Auto Optimize
- 8 repo stars
- Updated April 5, 2026
- bluuewhale/auto-optimize
auto-optimize is a Claude Code plugin that runs a profiling-driven benchmark loop so developers can iteratively improve measurable performance metrics with committed experiment history.
About
auto-optimize is a Claude Code plugin marketplace entry that brings autoresearch-style loops to performance engineering. You define a numeric goal and threshold; the plugin creates benchmark and regression infrastructure, locks a baseline, then autonomously profiles, reasons with Opus, applies changes, measures results, and reflects across git-committed iterations. Reach for it when you have a measurable latency or throughput problem and want disciplined, logged optimization instead of guess-and-check tuning.
- Autonomous optimization loop inspired by Karpathy autoresearch
- Requires working regression and benchmark commands before iterating
- Opus planner uses Step-Back, CoT, Self-Consistency, and Pre-mortem reasoning
- Each iteration is a git commit with experiments under experiments/
- Documented 13–32% benchmark gains on HashSmith SwissMap in one prompt
Auto Optimize by the numbers
- Data as of Jul 10, 2026 (Skillselion catalog sync)
/plugin marketplace add bluuewhale/auto-optimizeAdd your badge
Show developers this marketplace is listed on Skillselion. Paste this into your README.
| repo stars | ★ 8 |
|---|---|
| Last updated | April 5, 2026 |
| Repository | bluuewhale/auto-optimize ↗ |
How do I optimize code performance with baselines, benchmarks, and recorded experiments instead of shipping changes I never measured?
Run an autonomous profile-reason-benchmark loop that locks baselines, commits each experiment, and optimizes a numeric performance metric.
Who is it for?
Performance-minded developers with git repos, sub-agent support, and at least one numeric metric plus test commands to validate regressions.
Skip if: Projects lacking measurable benchmarks or teams unwilling to require explicit numeric success thresholds before optimization begins.
What you get
An experiments/ tree captures plans, profiles, iterations, leaderboard rankings, and a final-report.md summarizing the best configuration.
Plugins in this marketplace
1 plugin - install individually after you add the marketplace.
Recommended Marketplaces
FAQ
What must exist before the optimization loop starts?
Confirmed working Regression Test and Benchmark Test commands, plus a specific numeric success target—ambiguous goals like as fast as possible are rejected.
How does each iteration reason about changes?
A dedicated Opus sub-agent runs Step-Back, Chain-of-Thought, Self-Consistency, and Pre-mortem before any code changes, optionally using disassembly analysis.
Are failed experiments kept?
Yes. Every iteration is git-committed, leaderboard.md ranks attempts, and reflexion.md feeds lessons into the next cycle.