optimize-loop

# optimize-loop > Constraint + mechanical metric + fast verification = autonomous improvement. ## What this skill is - and isn't | Skill | Question it answers | Output | |---|---|---| | `/loop` | "Re-run this prompt every N minutes." | Cron-style recurrence | | `/ralph-loop` | "Bash while-loop external to Claude." | Subprocess churn | | `vd:auto-loop` | "Drive toward a goal until a verifier + audit both vote done." | Goal-pursuit (binary gate, hours-long) | | **`vd:optimize-loop`** | **"Improve a measurable metric over N bounded iterations, auto keep/discard."** | **Best metric value + git-committed wins** | `vd:optimize-loop` **optimizes a number**. It does not pursue subjective goals (`vd:auto-loop` / `vd:cook`) and does not poll on a clock (`/loop`). Each iteration makes one atomic change, commits it, measures, and keeps or reverts on the metric. ## When to Use / When NOT to Use | Use it for | Use something else | |---|---| | Coverage, bundle size, lint/type errors, latency, LOC | Subjective "make it cleaner" → `vd:cook` | | Autonomous bounded iteration (default 10) | Known-root-cause bug → `vd:fix` / `vd:debug` | | Git-tracked experiments with rollback | One-shot task, no repetition → `vd:cook` | | A search space with a consistent numeric evaluator | No mechanical metric → `vd:cook --interactive` | ## Configuration Parsed from the user message. Missing required fields trigger a single batched `AskUserQuestion` in Claude Code; ask the same questions in plain text e...