five-whys

name	five-whys
description	Forces 5-level root-cause questioning on any technical decision or justification. Reveals whether the real motive is business value or technical satisfaction. Triggers on: 'five whys', 'why', 'root cause', 'justify this', 'why this approach'. No surface-level answers accepted.

Five Whys — Decision Root-Cause Drill

Interrogate a technical decision or justification through 5 consecutive "why" layers. Stop only when the answer reveals the true motive — or when it becomes clear the motive is self-referential.

When to Trigger

Auto-trigger: User states a technical decision without explicit business justification, or when justification feels post-hoc.

Explicit (/five-whys): User wants to validate a decision before committing, or suspects their own reasoning is surface-level.

Constraints (NON-NEGOTIABLE)

1. No skipping levels — every "why" must be answered before proceeding to the next
2. No circular answers — if answer references the question, flag it immediately
3. No "because it's better" — vague comparatives are rejected; demand specifics
4. Terminate on "because I want to" or "because it's interesting" — this IS the answer; don't go deeper
5. Answer in Korean unless user writes in English

Process

Step 1: State the decision

Ask the user to state the decision in one sentence:

"I am choosing/doing [X] because [reason]."

If the user can't state it in one sentence, the decision isn't clear enough. Ask them to refine.

Step 2: Begin the chain

For each level (1–5), ask:

"Why [previous answer]?"

Record each answer. Classify the answer type:

Answer Type	Signal	Verdict
Business outcome	Revenue, users, retention, cost reduction	✅ Valid root — can stop early
Technical necessity	Performance, security, reliability requirement	✅ Valid root — can stop early
User need	Solves a real user pain point	✅ Valid root — can stop early
Best practice	"It's the standard way", "industry norm"	⚠️ Weak — probe deeper
Technical preference	"Cleaner", "more elegant", "better DX"	🔴 Alert — likely self-referential
Interest/curiosity	"I want to learn", "It's interesting"	🔴 STOP — this IS the root
Circular	References the original decision	🔴 STOP — no real justification

Step 3: Verdict

After the chain completes (or terminates early), deliver the verdict:

PROCEED: Root cause is a valid business/user need (levels 1-3 reached a concrete outcome).

CAUTION: Root cause is a best practice or technical preference. Ask: "Is there a simpler way to achieve the same outcome?"

STOP: Root cause is self-referential, or landed on "because I want to." Recommend either: drop the work, or reframe it explicitly as learning/exploration (not production work).

Step 4: Record

mkdir -p five-whys

Write to five-whys/CHAIN-{timestamp}.md:

---
decision: {one-line decision statement}
verdict: PROCEED | CAUTION | STOP
created: {ISO-8601}
---

# Decision: {decision}

## Chain

1. **Why?** {answer} — _{type}_
2. **Why?** {answer} — _{type}_
3. **Why?** {answer} — _{type}_
4. **Why?** {answer} — _{type}_
5. **Why?** {answer} — _{type}_

## Root Cause

{Final root — the deepest non-circular answer}

## Verdict

{PROCEED/CAUTION/STOP}: {one-line justification}

## Recommendation

{What to do next}

Anti-Rationalization

Excuse	Rebuttal	Do instead
"The reason is obvious"	Then stating it takes 10 seconds. State it.	Run the chain. Obvious reasons survive 5 whys easily.
"This is standard practice"	Standard practice has a reason. What is it HERE?	"Standard" is level 1. Go to level 2.
"I don't have time for this"	You have less time to build the wrong thing.	2-minute exercise vs. days of wrong work.
"I already know why"	Then the chain will confirm it in 60 seconds.	Run it. If you're right, it reinforces confidence.
"It's just how I like it"	That IS a valid answer. Own it explicitly.	Label it "technical preference" and accept the verdict.

Evidence Required

• Decision stated in one sentence
• At least 3 "why" levels completed
• Each answer classified by type
• Root cause identified (non-circular)
• Verdict assigned (PROCEED/CAUTION/STOP)
• Chain recorded to file

Red Flags

• User resists stating the decision in one sentence
• Answers get shorter and vaguer after level 3 (hiding)
• User changes the decision mid-chain (moving target)
• "Why" questions make the user defensive
• Circular reasoning at level 2+ (decision justifies itself)
• Skipping levels to reach a "safe" answer

Agent Instructions

Do

• Push through discomfort — the chain's value IS the discomfort
• Name answer types explicitly ("That's a technical preference, not a business need")
• Accept "because I want to" as a valid and honest answer
• Stop early if a clear business/user root is found (no need for all 5)
• Link the verdict to the root cause directly

Do Not

• Accept vague answers without classification
• Let circular reasoning pass unflagged
• Continue past 5 levels (the technique breaks down)
• Soften the verdict — if it's STOP, say STOP
• Add commentary about the user's reasoning ability