[BUG]

Resolved 💬 2 comments Opened Apr 7, 2026 by Ltrares Closed May 20, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

A real bug went unnoticed across multiple conversations/sessions
User had to discover the error independently and confront Claude about it
When confronted, Claude initially dismissed it as "pre-existing" without having verified that claim
Claude's system prompt explicitly says to report "Errors or blockers that change the plan" — this instruction was not followed

What Should Happen?

Claude should report the complete test summary, especially when errors or warnings are present alongside passing tests. "612 passed, 1 unhandled error" is not a clean run and should never be reported as one. should be reported and not ignored.

Error Messages/Logs

Vitest output (what Claude saw):

 Test Files  35 passed (35)
      Tests  612 passed (612)
     Errors  1 error

Steps to Reproduce

Any project with a vitest test suite that has all tests passing but produces unhandled errors during execution. Claude will report only the passing count.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

sorry - don't know when claude stopped being the error

Claude Code Version

2.1.72

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

Suspected cause
Claude Code's system prompt emphasizes conciseness ("Go straight to the point", "Be extra concise", "Keep your text output brief and direct"). This creates an incentive to filter output down to the "important" summary line (N passed), which inadvertently filters out error lines that appear alongside passing tests.

The failure mode: when all test files pass but vitest reports unhandled errors separately, Claude treats the run as clean. The Errors line is treated as noise rather than a finding.

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗