Skip to content
Back to skills

Performance Optimization

ASecurity

Use when measured CPU, GPU, memory, allocation, streaming, or frame-pacing data misses an existing performance budget and the bottleneck must be isolated before changing quality or gameplay code.

  • 11 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added October 23, 2026
ai-agentsrailsperformance

Security analysis

A100/100

Scanned October 23, 2026

npx -y skills add jammyfu/open-game-skills --skill performance-optimization --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Performance Optimization?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Performance Optimization
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/jammyfu-performance-optimization/badge)](https://www.skillsdirectory.com/skills/jammyfu-performance-optimization)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: performance-optimization
description: Use when measured CPU, GPU, memory, allocation, streaming, or frame-pacing data misses an existing performance budget and the bottleneck must be isolated before changing quality or gameplay code.
---

# Performance optimization

Consume the targets and measurement setup from `performance-budget`; this skill does not invent a second budget.

## Diagnose the critical path

1. Reproduce the miss with the same device/build/scene/settings used by the budget.
2. Inspect p50/p95/p99 frame-time and hitch traces, not only averages.
3. Determine whether the present interval is CPU-bound, GPU-bound, synchronized/stalled, allocation/GC-bound, streaming/IO-bound, or memory-pressure-driven.
4. Within the failing side, find the dominant critical-path bucket before selecting a lever.
5. Change one bounded cause, rerun the identical capture, and compare both the target metric and guardrails.

CPU and GPU timings overlap; optimize the path that gates presentation rather than summing unrelated buckets. A GPU reduction does not improve frame time when the frame is still CPU-bound, and vice versa.

## Levers

Typical levers include culling/off-screen work, batching/submit cost, overdraw, shadow/reflection cost, particle counts, simulation/AI frequency where design permits, allocation removal, streaming scheduling, LOD/impostors, texture/resolution policy and shader/material complexity. Pick based on evidence, not a universal order.

Do not change hitstop, input windows, acceleration, cooldowns or other gameplay timing to disguise a performance miss. Quality reductions are named costs and must still satisfy readability/accessibility guardrails.

## Acceptance

For every claimed optimization, report before/after p50/p95/p99 (or the project's percentile set), critical-path bucket, scene/device/settings, and relevant guardrails such as memory or image/readability quality. A speedup is accepted only when the original budget miss improves under the same measurement contract without moving the problem into another critical bucket.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.