failure-patterns

name	failure-patterns
description	Categorizes test failures as real_bug, test_bug, or flaky using heuristic pattern matching
user-invocable	false

Failure Pattern Categorization

When test failures are encountered, classify each failure using these heuristic patterns. Check in order — first match wins.

Category: test_bug

The test itself is broken (bad imports, wrong setup, stale mocks). Fix the test, not the source code.

Indicators (check error message and output for these strings, case-insensitive):

• modulenotfounderror — Python module import failure
• no module named — Python missing module
• import error — general import failure
• cannot find module — Node.js module resolution failure
• module not found — generic module not found
• syntaxerror — syntax error in test code
• referenceerror — undefined variable reference
• undefined is not — accessing property of undefined
• is not a function — calling non-function
• is not defined — using undeclared identifier
• mock combined with not called or expected — mock setup issue
• jest.fn() — Jest mock configuration error
• expected mock — mock expectation failure
• fixture combined with not found or error — test fixture problem
• before combined with hook or setup and error — setup hook failure
• no such file or directory — test referencing missing file
• permission denied — filesystem permission issue in test

Category: flaky

Non-deterministic failure caused by environment, timing, or external dependencies. Usually passes on retry.

Indicators:

• timeout — operation timed out
• timed out — alternative timeout phrasing
• econnrefused — connection refused (external service down)
• econnreset — connection reset
• network combined with error — network failure
• socket combined with error or hang up — socket issue
• connection refused — explicit connection failure
• epipe — broken pipe
• enotfound — DNS resolution failure (BUT check test_bug first — modulenotfounderror contains this)
• etimeout — connection timeout
• rate limit — API rate limiting
• too many requests — HTTP 429
• deadlock — database deadlock
• lock combined with timeout — lock acquisition failure
• resource temporarily unavailable — OS resource exhaustion
• flaky or intermittent — explicitly marked as flaky

Category: real_bug (potential)

The test may have caught an actual bug in the source code. These need investigation — don't auto-fix the test.

Indicators:

• assertionerror combined with expected and got or actual — assertion comparing values
• expect( combined with specific value mismatches — Jest/Vitest assertion
• assert combined with != or == — Python assertion
• Status code mismatches: expected 200 with got 4 or got 5 — wrong HTTP status
• null or undefined combined with expected and a specific value — null pointer issue
• typeerror: cannot read — accessing property of null/undefined (potential source bug)

Important: real_bug classification is a signal, not a certainty. Report it as "potential real bug — investigate" rather than a definitive categorization.

Category: unknown

If no patterns match, classify as unknown. Report the raw error and suggest the user investigate.

Confidence Levels

• high — Multiple indicators matched, or the indicator is very specific (e.g., modulenotfounderror for test_bug)
• medium — Single indicator matched, but it's a common pattern
• low — Weak signal (e.g., just assert without clear context)

Usage

When reporting failures in any agent:

Failed Tests:
  1. test_user_login (tests/test_auth.py:45)
     Category: real_bug (high confidence)
     Error: AssertionError: expected status 200, got 401
     Suggestion: The login endpoint returns wrong status. Investigate the auth handler.

  2. test_payment_webhook (tests/test_payments.py:112)
     Category: flaky (high confidence)
     Error: ConnectionRefusedError: [Errno 61] Connection refused
     Suggestion: External service dependency. Consider mocking the HTTP call.

  3. test_import_feature (tests/test_feature.py:8)
     Category: test_bug (high confidence)
     Error: ModuleNotFoundError: No module named 'features.new_thing'
     Suggestion: The import path is wrong. Check the module name and location.