5-whys

name	5-whys
description	Drill to root cause by asking why repeatedly until fundamentals emerge — use when a problem recurs or the quick fix keeps failing to hold
argument-hint	[problem or symptom to trace to its root cause]
user-invocable	true
license	MIT

Move past symptoms to find the root cause of a problem. Each "why" peels back a layer of causation. Usually 5 iterations reach the systemic cause that, if addressed, prevents recurrence. 1. State the problem or symptom clearly 2. Ask "Why does this happen?" — record the answer 3. Take that answer and ask "Why?" again 4. Repeat until you reach a root cause (typically 5 levels) 5. Verify: does addressing the root cause prevent the original symptom? 6. Check for multiple causal chains (problems often have more than one root) 7. Identify the most actionable root cause to address first

<output_format>

5 Whys Analysis: [Topic]

Problem Statement

[The symptom or issue observed]

Why Chain

1. Why? [Answer 1]
2. Why? [Answer 2]
3. Why? [Answer 3]
4. Why? [Answer 4]
5. Why? [Root cause]

Root Cause

[The fundamental issue identified]

Verification

[Would fixing this root cause prevent the original problem?]

Action

[Specific step to address the root cause] </output_format>

## 5 Whys Analysis: CI Builds Failing Intermittently

Problem Statement

Integration tests fail roughly 20% of the time on CI but pass locally.

Why Chain

1. Why? Tests depend on a shared database that sometimes has stale data from previous runs.
2. Why? Test teardown doesn't reset the database — it relies on transactions that sometimes don't roll back.
3. Why? Two test suites run in parallel and share a single test database instance.
4. Why? The CI config was copied from a project that ran tests sequentially.
5. Why? No one reviewed the CI config when we added the second test suite six months ago.

Root Cause

CI infrastructure was never updated when test parallelism was introduced. The shared database assumption held for sequential runs but breaks under concurrency.

Verification

Giving each parallel suite its own database instance would eliminate the shared-state race condition, preventing the intermittent failures.

Action

Create per-suite database instances in CI using a template database that's cloned at suite start and destroyed at suite end.

Key Instructions

• If the chain reaches "human error" or "bad luck," keep going — those are symptoms, not root causes; ask why that error was possible.
• Stop when fixing the identified root cause would demonstrably prevent the original symptom from recurring.

Anti-Pattern Guards

1. Stopping at symptoms — "the developer made a mistake" is a symptom; keep asking why that mistake was possible.
2. Single-chain tunnel vision — most real problems have multiple contributing causes; check for branching chains at each level (process step 6).

Handoff

Chainable to: consider (to apply additional mental models), inversion (to verify by imagining how to reproduce the problem)