name: reactlynx-best-practices description: Review, write, and refactor ReactLynx code using dual-thread, lifecycle, main-thread script, code-splitting, and profiling best practices.
ReactLynx Best Practices
Use this skill when writing, reviewing, or refactoring ReactLynx code. ReactLynx follows the React programming model, but Lynx's dual-thread runtime changes how side effects, lifecycle timing, event handlers, and main-thread scripts should be reasoned about.
This skill intentionally does not require @ast-grep/napi or any native parser at runtime. The bundled scanner is a lightweight heuristic helper for common issues. The agent must still read the code and apply the rule documents in rules/*.md.
When to Apply
- Writing new ReactLynx components or application code.
- Reviewing ReactLynx code for thread-boundary, lifecycle, event, code-splitting, or performance issues.
- Refactoring code that calls
lynx.getJSModule,NativeModules,runOnMainThread,runOnBackground,lazy,Suspense, oruseLayoutEffect. - Investigating performance traces that include ReactLynx render, diff, commit, patch, or setState events.
Official References
- Thinking in ReactLynx: https://lynxjs.org/next/react/thinking-in-reactlynx.html
- Rendering Process and Lifecycle: https://lynxjs.org/next/react/lifecycle.html
- Main Thread Script: https://lynxjs.org/next/react/main-thread-script.html
- Code Splitting: https://lynxjs.org/next/react/code-splitting.html
- Performance Profiling: https://lynxjs.org/next/react/performance/profiling
Workflow
1. Classify the task
Use one of these modes:
| Mode | Use when |
|---|---|
writing |
The user asks for new ReactLynx code or best-practice guidance |
review |
The user asks to check, audit, explain, or validate existing code |
refactor |
The user asks to fix or rewrite existing code |
If the mode is not explicit, infer it from the user's wording. Prefer review before refactor when code has not been inspected yet.
2. Inspect the code
For repository work, search before editing:
rg "lynx.getJSModule|NativeModules|useLayoutEffect|main-thread:|runOnMainThread|runOnBackground|lazy\\(|Suspense|background only" <target>
Read nearby components, custom hooks, custom components that forward event handlers, Rspeedy config, and performance-related code before making changes.
3. Run the helper scanner when source is available
The scanner catches common background-only and lifecycle issues. It is not a complete parser and must not replace code review.
node -e "
import fs from 'fs';
import { ReactLynxWorkflow, formatScanReport } from '<path_to_skill>/scripts/index.mjs';
const input = '<sourceCodeOrFilePath>';
const sourceCode = fs.existsSync(input) ? fs.readFileSync(input, 'utf-8') : input;
const workflow = new ReactLynxWorkflow('review');
const summary = workflow.reviewCode(sourceCode);
console.log(formatScanReport(summary));
"
4. Apply the rule checklist
Always combine scanner output with these manual checks:
- Dual-thread boundaries: render code may run on the main thread; side effects and native APIs must be background-only.
- Background-only propagation: code called only from background-only code is background-only, but custom prop and custom hook boundaries often need an explicit
'background only'directive. - Lifecycle:
useLayoutEffectis unsupported; useuseEffectfor background side effects or main-thread layout events/refs for layout reads. - Events: normal
bind*/catch*handlers run on the background thread;main-thread:*handlers require'main thread'and have stricter limitations. - MTS: captured values must be JSON-serializable, captured variables cannot be modified, nested main-thread functions are unsupported, and cross-thread calls must use
runOnMainThread()orrunOnBackground(). - Shared modules: import helpers with
with { runtime: 'shared' }only for code sharing, not state sharing. - Code splitting: lazy components need default exports,
Suspense, CSS scope awareness, and error handling for important boundaries. - Profiling: use trace events and readable
displayNamevalues to identify hot render/diff/update paths before optimizing.
5. Refactor safely
For refactor mode:
- Report current findings first.
- Explain which fixes are mechanical and which require human design judgment.
- Apply only scoped changes.
- Re-run the helper scanner or package tests after edits.
Use auto-fixes only as suggestions. The current auto-fixes are designed for detect-background-only diagnostics and should be reviewed before applying.
node -e "
import fs from 'fs';
import { ReactLynxWorkflow, formatFixPlan } from '<path_to_skill>/scripts/index.mjs';
const input = '<sourceCodeOrFilePath>';
const sourceCode = fs.existsSync(input) ? fs.readFileSync(input, 'utf-8') : input;
const workflow = new ReactLynxWorkflow('refactor');
workflow.reviewCode(sourceCode);
const plan = workflow.generateFixPlan();
if (plan) {
console.log(formatFixPlan(plan));
}
"
Rules
| Rule | Impact | Use for |
|---|---|---|
| detect-background-only | CRITICAL | lynx.getJSModule, NativeModules, 'background only', custom event/hook boundaries |
| avoid-use-layout-effect | MEDIUM | Lifecycle and layout reads |
| proper-event-handlers | MEDIUM | bindtap, catchtap, propagation, dataset, custom prop handlers |
| main-thread-scripts-guide | MEDIUM | main-thread:*, useMainThreadRef, cross-thread calls, shared modules |
| code-splitting | MEDIUM | lazy, Suspense, standalone lazy bundles, CSS bundle scope |
| performance-profiling | MEDIUM | ReactLynx trace events, flow IDs, displayName |
| hoist-static-jsx | LOW | Static JSX and render cost |
API Reference
function runSkill(source: string): Diagnostic[];
function runSkillWithFixes(source: string): DiagnosticWithFix[];
function analyzeBackgroundOnlyUsage(source: string): Diagnostic[];
function analyzeLifecycleUsage(source: string): Diagnostic[];
function generateFixes(source: string, diagnostic: Diagnostic): Fix[];
function applyFix(source: string, fix: Fix): string;
function applyFixes(source: string, fixes: Fix[]): string;
function formatScanReport(summary: ScanSummary): string;
function formatFixPlan(plan: FixPlan): string;
class ReactLynxWorkflow {
constructor(mode: WorkflowMode);
reviewCode(source: string): ScanSummary;
generateFixPlan(): FixPlan | null;
applyAutoFixes(source: string): { fixed: string; appliedFixes: Fix[] };
}