my-optimization

What problem does it solve?

It helps you iteratively improve a codebase by exploring multiple candidate changes and selecting the best one, instead of relying on a single-shot manual refactor or one-off optimization attempt.

Core Features & Use Cases

• Multi-idea sampling and parallel rollouts: generates multiple optimization ideas, promotes them to tasks, and runs them in isolated worktrees to avoid cross-contamination.
• Reward-based selection (PPO-like): computes multi-objective rewards from CI results and reviews, penalizes overly risky diffs, and selects the best-performing PR.
• Convergent continuous evolution loop: repeats sampling, execution, reward computation, selection, and merge/cleanup until convergence criteria or limits are reached.

Quick Start

Use the skill to optimize performance by generating ideas and running selection until the best PR is merged: assign a FileEvo target in your task context and then start the loop by promoting generated ideas to worktree tasks and running reward selection with your threshold.