performance-optimization

performance-optimization

A Claude Code plugin that provides language-agnostic performance optimization guidance for C++, Rust, and TypeScript.

Attribution

The content in this skill is distilled from Google's Abseil Performance Guide — specifically the Performance Hints page and the 25 Fast Tips of the Week articles (#7, #9, #21, #26, #39, #52, #53, #60, #62, #64, #70, #72, #74, #75, #79, #83, #87, #88, #90, #93, #94, #95, #97, #98, #99). All credit for the original ideas, techniques, case studies, and latency reference numbers belongs to the Abseil authors (Jeff Dean, Sanjay Ghemawat, Chris Kennelly, and contributors).

This plugin repackages that wisdom into a progressive-disclosure skill format suitable for Claude Code, with cross-language equivalents added for Rust and TypeScript.

Installation

Add the marketplace in Claude Code:

/plugin marketplace add timbrinded/performance-plugin

Then install the plugin:

/plugin install performance-optimization@timbrinded-performance-plugin

Or use the interactive plugin manager (/plugin) and browse the Discover tab to find and install it.

What it covers

Reference file	Topic
philosophy.md	The critical 3%, performance-aware design, back-of-envelope estimation, latency numbers
measurement.md	Profiling, microbenchmarking, hardware counters, measurement ROI
techniques-algorithms.md	Algorithmic complexity, data structure selection, bulk APIs, regex
techniques-memory.md	Compact structs, field layout, arenas, bit vectors, reducing indirections
techniques-execution.md	Fast paths, precomputation, deferred work, specialization, caching
techniques-allocations.md	Allocation reduction, copy avoidance, temporary reuse, object pooling
optimization-process.md	Project selection, rollout strategy, automation, estimation calibration
language-equivalents.md	C++ / Rust / TypeScript mapping table for all techniques

When to use this

Good fit:

• "This endpoint is slow, help me figure out why" — the skill walks through measure-first methodology, helps pick the right profiling tool, and suggests technique categories based on what the profile reveals.
• "I need to pick between a HashMap and a BTreeMap for this lookup table" — the algorithms reference has a data structure decision framework with language-specific recommendations.
• "Estimate whether this batch job can finish in under 10 minutes" — the philosophy reference has latency numbers and back-of-envelope estimation techniques to sanity-check feasibility before writing code.
• "I'm about to optimize this hot loop, what should I think about" — the process reference covers project selection, one-tradeoff-at-a-time rollouts, and reversibility thinking that prevents wasted effort.
• "Port this C++ optimization to Rust" — the language equivalents table maps specific types and patterns (InlinedVector → SmallVec, flat_hash_map → hashbrown, arena → bumpalo).

Poor fit:

• You need a specific profiling tutorial. The skill explains what to measure and why, but does not walk through tool-specific steps like "how to use perf record on this binary" or "how to read a Chrome DevTools flame chart." Use the tool's own documentation for that.
• You need domain-specific optimization. Database query tuning, GPU shader optimization, ML model inference, network protocol optimization — these have their own bodies of knowledge that this skill does not cover.
• You are writing TypeScript and want memory layout control. Roughly 20% of the techniques (cache line optimization, SoA/AoS, SIMD, compiler hints) do not apply to TypeScript at all. The skill marks these honestly as "N/A" but cannot compensate for the runtime not exposing those levers.
• You want a linter or automated checker. This is reference material for a thinking agent, not a static analysis tool. It will not flag performance issues in code unprompted.

Known limitations

Condensed source material. The original Abseil content spans 25+ detailed articles with rich case studies, production anecdotes, and nuanced discussion. This skill distills that into ~9,000 words of reference material. Some depth and context from the originals is inevitably lost — particularly the specific Google-scale examples that motivate each technique. When a recommendation here seems surprising or under-justified, read the original article at abseil.io/fast for the full reasoning.