ios-perceived-performance
Use this skill for product-level iOS responsiveness and loading/feedback flows, including perceived latency, time to first feedback, progressive rendering, loading states, skeletons, placeholders, optimistic updates, rollback behavior, high-stakes actions, UI continuity, and responsiveness validation. Do not use it for low-level CPU, memory, rendering, Swift Concurrency, or launch-performance diagnostics unless the task is about user-perceived speed.
Livsy90 · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add Livsy90/iOS-Performance-Agent-Skills --skill ios-perceived-performance技能说明
name: ios-perceived-performance description: Use this skill for product-level iOS responsiveness and loading/feedback flows, including perceived latency, time to first feedback, progressive rendering, loading states, skeletons, placeholders, optimistic updates, rollback behavior, high-stakes actions, UI continuity, and responsiveness validation. Do not use it for low-level CPU, memory, rendering, Swift Concurrency, or launch-performance diagnostics unless the task is about user-perceived speed.
iOS Perceived Performance
Purpose
Use this skill to review whether an iOS screen or flow feels responsive to users, even when raw execution time does not change.
This skill focuses on user-visible feedback, loading behavior, staged content, continuity, optimistic UI, high-stakes actions, and validation of perceived responsiveness.
When to use this skill
Use this skill when the task involves:
- a screen that feels slow, stuck, blank, jumpy, or unresponsive;
- no visible feedback after a tap or gesture;
- delayed first meaningful content;
- loading states, placeholders, skeletons, empty states, retry states, or refresh states;
- progressive rendering, partial content, section-level loading, or stale-while-refreshing UI;
- optimistic updates, pending state, rollback, retries, or local/server reconciliation;
- high-stakes, irreversible, destructive, financial, legal, medical, identity, or security-sensitive actions;
- duplicate submissions or repeated taps during async work;
- perceived latency trade-offs where the product behavior matters as much as the raw duration;
- validation using recordings, UI tests, Instruments, MetricKit, production logs, or user-visible signals.
When not to use this skill
Do not use this skill for:
- low-level CPU, allocation, ARC, memory, runtime, or compiler-performance investigations;
- SwiftUI invalidation, identity, layout, or scrolling problems unless the question is about perceived responsiveness;
- Swift Concurrency internals unless task behavior affects visible UI feedback or staged loading;
- app launch performance unless the question is about first useful screen or first interaction from the user's point of view;
- visual redesign requests with no responsiveness, loading, or feedback concern;
- claims that require running the app when no runtime evidence is available.
Use ios-performance-profiling, swiftui-performance, swift-concurrency-performance, swift-runtime-performance, or ios-launch-performance when those domains are the primary problem.
Core model
Users do not experience CPU samples, network traces, database timings, or async task graphs directly.
They experience:
- whether the app reacts immediately;
- when useful content first appears;
- whether progress is clear;
- whether the screen stays stable while work continues;
- whether failures and retries are understandable;
- whether pending work feels trustworthy;
- whether actions can be repeated accidentally;
- whether the UI tells the truth about the server-authoritative outcome.
Start with the user-visible delay or uncertainty before proposing lower-level optimization.
Agent capability boundaries
The agent can usually inspect and improve:
- state models for loading, loaded, empty, error, refreshing, pending, syncing, and failed states;
- whether UI changes happen only after async work completes;
- whether user actions receive immediate visual feedback;
- whether refresh clears useful existing content unnecessarily;
- whether a screen can render critical content before secondary content;
- whether optimistic updates have pending, rollback, retry, and conflict behavior;
- whether high-stakes actions wait for confirmation before showing final success;
- whether duplicate submissions are prevented.
The agent can suggest, but cannot prove without runtime evidence:
- whether a screen feels fast enough;
- whether a skeleton improves perception;
- whether a deliberate delay is appropriate;
- whether Low Power Mode exposes low performance headroom;
- whether older devices perform acceptably;
- whether a product flow feels trustworthy to users.
Do not claim that manual or device-based validation was performed unless the user supplied evidence such as a recording, trace, test result, profiling output, or production signal.
Review workflow
- Identify the user-visible symptom: no feedback, blank screen, all-or-nothing loading, jumpy update, unclear progress, unsafe optimism, repeated submission, or distrust.
- Locate the state transition that creates the symptom.
- Decide whether the first useful improvement is product/UI feedback or low-level performance work.
- Check whether the UI acknowledges user actions immediately.
- Check whether critical content can appear before secondary content.
- Check whether refresh can preserve useful existing content.
- Check whether loading, empty, error, retry, refreshing, pending, synced, and failed states are distinct enough.
- Decide whether optimistic UI is safe, reversible, and honest.
- For high-stakes actions, prefer explicit progress and server-confirmed success over optimistic final states.
- Propose the smallest change that improves user-visible responsiveness.
- State what can be determined from code and what needs runtime validation.
- Recommend validation using evidence that reflects user experience.
Decision rules
- If the user sees no response after an action, add immediate feedback before optimizing the operation itself.
- If the screen is blank while work is happening, consider cached content, placeholders, skeletons, or progressive rendering.
- If the screen waits for unrelated data before showing anything, consider section-level loading or staged rendering.
- If refresh removes useful content, preserve stale content when it is safe and mark it as refreshing.
- If loading, empty, error, and retry states collapse into one state, separate them before tuning performance.
- If an action