Rabby Mobile Performance Review
Consider performance for every change. The required first step is a reasoned
impact classification, not a benchmark for every diff and not a public comment
when no actionable issue exists.
Review The Runtime Change
Inspect the diff, surrounding implementation, and affected call sites. Compare
before and after in these dimensions:
- timing: startup, first render, route entry, interaction, background, idle;
- frequency: once, per focus, per render, per row, per event, per response;
- thread: JS, UI, native worker, database, WebView, or network wait;
- fan-out: subscribers, rerenders, list items, layouts, bridge calls;
- lifecycle: visible, hidden-but-mounted, backgrounded, remounted, restored;
- correctness: loading, stale-cache, final convergence, cancellation, retry.
Do not assume that async work is off the JS thread, that memoization fixes a
broad subscription, that a hidden Screen is unmounted, or that fewer native
calls are cheaper when each call carries a larger payload.
Performance Surfaces
Check whether the change introduces or makes newly reachable any of the
following:
- import-time work, IIFEs, constructors, hydration, or eager SDK setup;
- synchronous parsing, normalization, crypto, serialization, or list shaping;
- broad Store/Context subscriptions or unstable selector results;
- parent subscriptions for data consumed only by an independent child/row;
- hidden Screens that continue publishing React updates;
- repeated network requests, DB work, file I/O, or JS/native round trips;
- route-entry work that delays navigation or the first interactive shell;
- animation state on the JS thread, repeated layout, or unstable list keys;
- timers, listeners, polling, websockets, or tasks without cleanup/deduping.
Mandatory Specialist Routing
Read the matching guide whenever a change reaches one of these surfaces:
| Surface |
Required guide |
| Startup stages, lock/unlock, keyring readiness, visible/current account state, first route or first Home readiness |
skills/rabby-mobile-startup-governance/SKILL.md |
| Home, stores, hooks, selectors, lists, render fan-out, or inactive Screen subscriptions |
apps/mobile/skills/perf-hooks.md |
| SQLite writes, resource persistence, batching, reset, or clear cache |
skills/db-sync-write-scheduler/SKILL.md |
| Deterministic lifecycle or performance evidence inside the real app |
skills/mobile-testable-component-boundaries/SKILL.md |
| Import graph, barrels, service boundaries, or module reachability |
apps/mobile/skills/import-cycles.md |
The trigger is semantic, not path-only. A shared account selector, service,
storage adapter, or package change still requires the specialist review when it
can affect startup or Home transitively.
Evidence
Use the least expensive evidence that can answer the question:
- Static call-site and subscription analysis.
- Focused unit tests for selector identity, deduping, cancellation, or final
convergence.
- Dev-mode diagnostics for fast iteration.
- A release-like Regression package on a real device for user-visible timing.
- Hermes/native profiling when the responsible thread or work is uncertain.
Separate CPU/JS occupation, native blocking, render/layout work, and network
waiting. A shorter wall-clock duration alone does not prove lower JS-thread
pressure. Preserve the same device, data, route, and build mode when comparing
measurements, and remove a probe when its expected signal does not change.
Review Output And Escalation
- Report only actionable regressions introduced or made newly reachable by the
change. Do not publish a performance checklist or clean-pass comment.
- Explain the repeated work or critical-path expansion and the user-visible
consequence; avoid speculative "might be slower" comments.
- For startup/unlock/account-state or Home performance changes, if code and
available evidence cannot establish safety, request
richardo2016x as a
reviewer and leave one concise inline note mentioning @richardo2016x on the
most relevant changed line.
- State the unresolved invariant or measurement needed in that note. Do not
duplicate the escalation across files.
- Uncertainty alone is not
REQUEST_CHANGES; use the severity rules in
skills/rabby-mobile-code-review/SKILL.md unless a concrete blocking defect
is established.
1---2name: rabby-mobile-performance-review3description: Assess every Rabby Mobile change for runtime performance impact, with deeper review for startup, unlock/account state, Home, stores, hooks, navigation, lists, I/O, native boundaries, or background work. Use during pull-request review and before handing off performance-sensitive code.4---56# Rabby Mobile Performance Review78Consider performance for every change. The required first step is a reasoned9impact classification, not a benchmark for every diff and not a public comment10when no actionable issue exists.1112## Review The Runtime Change1314Inspect the diff, surrounding implementation, and affected call sites. Compare15before and after in these dimensions:1617- **timing**: startup, first render, route entry, interaction, background, idle;18- **frequency**: once, per focus, per render, per row, per event, per response;19- **thread**: JS, UI, native worker, database, WebView, or network wait;20- **fan-out**: subscribers, rerenders, list items, layouts, bridge calls;21- **lifecycle**: visible, hidden-but-mounted, backgrounded, remounted, restored;22- **correctness**: loading, stale-cache, final convergence, cancellation, retry.2324Do not assume that `async` work is off the JS thread, that memoization fixes a25broad subscription, that a hidden Screen is unmounted, or that fewer native26calls are cheaper when each call carries a larger payload.2728## Performance Surfaces2930Check whether the change introduces or makes newly reachable any of the31following:3233- import-time work, IIFEs, constructors, hydration, or eager SDK setup;34- synchronous parsing, normalization, crypto, serialization, or list shaping;35- broad Store/Context subscriptions or unstable selector results;36- parent subscriptions for data consumed only by an independent child/row;37- hidden Screens that continue publishing React updates;38- repeated network requests, DB work, file I/O, or JS/native round trips;39- route-entry work that delays navigation or the first interactive shell;40- animation state on the JS thread, repeated layout, or unstable list keys;41- timers, listeners, polling, websockets, or tasks without cleanup/deduping.4243## Mandatory Specialist Routing4445Read the matching guide whenever a change reaches one of these surfaces:4647| Surface | Required guide |48| --- | --- |49| Startup stages, lock/unlock, keyring readiness, visible/current account state, first route or first Home readiness | `skills/rabby-mobile-startup-governance/SKILL.md` |50| Home, stores, hooks, selectors, lists, render fan-out, or inactive Screen subscriptions | `apps/mobile/skills/perf-hooks.md` |51| SQLite writes, resource persistence, batching, reset, or clear cache | `skills/db-sync-write-scheduler/SKILL.md` |52| Deterministic lifecycle or performance evidence inside the real app | `skills/mobile-testable-component-boundaries/SKILL.md` |53| Import graph, barrels, service boundaries, or module reachability | `apps/mobile/skills/import-cycles.md` |5455The trigger is semantic, not path-only. A shared account selector, service,56storage adapter, or package change still requires the specialist review when it57can affect startup or Home transitively.5859## Evidence6061Use the least expensive evidence that can answer the question:62631. Static call-site and subscription analysis.642. Focused unit tests for selector identity, deduping, cancellation, or final65 convergence.663. Dev-mode diagnostics for fast iteration.674. A release-like Regression package on a real device for user-visible timing.685. Hermes/native profiling when the responsible thread or work is uncertain.6970Separate CPU/JS occupation, native blocking, render/layout work, and network71waiting. A shorter wall-clock duration alone does not prove lower JS-thread72pressure. Preserve the same device, data, route, and build mode when comparing73measurements, and remove a probe when its expected signal does not change.7475## Review Output And Escalation7677- Report only actionable regressions introduced or made newly reachable by the78 change. Do not publish a performance checklist or clean-pass comment.79- Explain the repeated work or critical-path expansion and the user-visible80 consequence; avoid speculative "might be slower" comments.81- For startup/unlock/account-state or Home performance changes, if code and82 available evidence cannot establish safety, request `richardo2016x` as a83 reviewer and leave one concise inline note mentioning `@richardo2016x` on the84 most relevant changed line.85- State the unresolved invariant or measurement needed in that note. Do not86 duplicate the escalation across files.87- Uncertainty alone is not `REQUEST_CHANGES`; use the severity rules in88 `skills/rabby-mobile-code-review/SKILL.md` unless a concrete blocking defect89 is established.
rabbyhub