Optimize anything you can measure. Define a clear goal and run experiments in parallel. Each experiment is tested against a hard threshold or an AI judge score. Keep only the improvements that work. Repeat until you find the best solution.
This approach works for clustering quality, search relevance, build performance, prompt quality, and many other outcomes. The system tracks every result on disk so nothing is lost. Even if your session crashes, the data survives.
Start small and safe. Use simple metrics first. Add AI judgments only when you need them. Run experiments one at a time with a low iteration limit. This builds trust in your measurement process before scaling up.
Global
mkdir -p ~/.claude/skills/ce-optimizeProject
mkdir -p .claude/skills/ce-optimizeSource Repository
Lark Drivelarksuite/cli
Organize Lark cloud files import export manage permissions and versions
Lark Eventlarksuite/cli
Stream Lark events in real time with simple CLI commands
Lark Workflow Meeting Summarylarksuite/cli
Turn your Lark meeting notes into organized weekly summaries with ease
Lark Approvallarksuite/cli
Easily manage Lark approval instances and tasks from your command line
Executing Plansobra/superpowers
Load a plan review critically then execute each task with verifications
Subagent Driven Developmentobra/superpowers
Execute plans with fresh subagents per task and thorough reviews for fast iteration
Churn Preventioncoreyhaines31/marketingskills
Reduce customer churn with smart cancel flows and failed payment recovery
Baoyu Danger X To Markdownjimliu/baoyu-skills
Save X tweets and articles as markdown files with YAML front matter