain this feature
Project: <ProjectName>
Feature: <feature or flow>
Goal: understand current user-visible behavior

Explain ownership

xuunity product brief this system
Project: <ProjectName>
System: <system name>
Goal: understand ownership and responsibility split

Understand change impact

xuunity product impact this flow change
Project: <ProjectName>
Change: <proposed change>
Goal: understand blast radius before approval

Understand dependencies

xuunity product deps this popup
Project: <ProjectName>
Popup/Feature/Flow: <name>
Goal: understand upstream and downstream dependencies

Decide rollout readiness

xuunity product rollout this feature
Project: <ProjectName>
Feature: <feature or change>
Goal: decide whether rollout is safe now

Understand bug impact

xuunity product bug this issue
Project: <ProjectName>
Issue: <bug description>
Goal: decide urgency and business impact

Pressure-test a proposal

xuunity po evaluate this feature
Project: <ProjectName>
Feature: <proposal>
Goal: pressure-test scope and acceptance criteria

Check whether product answers are trustworthy

xuunity product health this project
Project: <ProjectName>
Goal: understand whether AI product answers are reliable enough to use

xuunity project memory freshness this project
Project: <ProjectName>
Goal: decide whether memory-first answers are still safe

Infrastructure-Sensitive Rule

For startup-, SDK-, manifest-, plist-, entitlement-, privacy-, or compliance-sensitive questions:

• treat the answer as code-first
• use project memory for navigation and scoping, not as final proof
• if project memory and code disagree, code wins

How To Read The Answer

Verification status:

• verified in source code
  • highest confidence
• based on project memory
  • useful, but may be outdated
• partially inferred
  • directional only; ask for a verification pass

Most useful sections:

• What it does
• What can break
• Dependencies
• Rollout risk
• Recommended next step

If The Answer Needs Adjustment

If the answer is too technical:

Re-answer for a product-facing teammate.
Keep code details minimal.
Keep product risks, dependencies, and rollout implications.

If the answer is not verified enough:

Do a quick source-code verification pass.
If project memory is outdated, say so clearly and answer based on code.

Main Working Rules

• Prefer repo root when possible.
• Keep one project per chat.
• If the answer starts mixing project contexts, restart the chat and restate Project: <ProjectName>.
• Use product health when you are unsure whether the project is structurally ready for trustworthy AI product work.
• Use project memory freshness when the answer feels stale or memory-first trust is questionable.