This skill should be used when the user asks to "benchmark the judge", "compare LLM judges", "pick a new classifier model", "replace the eval judge", or wants to know how accurate, fast and cheap a candidate CLASSIFIER_MODEL is on real eval outputs.
This skill should be used when the user asks to "review an eval run", "check a benchmark run", "investigate eval results", "are these failures real", or shares a GitHub Actions eval/benchmark run URL or Braintrust experiment and wants to know whether failures…
Any Elasticsearch error-log query against app-279-logs-* — these indices use ``lvl`` (keyword) for log severity, not ``severity`` or ``level``.
Guide for investigating slow PostgreSQL queries
AWS Lambda High Latency Troubleshooting
Troubleshooting Application Gateway Problems
Database Connection Troubleshooting
Python Application Memory Troubleshooting