experiment-search-eval
Use this skill when working with A/B experiments related to search or filter features at Cho Tot and the user wants to evaluate keyword detection quality, compare treatment vs control behavior, identify bad auto-filter cases, or decide whether to ship an experiment. Always trigger when the user mentions an experiment_id alongside any of: keyword quality, auto-filter, dead session, filter rate, "ra soat keyword", "xuat keyword", "danh gia experiment", "keyword detect sai", "treatment vs control", "co nen ship khong", "keyword nao bi sai". Trigger also for: "filter_automation_based_on_keyword", "search_foundation", keyword eval, experiment keyword review. This skill runs the full pipeline: BigQuery query (under 20GB), pivot treatment/control, flag bad keywords, 5-sheet Excel export, deep PM-ready summary with verdict (ship/tune/rollback).
@ndmhoang54How to install
This skill is part of the product-delta-force plugin. To use it, install the plugin via one of these methods:
Claude Code: git clone git@github.com:carousell/ct-builder-os.git ~/.builder-os && ~/.builder-os/scripts/bootstrap.sh <your-role>
Claude Desktop: Download product-delta-force.zip → Settings → Customization → Upload
Full skill source is available to team members with repository access.