Builder School — Build Your AI Audit & Fix Agent System

MODULE 01

Watch it happen · Understand the roles

Understand Your
Audit & Fix Agents.

Builder School shows how the independent checker and repairer work together so you can focus on understanding the two roles and why they stay separate.

MODULE 01 · WATCH THE QUALITY TEAM WORK

One agent checks. One agent fixes. Then the checker comes back.

You are watching a complete example before Builder School builds both reusable agents in Module 2. The important idea is that the two jobs stay separate.

1

Audit Agent = independent checker

It inspects the product, tests what actually works, verifies evidence, and reports confirmed failures. It does not repair the product.

2

Fix Agent = focused repairer

It receives a confirmed failure, finds the root cause, repairs that problem, protects working areas, and sends the product back for retesting. It does not approve itself.

AUDIT→CONFIRMED FAILURE→FIX→RETEST→PASS

Keep these four rules:

Audit checks and proves. Audit does not repair.
Fix repairs only confirmed failures.
Fix protects parts that already work.
Audit retests the repair. Fix never declares its own work successful.

Builder School exampleSimple Expense Tracker Quality Check

Ready

Carried forward from Course 2Your Approved Product State

Automatic · Read only

Checking Course 2 completion.

Official Product Foundation — Final

Complete Course 2 first.

Product Hard Rules

Complete Course 2 first.

Complete current build state

Complete Course 2 first.

Watch the roles workExample Audit → Fix → Retest Cycle

Learning example

This is a teaching example, not a real bug found in the Simple Expense Tracker. Builder School is showing what the two agents would do if an expense were deleted but Total Spent did not update.

1

Audit checks the rule

The Audit Agent checks Product Hard Rule 5: deleting an expense must remove it from the list and recalculate Total Spent immediately.

AUDIT RESULT: Example failure confirmed with evidence.

2

Audit hands off only the confirmed failure

The Audit Agent reports the exact problem to the Fix Agent. It does not start changing code or redesigning the expense tracker.

HANDOFF: Delete works, but the total did not recalculate.

3

Fix finds the root cause and repairs it

The Fix Agent finds that the delete flow removed the expense but skipped the total-recalculation step. It repairs only that failure and leaves working areas alone.

FIX: Restore the total recalculation after delete.

4

Audit retests the repaired build

The Audit Agent checks Delete again and verifies that Total Spent now changes immediately. It also checks nearby working behavior for regression.

RETEST: Delete and Total Spent now work together.

5

Audit gives the final result

Only the independent Audit Agent decides whether the repaired build passes. The Fix Agent never approves its own repair.

FINAL AUDIT: PASS.

What to rememberThe Whole Lesson in Four Lines

Understanding summary

Audit AgentIndependent checker. Finds and proves real failures.

Fix AgentRepairer. Fixes confirmed failures and protects working areas.

Why separate?The repairer should not be the one that declares its own repair successful.

Finish lineAudit retests the newest repaired build and reports PASS or the next confirmed failure.

Read the example. If anything is unclear, use the Teacher. When the two roles make sense, confirm your understanding.

MODULE 02

Watch it happen · See how both agents are built

How to Build Your
Audit & Fix Agents.

Builder School builds both reusable agent instructions from the current product state while you follow how each role is constructed.

MODULE 02 · WATCH BUILDER SCHOOL BUILD BOTH ROLES

Build the checker first. Build the repairer second. Keep their jobs separate.

The product work from Course 2 and the quality-control lesson from Module 1 are already here. Builder School uses that exact state to create both agent instructions for you.

1

Carry forward the source of truth

Use the approved Product Foundation, Product Hard Rules, complete current build state, and Module 1 quality-control lesson. Nothing gets re-entered or guessed.

2

Build the Audit Agent first

Give the checker one job: inspect, test, verify evidence, classify results, and report confirmed failures without repairing the product.

3

Build the Fix Agent second

Give the repairer one job: accept confirmed Audit failures, find root cause, make the smallest complete repair, protect working areas, and return the build to Audit.

4

Lock the separation

Audit never repairs. Fix never self-approves. Every repair goes back to Audit for retesting and regression checking.

AUDIT AGENT

The checker needs these parts.

Role: independent quality checker.
Mission: find what actually works or fails.
Source of truth: foundation, hard rules, and current build.
Duties: inspect, test, verify, and collect evidence.
Boundary: never repair or quietly change the product.
Output: passes, confirmed failures, unverified items, evidence, and handoff.

FIX AGENT

The repairer needs these parts.

Role: focused repair specialist.
Mission: repair confirmed failures without new damage.
Source of truth: product state plus confirmed Audit findings.
Duties: diagnose root cause, repair, protect, and document.
Boundary: never invent failures or approve its own work.
Output: repair record and return to Audit for retesting.

Keep this simple:

Audit proves. Fix repairs. Audit checks again. Builder School is doing the setup work so you can see exactly what a good pair of agents looks like.

Builder School exampleComplete Audit & Fix Agent System

Ready

Carried forward automaticallyThe Approved Simple Expense Tracker State

Read only · Current state

Checking Module 1.

Official Product Foundation — Final

Complete Module 1 first.

Product Hard Rules

Complete Module 1 first.

Complete current build state

Complete Module 1 first.

Module 1 — Audit & Fix knowledge

Complete Module 1 first.

Builder School builds the checker firstAudit Agent — Version 1

Independent checker

Why it is built this way

The checker gets the approved product state as its source of truth, but it is forbidden from repairing anything. That keeps its evidence independent.

Complete Audit Agent instructions

Builder School builds the repairer secondFix Agent — Version 1

Focused repairer

Why Audit comes first

The Fix Agent receives confirmed Audit failures instead of hunting for its own problems. It must find root cause, repair only what is confirmed, protect working areas, and return the result to Audit.

Complete Fix Agent instructions

See the separationFour Rules That Keep the System Trustworthy

Understanding summary

1. Audit checksIt inspects, tests, verifies, and records evidence.

2. Audit does not repairThe checker never changes the product it is judging.

3. Fix repairs confirmed failuresIt diagnoses root cause and protects working areas.

4. Audit retestsThe repairer never declares its own repair successful.

What you just watchedAudit & Fix Agent System — Version 1

Carries forward automatically

ProductSimple Expense Tracker

Audit AgentComplete independent checker instructions are ready.

Fix AgentComplete focused repair instructions are ready.

NextModule 3 can use both agents without rebuilding them.

Read both finished agent instructions. If anything is unclear, use the Teacher. When the separation makes sense, confirm your understanding.

MODULE 03

Watch it happen · See the full loop finish itself

How to Loop Your
Audit & Fix Agents.

Builder School runs one complete Audit → Fix → Retest loop while you follow how the newest build moves through the quality process.

MODULE 03 · WATCH THE AUTOMATIC LOOP

Start with the current build. End only when Audit says PASS.

Builder School runs the example for you. The important idea is that the two agents keep their separate jobs while the newest build moves through the loop automatically.

The rule: AUDIT → CONFIRMED FAILURE → FIX → RETEST → PASS.

If Audit finds another confirmed failure after retesting, the newest repaired build goes through the same cycle again. The newest build moves automatically between Audit and Fix until Audit reports PASS.

1

Use the newest current build.

The loop always starts from the latest approved Simple Expense Tracker state. Never go backward to an older build after a repair creates a newer one.

2

Audit checks first.

The Audit Agent inspects, tests, verifies evidence, and reports PASS, a confirmed failure, or something it cannot verify. Audit does not repair.

3

Only a confirmed failure moves to Fix.

The Fix Agent receives the exact proven problem and evidence. Unverified guesses do not become repair work.

4

Fix repairs the root cause.

The Fix Agent makes the smallest complete repair, protects working areas, records what changed, and never approves its own work.

5

Audit retests the newest repaired build.

The repaired build goes back to the independent Audit Agent. Audit checks the repair and nearby behavior for regression.

6

Stop only at PASS.

The loop ends when Audit reports PASS with zero confirmed failures remaining. If another confirmed failure exists, the loop repeats automatically.

Keep these four rules:

Audit proves. Audit never repairs.
Fix repairs only confirmed failures.
The newest repaired build always goes back to Audit.
The loop stops only at PASS with zero confirmed failures.

Builder School exampleAutomatic Audit & Fix Loop

Ready

Carried forward automaticallyThe Current Simple Expense Tracker State

Read only · Current state

Checking Module 2.

Official Product Foundation — Final

Complete Module 2 first.

Product Hard Rules

Complete Module 2 first.

Complete current build state

Complete Module 2 first.

Audit Agent — complete instructions

Complete Module 2 first.

Fix Agent — complete instructions

Complete Module 2 first.

Watch the loop runOne Complete Guided Quality Cycle

Automatic handoff

Teaching example: this demonstrates the loop. It is not claiming that a new live failure was found in the current Simple Expense Tracker build.

AUDIT

Confirmed failure in the teaching example

Deleting an expense removes it from the list, but Total Spent does not recalculate immediately. Audit ties the evidence to Product Hard Rule 5 and reports the failure without changing the product.

FIX

Root cause repaired

Fix finds that the delete flow skipped the total-recalculation step. It repairs only that failure, protects the working add/category/save behavior, and returns the newest repaired build.

RETEST

Audit checks the newest repaired build

Audit verifies that Delete now removes the expense and immediately recalculates Total Spent. It also checks nearby behavior for regression.

PASS

Zero confirmed failures remain in the example

Audit reports PASS. The loop stops because the independent checker—not the repairer—confirmed the repaired behavior.

Finished exampleAutomatic Loop Run Report

Builder School completed

What the system followsAutomatic Loop Operating Rules

Carried forward

What you just watchedThe Whole Loop in Four Lines

Understanding summary

Audit first

The independent checker tests and proves what is real.

Fix second

The repairer works only on a confirmed failure and root cause.

Audit again

The newest repaired build goes back for retesting and regression checking.

PASS stops the loop

The cycle ends only when Audit reports zero confirmed failures remaining.

Watch the complete loop. If anything is unclear, use the Teacher. When the automatic handoff between Audit and Fix makes sense, confirm your understanding.

Build Your AI Audit & Fix Agent System

Understand Your
Audit & Fix Agents.

One agent checks. One agent fixes. Then the checker comes back.

Audit Agent = independent checker

Fix Agent = focused repairer

Need a simpler explanation of Audit and Fix Agents?

Audit checks the rule

Audit hands off only the confirmed failure

Fix finds the root cause and repairs it

Audit retests the repaired build

Audit gives the final result

You understand how Audit & Fix Agents work together.

How to Build Your
Audit & Fix Agents.

Build the checker first. Build the repairer second. Keep their jobs separate.

Carry forward the source of truth

Build the Audit Agent first

Build the Fix Agent second

Lock the separation

The checker needs these parts.

The repairer needs these parts.

Need a simpler explanation of how the two agents are built?

You understand how to build separate Audit & Fix Agents.

How to Loop Your
Audit & Fix Agents.

Start with the current build. End only when Audit says PASS.

Use the newest current build.

Audit checks first.

Only a confirmed failure moves to Fix.

Fix repairs the root cause.

Audit retests the newest repaired build.

Stop only at PASS.

Need a simpler explanation of the automatic loop?

Confirmed failure in the teaching example

Root cause repaired

Audit checks the newest repaired build

Zero confirmed failures remain in the example

You understand how the Audit & Fix loop keeps running until PASS.