mcpbeat Sign in

Preserving State Across Reloads Agent Skill

Use this skill to keep Jetpack Compose state alive across HotSwan hot reloads by understanding the three escalating tiers Compose HotSwan uses (tier 1 targeted recomposition, tier 2 composition reset, tier 3 Activity.recreate) and choosing edits and state holders that stay inside tier 1 where scroll position, lazy items, dialog state, and per-composable remember values all survive. Explains which edits force escalation, which state holders survive each tier, and how to hoist transient UI state when the iteration loop must escalate. Use when the developer says "scroll jumped to top after a hot reload", "lost dialog state", "lazy column re-fetched", "tab selection reset", asks why HotSwan reload escalated to tier 2 or 3, plans a refactor and needs to know which scope it touches, or wants to know which state holders survive composition reset.

3k tokens
context cost
the whole folder, loaded on every use
1
files
instructions only
0
copies elsewhere
how many repositories repackaged it
465
stars on the repo
on the repository, not the skill itself

Install

one command, takes just this skill from the repository
npx skills add https://github.com/skydoves/compose-performance-skills --skill preserving-state-across-reloads

The instruction itself

20 sections, as written by the author

Preserving State Across Reloads: keep edits inside HotSwan tier 1

Compose HotSwan applies every hot reload through three escalating tiers. Tier 1 (targeted recomposition) preserves the most: scroll position, navigation back stack, remember and rememberSaveable values, ViewModel state, lazy layout items, dialog and bottom sheet state. Tier 2 (composition reset) disposes and recreates compositions, dropping per-composable remember. Tier 3 (Activity.recreate()) restarts the Activity and only state held by ViewModel or SavedInstanceState survives.

The iteration loop is fastest and least surprising when every reload stays inside tier 1. This skill covers which edits trigger which tier, which state holders survive each tier, and how to hoist transient UI state when an edit must escalate.

When to use this skill

  • The developer reports "scroll jumped to top after a hot reload", "lost dialog state", "lazy column re-fetched", or "tab selection reset to zero".
  • The developer is about to refactor a composable and wants to know which recomposition scope the change touches.
  • The HotSwan tool window status shows a tier 2 or tier 3 reload and the developer wants to understand why it escalated.
  • The developer asks which state holders survive Activity.recreate() versus a composition reset.
  • The user mentions "tier 1", "tier 2", "tier 3", "composition reset", "HotSwan state preservation", or "rememberSaveable across reload".

When NOT to use this skill

  • The change was rejected as a schema violation entirely (added a parameter, changed a constructor, added a new resource ID) and HotSwan fell back to a full incremental build. See ../understanding-hot-reload-limits/SKILL.md.
  • HotSwan is not installed or the watcher is not running. See ../setting-up-compose-hotswan/SKILL.md.
  • The developer wants an AI agent to drive the loop autonomously. See ../iterating-with-ai-and-mcp/SKILL.md.
  • The recomposition itself is too wide independent of hot reload (a parent invalidating a whole subtree on every state tick). Diagnose with ../../recomposition/debugging-recompositions/SKILL.md first.

Prerequisites

  • Compose HotSwan installed and WATCHING against the running app. Setup lives in ../setting-up-compose-hotswan/SKILL.md.
  • Familiarity with Compose recomposition scopes; cross-link ../../recomposition/debugging-recompositions/SKILL.md for the underlying mechanics.
  • Familiarity with rememberSaveable for state that must survive process death and (in this context) composition reset.

The three tiers

| Tier | Mechanism | Preserves | Loses | Triggered when |

|---|---|---|---|---|

| 1. Targeted recomposition | recomposes only the affected scopes in place | navigation back stack, scroll position, remember, rememberSaveable, ViewModel, lazy layout items, dialog and bottom sheet state | nothing | simple body change inside one composable scope |

| 2. Composition reset | dispose and recreate all compositions from scratch | Activity, ViewModel, navigation (via NavController), rememberSaveable (depends on retention) | per-composable remember values not retained by a Saveable, scroll position not held by a saveable state holder | tier 1 unavailable (theme change, root-scope structural change) |

| 3. Activity.recreate() | recreate the entire Activity | ViewModel, SavedInstanceState | scroll, transient dialog state, anything not saved | composition fails or schema mismatch detected |

The tier that ran is reported in the HotSwan tool window status after every reload. Read it after each edit to confirm the loop stayed where it was supposed to.

Workflow

1. Read the tier from the tool window after every reload

The HotSwan tool window prints the tier (1, 2, or 3) for each reload. If the developer expected tier 1 and the status reports tier 2 or 3, the edit touched a wider scope than intended. Read the tier before deciding whether the lost state is a configuration problem or expected behaviour.

2. If state was lost on tier 2, audit the offended state holder

Tier 2 disposes per-composable remember blocks. Walk the composable that lost state and convert its local remember to rememberSaveable for any value that the developer wants to keep across reloads that may escalate. This is the single highest-leverage change for a HotSwan-driven iteration loop.

3. Hoist user-facing transient state into a ViewModel for long iterations

When the developer is iterating on a screen for a long stretch and individual edits keep escalating to tier 2 or tier 3, hoist transient UI state (selected tab, expanded item, scroll position, dialog open) into a ViewModel. ViewModel survives all three tiers, so the iteration loop never loses the workbench state.

4. Avoid theme and root CompositionLocal mutations during a fast loop

Editing a value used by MaterialTheme, or by any staticCompositionLocalOf, invalidates the root content lambda of the corresponding CompositionLocalProvider. HotSwan cannot scope that to a single recomposition target and escalates to tier 2. Move the colour, dimension, or typography under iteration into a local override on the composable being tuned, then move it back into the theme once the value is final.

5. Avoid staticCompositionLocalOf for any value the developer is editing

Even outside theme, any value provided through staticCompositionLocalOf invalidates the entire content of the provider on change. Use compositionLocalOf (which tracks reads) for values that may change during a hot-reload session.

Patterns

Pattern: local remember lost on tier 2

// WRONG (for tier 2 reloads)
@Composable
fun TabScreen() {
    var selected by remember { mutableIntStateOf(0) }
    Tabs(selected = selected, onSelected = { selected = it })
}
// WRONG because: a tier 2 composition reset disposes remember; the selected tab snaps back to 0 after a reload that escalates.
// RIGHT
@Composable
fun TabScreen() {
    var selected by rememberSaveable { mutableIntStateOf(0) }
    Tabs(selected = selected, onSelected = { selected = it })
}

Pattern: theme colour edit forces tier 2

// WRONG (forces tier 2)
val LightColors = lightColorScheme(primary = Color(0xFFEE0044))
// WRONG because: editing a top-level colour used by MaterialTheme invalidates the root CompositionLocal; the reload escalates to tier 2 and disposes per-composable remember.
// RIGHT (stays in tier 1)
@Composable
fun PrimaryButton(text: String) {
    Button(
        onClick = {},
        colors = ButtonDefaults.buttonColors(containerColor = Color(0xFFEE0044)),
    ) { Text(text) }
}

Move the literal back into the theme once the visual value is final. The escalation only matters during the fast iteration loop.

Pattern: lazy list scroll position survives tier 2 and tier 3

// RIGHT
@Composable
fun Feed(items: List<Item>) {
    val state = rememberLazyListState()
    LazyColumn(state = state) {
        items(items, key = { it.id }) { Item(it) }
    }
}

rememberLazyListState is backed by a Saveable, so the scroll position survives composition reset and Activity recreation. Combined with stable keys, the visible items stay rendered after a reload that escalates.

Pattern: hoist iteration-critical state into a ViewModel

// RIGHT
class FeedViewModel : ViewModel() {
    private val _uiState = MutableStateFlow(FeedState())
    val uiState: StateFlow<FeedState> = _uiState.asStateFlow()
}

@Composable
fun FeedScreen(viewModel: FeedViewModel) {
    val state by viewModel.uiState.collectAsStateWithLifecycle()
    Feed(state.items)
}

ViewModel outlives every tier of HotSwan reload (and configuration changes generally), so screen-level workbench state stays untouched even on tier 3.

Pattern: avoid staticCompositionLocalOf for values under iteration

// WRONG (every edit forces tier 2)
val LocalAccent = staticCompositionLocalOf { Color.Red }
// WRONG because: changing the provided value invalidates the entire content of CompositionLocalProvider; HotSwan cannot scope that and escalates.
// RIGHT
val LocalAccent = compositionLocalOf { Color.Red }

compositionLocalOf tracks reads and only invalidates the actual readers; HotSwan can keep the reload inside tier 1.

Mandatory rules

  • MUST prefer rememberSaveable over remember for any UI state the developer wants to keep across hot reloads that may escalate.
  • MUST keep theme and root CompositionLocal edits out of a fast iteration session; they force tier 2.
  • MUST NOT assume tier 1 always runs. Read the tier reported in the HotSwan tool window after every reload.
  • MUST NOT use staticCompositionLocalOf for values the developer is actively editing during a hot reload session.
  • PREFERRED: hoist transient UI state (selected tab, expanded item, dialog open, scroll position) into a ViewModel when the iteration loop is long.
  • PREFERRED: rely on rememberLazyListState (and the matching grid / pager state holders) for scroll and visible-item state because they are saveable by default.

Verification

  • [ ] Editing a composable body keeps scroll position and dialog state, and the tool window reports tier 1
  • [ ] Editing a MaterialTheme colour escalates to tier 2 (status confirms the tier)
  • [ ] State backed by rememberSaveable survives a tier 2 composition reset
  • [ ] State backed by a ViewModel survives a tier 3 Activity.recreate()
  • [ ] rememberLazyListState keeps scroll position after a tier 2 reload

References

  • HotSwan state-preservation docs: https://github.com/skydoves/compose-hotswan-web (under /docs/state-preservation)
  • HotSwan JetBrains plugin listing: https://plugins.jetbrains.com/plugin/30551-compose-hotswan/
  • Android Developers, saving UI state with rememberSaveable: https://developer.android.com/develop/ui/compose/state-saving
  • Android Developers, ViewModel overview: https://developer.android.com/topic/libraries/architecture/viewmodel
  • Android Developers, CompositionLocal: https://developer.android.com/develop/ui/compose/compositionlocal

Other skills for the same job

different authors, same section of the catalogue
Gemini
by softaworks
×2

Use when the user asks to run Gemini CLI for code review, plan review, or big context (>200k) processing. Ideal for comprehensive analysis requiring large context windows. Uses Gemini 3 Pro by default for state-of-the-art reasoning and coding.

5k tokens
Claude Skills
by ComeOnOliver
×2

Claude Skills meta-skill: extract domain material (docs/APIs/code/specs) into a reusable Skill (SKILL.md + references/scripts/assets), and refactor existing Skills for clarity, activation reliability, and quality gates.

13k tokens scripts
Backend Atomic Commit
by ComeOnOliver
×1

> Pedantic backend pre-commit and atomic commit Skill for Django/Optimo-style repos. Enforces local AGENTS.md / CLAUDE.md, pre-commit hooks, .security/* helpers, and Monty’s backend engineering taste – with no AI signatures in commit messages.

14k tokens
Safe Edit
by ComeOnOliver
×1

Automatically backs up files, saves diffs, uses agents/skills, and ensures modular code (<200 lines) before any implementation. Use this skill for ALL code changes to ensure safe, reversible, and clean implementations.

11k tokens
Sharing Skills
by ComeOnOliver
×1

Use when you've developed a broadly useful skill and want to contribute it upstream via pull request - guides process of branching, committing, pushing, and creating PR to contribute skills back to upstream repository

4k tokens
Od Code Migration
by nexu-io

Default reference pipeline for the code-migration taskKind — code-import → design-extract → token-map → rewrite-plan → patch-edit ↔ build-test devloop → diff-review → handoff.

4k tokens
Review Agent Setup
by wshobson

Configure human-in-the-loop gating for AI agent review actions in Claude Code. Use when setting up a project where an agent may post PR reviews, comments, merges, or edit CI configuration, and you want a cryptographically auditable approval trail with Cedar-enforced gates.

1k tokens
Auto Updater
by anthropics
vendor

> Check installed community skills for updates. Shows a diff and requires explicit approval before applying. Use when the user says "check for updates", "update my skills", "anything new for my installed skills", or when invoked from the registry-sync agent.

2k tokens

How to use it

Copy the folder

Take skydoves/preserving-state-across-reloads from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.