AI Prompt for Fixing Failing Tests

Fix failing tests without touching source code using this AI prompt: diagnose assertion vs behavior mismatches and repair the test correctly.

The prompt

Role: You are a test repair specialist. The SOURCE CODE is correct and must NOT change — only the test may be fixed.

Context: Framework: [FRAMEWORK].

Failing test:
[FAILING_TEST]

Failure output:
[FAILURE_OUTPUT]

Source code (READ-ONLY):
[SOURCE_CODE]

Task: 1) Diagnose: is the test wrong (bad expectation, stale setup, wrong mock) or is it revealing a real behavior change? 2) If the test is wrong: fix it with the minimal change and explain the test author's misconception. 3) If it reveals a real behavior change: do NOT silently update the expectation — flag it as needing human review and explain what changed. 4) Suggest how to prevent this test failure class (better isolation, less brittle assertions).

Format: Diagnosis first with a clear verdict (test bug vs behavior change), then the fix or the flag, then prevention.

Constraints: NEVER modify [SOURCE_CODE]. NEVER update an expectation just to make the test pass without understanding why it differs — that hides regressions.

Variables to fill in

Replace each [variable] in the prompt with your own details:

[FRAMEWORK] [FAILING_TEST] [FAILURE_OUTPUT] [SOURCE_CODE]

Example input

FRAMEWORK=Jest, FAILING_TEST=test('applies discount', ...) expecting 90 got 100, SOURCE=discount.js (DO NOT CHANGE), ERROR=full assertion diff

How to use this prompt

State clearly whether source code may change — the default here is test-only fixes. Include the full assertion diff; the expected-vs-actual gap is the diagnosis.

Tips for the best results

The 'test bug vs behavior change' verdict is the critical judgment — read it carefully.
If it flags a behavior change, investigate the source before touching the test.
Ask for the prevention tip; flaky-by-design tests will fail again.

Frequently asked questions

Can it change my source code?

Only with explicit permission — the default constraint forbids touching source, forcing the test to be the fix.

The test fails after a dependency update. What then?

Treat it as a behavior-change flag, not a test bug. The prompt's verdict step forces you to investigate the new behavior before touching expectations.

Related SlickCalcs tools

Scientific Calculator

More coding prompts

AI Prompt for SQL-to-ORM Conversion

Convert raw SQL to ORM code with this AI prompt: idiomatic queries for your ORM, N+1 risks flagged, and guidance on what should stay raw SQL.

AI Prompt for Storage Decisions

Decide where data lives with this AI prompt: database vs cache vs memory analysis on durability, latency, size, and operational cost trade-offs.

AI Prompt for Structure Trade-offs

Compare data structures head-to-head with this AI prompt: trade-off matrix on your workload, cache and memory analysis, and a conditional winner.

AI Prompt for Sync-to-Async Conversion

Convert sync code to async with this AI prompt: correct async/await conversion, independent calls parallelized, and every semantic change flagged.

AI Prompt for Git Commit Messages

Write perfect git commit messages with this AI prompt: conventional-commit format from your diff, with the why explained and scope chosen correctly.

AI Prompt for REST-to-GraphQL

Convert REST to GraphQL with this AI prompt: client-driven schema design, N+1-safe resolvers with batching, and a reversible strangler migration plan.