Guides iterative performance optimization with stable benchmarks and correctness gates.
npx skills add https://github.com/fallow-rs/fallow --skill perf-loop
Please help me install this Agent Skill. Skill: perf-loop Source: https://github.com/fallow-rs/fallow/tree/main/.agents/skills/perf-loop Command: npx skills add https://github.com/fallow-rs/fallow --skill perf-loop
Performance work often fails because benchmarks are unstable, baselines are missing, or optimizations are kept without reproducible evidence. This Skill enforces a disciplined loop so every optimization is measured, verified, and regression-free.
Use the perf-loop skill to guide my next performance optimization on this codebase with a stable benchmark and correctness checks.
High-intent search queries and answers about installing and using this skill.
Follow a measured loop: pick a stable benchmark, record a baseline, profile the hot path, make one bounded change, then re-run the same benchmark and correctness checks. Keep the change only if the improvement reproduces and no contract regresses.
Preserve the benchmark's identity and workload, record a statistically useful baseline before editing, and re-run the identical benchmark after the change. Use a new benchmark identifier only when the workload materially changes.
Gains often vanish when measured from debug builds or incomparable fixtures. Always measure optimized builds against the same workload, and discard results that cannot be reproduced across runs.
Create a new benchmark identifier when the workload changes materially, since comparing against the old baseline would be invalid. Keeping benchmark identity stable is what makes before-and-after evidence meaningful.
Benchmark results only reflect the measured workload and environment, so they can miss real-world variance. This loop mitigates that by requiring profiling first, correctness checks, and a final review step before accepting changes.
cecon123Profile Rust hot paths and validate performance improvements with flamegraph and criterion.
dogearedCollect navigation timing metrics and resource usage to benchmark web application performance.
dhntUpdate failing test suites after intentional contract changes by correcting assertions, fixtures, and golden files.
ntholm86Evaluate system architecture for redesign versus incremental improvement.
