Hand-curated regression questions Ask Mkl should always answer correctly. A nightly cron (/claude-goldset-cron)
runs each one against the live agent and Slack-pings #mkl on regression.
Each question burns ~$0.02–0.10 per run. Use expected keywords for substrings the reply must contain (case-insensitive),
and expected tools for tool names that should fire (e.g. getOrder, searchDeals).
Question
Expected keywords
Expected tools
Last run
Status
Loading…
New goldset question
Saved questions run nightly via /claude-goldset-cron.
Substrings the reply MUST contain (case-insensitive). Leave blank to skip keyword check.
Tool names that should fire. Leave blank to skip tool check.
