cursor.yml code review workflow fails on every PR (100% failure rate since Jan 2026)

[hampsterx]

Summary

The cursor.yml workflow ("Cursor Code Review") has a 100% failure rate. Out of 100 recent runs: 93 failures, 6 skipped, 1 success (Jan 11, the day it was introduced). Every PR since then shows a failing check.

Failure Details

The cursor-agent CLI installs and verifies successfully, but crashes immediately (exit code 1, ~2s) when executing the review prompt:

code-review  Verify Cursor CLI installation  2026.03.25-933d5a6
code-review  Perform automated code review   ##[error]Process completed with exit code 1.

There is zero output from cursor-agent, so the $OUTPUT variable is empty. Combined with set -e in the shell, the script fails before reaching the CRITICAL_ISSUES_FOUND check.

Likely Cause

The CURSOR_API_KEY secret is either missing, invalid, or expired. The instant exit with no output is consistent with an authentication failure.

Impact

• Every PR shows a failing "Cursor Code Review" check, creating noise for contributors
• BLOCKING_REVIEW is currently false, so it doesn't block merges, but the persistent red X is confusing

Suggestions

1. Verify CURSOR_API_KEY is set and valid in repo secrets
2. Add error handling for empty cursor-agent output:

   OUTPUT=$(cursor-agent --force --model "$MODEL" --output-format=text --print "$PROMPT" 2>&1) || true
   if [ -z "$OUTPUT" ]; then
     echo "::warning::cursor-agent produced no output (possible auth failure)"
     echo "CRITICAL_ISSUES_FOUND=false" >> $GITHUB_ENV
     exit 0
   fi

3. Consider disabling the workflow until the API key issue is resolved

Evidence

# 100 most recent runs
Success: 1  (2026-01-11, the day the workflow was added)
Failed:  93
Skipped: 6

# First failure: 2026-01-11T20:11:34Z (same day as the only success)

[tuannvm]

Thanks @hampsterx , fixed