Jetpack Compose Performance logo

Jetpack Compose Performance

OrganizationPopular
android
jetpack-compose-performance

Diagnose and optimize Android Jetpack Compose UI code. Use this skill when analyzing performance bottlenecks of existing code to minimize recomposition overhead and prevent dropped frames and UI lag.

Overview

Publisherandroid
Repositoryskills
Skill namejetpack-compose-performance
Stars
10K
Forks
654
Bundled files
9
LicenseApache-2.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 9 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by android on GitHub. Read the source before you install it.

Installation

Install the Jetpack Compose Performance AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/android/skills/tree/main/jetpack-compose/jetpack-compose-performance
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/android/skills.git /tmp/skills
mkdir -p .claude/skills
cp -r /tmp/skills/jetpack-compose/jetpack-compose-performance .claude/skills/jetpack-compose-performance
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Jetpack Compose Performance in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Jetpack Compose Performance on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Jetpack Compose Performance is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Foundational prerequisites

Before micro-optimizing composables, ensure standard production flags are in place:

  1. Release Build: Never profile or benchmark debuggable builds; ART runtime verification severely degrades performance.
  2. R8 Minification: Keep R8 enabled to optimize bytecode and coroutine dispatch.
  3. Baseline Profiles: Generate baseline profiles for critical user journeys (cold start, key scrolling feeds) so the ART VM pre-compiles and preloads classes ahead of time.

Optimization audit workflow

  1. Exhaustive Scan : Scan the target code against all 7 core categories in the following sections. Evaluate every @Composable function, modifier chain, and lazy-layout data model against all checklist rules rather than stopping after the first issue found.
  2. Mandatory Reference Read Before Flagging or Editing : Whenever a file matches a rule in Categories 1--7, you MUST open and read the corresponding references/*.md file for the approved fix pattern and full scope guards before reporting a finding or editing code. Never rely on the summary bullets alone to write fixes.

1. State, snapshots, and stability (references/state-and-stability.md)

MANDATORY : Before flagging or fixing any issue in this category, you MUST read references/state-and-stability.md.

  • Phase Deferral (Lambdas) : Wrap frequently changing state reads in lambdas (such as Modifier.offset or Modifier.graphicsLayer) so reads occur in Layout or Draw phase rather than Composition.
  • Prohibit Backwards Writes (Verify Read-Before-Write Order) : Never modify state in a later phase or after it has already been read in the same composition pass (Read -> Write, or Composition Read -> Layout or Draw Write). Prevent backwards writes across phases to avoid first-frame popping, jank, and infinite recomposition loops (Backwards Write). Do not flag a state assignment at the start of a composable (Write -> Read or Composition Write -> Draw Read) as a backwards write---a backwards write strictly requires a read of that state to precede the write.
  • Avoid Redundant State Wrapping in Composition (Write -> Read) : Do not wrap synchronous values in remember { mutableStateOf(...) } only to overwrite them immediately at the top of composition before reading them (for example, var path by remember { mutableStateOf(Path()) }; path = viewModel.getPath()). While not a backwards write, this adds unnecessary slot table and MutableState snapshot overhead (and risks becoming a backwards write if a read is later added preceding the assignment); use a plain local val or read the value directly inside the target phase (such as Modifier.drawWithCache).
  • Explicit remember Keys & Slot Overhead : Always pass all input variables and composable parameters referenced inside remember as calculation keys. Remember that remember stores keys + 1 objects in the slot table and checks key equality on every recomposition; only remember when allocations are heavy or identity must be preserved. Do not pass snapshot state values/properties (e.g. state.value) as keys when wrapping derivedStateOf (passing the State instance itself is fine if the instance reference can change).
  • computedStateOf vs derivedStateOf (Compose 1.13+) : On Compose 1.13+, use computedStateOf for simple condition checks and filtered state from fast-changing inputs (such as scrollState.value > 0), which provides significantly faster invalidation than derivedStateOf (on Compose < 1.13, use remember { derivedStateOf { ... } }). Reserve derivedStateOf for heavy calculations that benefit from multi-read caching, or trivial calculations read frequently enough per composition (roughly 10 / numberOfStatesReadPerCalculation times without invalidating between reads) to make up for the invalidation cost.
  • Synchronous State Derivation : Audit for asynchronous produceState + snapshotFlow blocks observing Compose states (such as scroll positions or collection sizes); replace them with synchronous remember { derivedStateOf { ... } } (or computedStateOf on Compose 1.13+).
  • Trivial Primitives : Prefer plain custom Kotlin getters (get()) over derivedStateOf or computedStateOf for simple boolean expressions and property lookups.
  • Collection State Tracking : Back whole-list updates with standard mutableListOf() and mutableStateOf(list, neverEqualPolicy()) with load-bearing assignments instead of mutableStateListOf().
  • Snapshot Collections : Use SnapshotStateList.toList() for repeated queries or iterations to avoid per-element read observation overhead.
  • Repeated State Reads : Extract repeated Compose State property reads into a local variable at the top of the block.
  • State in Setters : Avoid reading Compose State inside custom property setters; use a raw backing field when triggering change side-effects and batch multi-property updates.
  • Stability & Skipping Trade-Offs (Skipping Isn't Free) : Do not fixate on stability or treat @Immutable, @Stable, or ImmutableList as default solutions. Skipping a composable is not free: each parameter is stored in the slot table as if remembered, and all inputs must be compared on recomposition. Strong Skipping Mode already skips composables with unstable parameters (including standard kotlin.collections.List) when instance equality (===) matches. Annotating models or using ImmutableList forces structural equality (.equals()), which can be significantly more expensive than recomposition itself for large collections or complex classes.
  • Never Add @Stable, @Immutable, or ImmutableList to Collection Models : When writing or optimizing code, never add @Immutable or @Stable to models containing collections (List, Set, Map), and never recommend replacing kotlin.collections.List<T> with ImmutableList<T> (or adding ImmutableList to stability_config.txt). Calling O(N) .equals() on collections on every recomposition pass is more expensive than O(1) instance equality (===) under Strong Skipping Mode.
  • Scope Guard for Removing Existing @Stable / @Immutable Annotations (Large Lazy Datasets Only) :
    • When to Remove : Only recommend removing existing @Stable or @Immutable annotations from data models when both conditions are met: (1) the model is used within a lazy layout (LazyColumn, LazyRow, LazyGrid, Pager, ScalingLazyColumn) as the input feed or item model, and (2) it contains a large dataset or deep nested collection hierarchy (e.g., feed streams, large lists of tags/items, or multi-level nested lists like Post with List<Paragraph> and List<Markup>), where recursive O(N) .equals() checks degrade scrolling and recomposition performance.
    • When to Leave Untouched : For all other existing @Stable, @Immutable, or ImmutableList usages in a codebase---such as non-lazy UI models, standalone components, or models with small/bounded collections (e.g., a card with a few badges, thumbnails, or metadata rows) where dataset size is small or unknown---leave the existing annotations untouched and do not flag them . Never blindly grep and flag all @Stable/@Immutable classes across a codebase without verifying they feed a lazy layout with a large dataset.
  • When to Add @Stable / @Immutable Under Strong Skipping Mode : Flat data classes with val primitive/String properties in Compose modules are already inferred as stable by the compiler and use .equals() automatically---do not add redundant @Immutable annotations to them. Only add @Stable or @Immutable when the compiler cannot infer stability (e.g. classes from non-Compose modules or interfaces) and a data source (such as Room or MVI reducers) allocates new object instances for unchanged items on every emission, provided the class is small, flat, has no collections (List, Set, Map), and its .equals() check is trivial O(1).
  • Make Composables Cheaper Instead of Forcing Skipping: Do not overindex on recomposition counts. Skipping one composable during an active recomposition pass saves little time once composition is scheduled (scheduling and starting the recomposition pass is the bulk of the cost). Focus on making composables cheap to execute, deferring state reads to Layout/Draw, and verifying with benchmarks rather than assuming skipping improves performance.

2. Layouts and measurement optimization (references/layouts.md)

MANDATORY : Before flagging or fixing any issue in this category, you MUST read references/layouts.md.

  • Coordinates & Placement : Avoid onGloballyPositioned; use Modifier.onVisibilityChanged for viewport tracking, Modifier.onPlaced or Modifier.onLayoutRectChanged for local bounds, and query requireLayoutCoordinates() on demand inside Modifier.Node event handlers.
  • Anchor Floating Content via Custom Layout : When attaching floating labels or tooltips to a component, use a custom Layout measuring and placing both children relative to each other. Never track onGloballyPositioned coordinates in MutableState to position floating content with offset, which causes a 1-frame lag and schedules a second composition pass.
  • Phase Deferral for Offsets : Wrap dynamic offsets in lambdas (Modifier.offset { IntOffset(...) }) to skip Composition straight to Layout.
  • Tree Depth & Decorative Wrappers : Avoid redundant wrapper Boxes and decorative Canvas nodes; attach modifiers directly or use Modifier.drawBehind.
  • Subcomposition Overhead : Avoid BoxWithConstraints / SubcomposeLayout inside hot paths and lazy items; prefer standard modifiers, Modifier.layout, or LayoutModifierNode.
  • Intrinsic Measurements: Use intrinsic measurements judiciously and avoid deeply nested intrinsic chains.

3. Lazy layouts optimization (references/lazy-layouts.md)

MANDATORY : Before flagging or fixing any issue in this category, you MUST read references/lazy-layouts.md.

  • Lazy Layout Scope Guard : Directives for key and contentType apply strictly to Lazy containers (LazyColumn, LazyRow, LazyGrid). Never flag standard non-lazy containers (Column, Row, Box) iterating over collections.
  • Mandatory contentType and Primitive key : Always specify both primitive keys (Int, Long, String) and contentTypes for all items in lazy layouts (LazyColumn, LazyRow, LazyGrid). When using items or itemsIndexed, never omit contentType: contentType = { _, item -> item.type }. Omitting contentType prevents Compose from reusing composition slots during scrolling.
  • In-Item Index Lookups : Replace list.indexOf(item) with itemsIndexed to eliminate O(N²) lookups during layout passes.
  • Observable Selection & State Queries : Convert selection tracking from mutableStateListOf to an immutable Set in MutableState (mutableStateOf(emptySet<String>())), mutableStateSetOf(), or Map for O(1) lookups. Pass lambda providers (isSelected: () -> Boolean) to leaf composables to isolate recomposition from parent items scopes.
  • Propagate CompositionLocals in Lazy Items : Avoid querying CompositionLocals (LocalContext.current, LocalDensity.current, LocalSharedTransitionScope.current, scopes, etc.) inside individual lazy item composables; resolve them once in the parent container or screen and propagate the resolved values down as parameters.
  • Defer Shared Transitions in Lazy Items : Put Modifier.sharedBounds or Modifier.sharedElement on lazy items behind visibility indicators (such as Modifier.onVisibilityChanged or viewport visibility checks) so offscreen or pre-fetched items do not instantiate transition scopes.
  • Item Count Derivations : Read collection .size directly instead of wrapping list counts in derivedStateOf.
  • Small & Bounded Lists : Use standard Row or Column with for or forEach loops for small or bounded collections (e.g. 1 to 10 items) to eliminate lazy virtualization, measuring, and pre-fetching overhead.
  • Hoist Shared Flows : Collect shared Flows once in the parent container preceding lazy layouts; never call collectAsStateWithLifecycle() per visible item.
  • Avoid Side Effects in Pre-Fetched Items : Never trigger analytics in LaunchedEffect(Unit) inside lazy items; use Modifier.onVisibilityChanged with dwell thresholds instead.

4. Animation and shared transitions (references/animation.md)

MANDATORY : Before flagging or fixing any issue in this category, you MUST read references/animation.md.

  • Phase Deferral for Animated Properties : Wrap fast-changing animated state reads (alpha, scale, rotation, color) in Modifier.graphicsLayer { ... } to bypass Composition and evaluate directly in the Draw phase.
  • Phase Deferral for Offsets & Size : Wrap animated offsets in Modifier.offset { IntOffset(...) } and animate constraints using Modifier.layout { ... } or Modifier.animateBounds to bypass recomposition during size transitions.
  • Defer Shared Transitions in Fast-Scrolling Lists : Attach Modifier.sharedBounds or Modifier.sharedElement only when visible using Modifier.onVisibilityChanged to avoid modifier and observation instantiation costs on discarded pre-fetched items.
  • Lambda Providers : Pass lambda providers (() -> Float) instead of raw animated values to child composables to keep state reads deferred to Draw or Layout.
  • Persistent Animatable in Nodes : Store Animatable as a persistent member property on Modifier.Node and launch in onAttach() using coroutineScope.launch.

5. Effects, lifecycle, and threading (effects-and-threading.md)

MANDATORY : Before flagging or fixing any issue in this category, you MUST read references/effects-and-threading.md.

  • Prohibit Direct Side-Effects in Body : Never perform side-effects directly in the Composable body during composition; wrap them in SideEffect, LaunchedEffect, or user event callbacks.
  • Prohibit Non-Suspending LaunchedEffect & Empty DisposableEffect : Never use LaunchedEffect for synchronous operations (e.g. logging views or updating objects). Never use DisposableEffect with an empty onDispose {} block as a workaround. Use SideEffect instead (in Compose 1.12+, SideEffect(keys) { ... } supports keys directly; on Compose < 1.12, SideEffect takes no arguments, so check the project's Compose version and track key changes with a remembered holder inside SideEffect { ... }).
  • Cache Allocations in Body : Wrap heavy object allocations (such as date formatters or complex paths) and expensive computations in remember.
  • Combine Related LaunchedEffects : Combine multiple LaunchedEffect(Unit) calls into a single LaunchedEffect with nested launch { ... } blocks to reduce effect node allocations and coroutine launcher overhead.
  • Effect Ordering & Teardown Costs : Understand effect execution order when composition changes are applied (DisposableEffect runs first, then SideEffect, while LaunchedEffect launches a coroutine on the composition's CoroutineContext that is dispatched asynchronously and is not guaranteed to wait until after Draw). Note that canceling a LaunchedEffect incurs coroutine cancellation and teardown overhead; avoid cycling LaunchedEffect keys frequently during animations or scrolling.
  • Main Thread Offloading : Offload blocking I/O, heavy file parsing, and database operations to withContext(Dispatchers.IO) using produceState.
  • No Item-Level Impression Effects in Lazy Layouts : Never use LaunchedEffect or DisposableEffect inside lazy items for impression logging. Lazy items are pre-composed before entering the viewport, causing premature or false impressions. Use Modifier.onVisibilityChanged with dwell thresholds instead.
  • Lifecycle-Aware Flow Collection : Use collectAsStateWithLifecycle() for UI Flow collections that actively produce background work on Android, and collectAsState() for lightweight in-memory StateFlows or non-Android scopes.
  • Hoist Receivers & Listeners : Register system listeners and BroadcastReceivers once at the screen level using DisposableEffect, not inside lazy list items.
  • Guard Optional Effects : Guard effects with if (callback != null) before launching to prevent allocating empty coroutines and effect nodes.
  • Prevent Stale Captures : Use rememberUpdatedState when long-running or looping coroutines in LaunchedEffect reference callbacks or parameters.
  • Replace Legacy Timers : Replace Handler.postDelayed or TimerTask with LaunchedEffect and delay().

6. Graphics and custom drawing (references/graphics-performance.md)

MANDATORY : Before flagging or fixing any issue in this category, you MUST read references/graphics-performance.md.

  • Cache Drawing Allocations : Replace decorative Canvas nodes with Modifier.drawWithCache to cache Path, Paint, and Brush allocations.
  • Prefer AsyncImage instead of Image : Use Coil's AsyncImage for loading images from the internet and painterResource for drawables. Image is not optimized for caching and loading images and may cause performance issues.
  • Scope Fast-Changing Uniforms : remember the RuntimeShader and ShaderBrush in composition, update size uniforms in drawWithCache, and per-frame uniforms in onDrawBehind.
  • Graphics Path Object Reuse : Prefer Path.rewind() over reset() in hot draw loops to retain allocated buffers without reallocation.
  • Value Classes & Primitive Collections : Avoid generic tuples (Pair, Triple) and generic sets (HashSet<Long>); use androidx.collection.MutableLongSet and androidx.compose.ui.util.packInts.

7. Custom modifiers: Modifier.Node (Custom modifiers guide)

  • Modifier.Node Migration : Migrate legacy Modifier.composed to ModifierNodeElement + Modifier.Node (or DelegatingNode when delegating to sub-nodes such as SuspendingPointerInputModifierNode) with capability interfaces (DrawModifierNode, LayoutModifierNode, SemanticsModifierNode, CompositionLocalConsumerModifierNode using currentValueOf(...)) to eliminate per-node subcomposition overhead and enable node reuse (see Create custom modifiers with Modifier.Node):

    kotlin
    // Modifier factory
    fun Modifier.circle(color: Color) = this then CircleElement(color)
    
    // ModifierNodeElement
    private data class CircleElement(
        val color: Color
    ) : ModifierNodeElement<CircleNode>() {
        override fun create() = CircleNode(color)
    
        override fun update(node: CircleNode) {
            node.color = color
        }
    }
    
    // Modifier.Node
    private class CircleNode(
        var color: Color
    ) : DrawModifierNode, Modifier.Node() {
        override fun ContentDrawScope.draw() {
            drawCircle(color)
        }
    }
  • On-Demand Coordinates : Query requireLayoutCoordinates() on demand inside pointer or draw handlers in Modifier.Node rather than implementing GlobalPositionAwareModifierNode.onGloballyPositioned.

IMPORTANT : For more examples on how to migrate common modifiers to Modifier.Node, you must read Create custom modifiers with Modifier.Node.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Jetpack Compose Performance AI skill do?

Diagnose and optimize Android Jetpack Compose UI code. Use this skill when analyzing performance bottlenecks of existing code to minimize recomposition overhead and prevent dropped frames and UI lag.

Why use Jetpack Compose Performance on TypingMind?

Because you install it once and use it with any model. Jetpack Compose Performance is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Jetpack Compose Performance in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/android/skills/tree/main/jetpack-compose/jetpack-compose-performance. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Jetpack Compose Performance?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Jetpack Compose Performance?

As many as you like. As long as a model supports skills, you can use Jetpack Compose Performance with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Jetpack Compose Performance AI skill free?

Yes. It is published on GitHub by android under the Apache-2.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇