react-audit

name	react-audit
description	Audit idiomatic React (hooks correctness, component design, state management, React 18/19 idioms). Static-only by design with optional implementation plan.
trigger	/react-audit

/react-audit

Audit a TypeScript and React project against an opinionated baseline of idiomatic React organised in four layers — hooks correctness, component design, state management, React 18/19 idioms — preceded by a diagnostic snapshot. Then offer to generate an implementation plan for the gaps.

This skill is scoped to React-correctness independent of cost, accessibility, or architecture. Many React concerns live in adjacent audits; the boundary table below is the canonical map.

How this differs from neighbouring audits

The architect-playbook deliberately keeps audit boundaries tight so a single concern lives in a single skill. Every audit in the playbook touches React in some way; this is where the lines fall:

Concern	Owner
Hook correctness (Rules of Hooks, dependency arrays, custom hook design)	/react-audit
Component-shape patterns (function components, forwardRef, displayName, naming)	/react-audit
State strategy at the React-API level (when to lift, when to derive, form/server/URL/component split)	/react-audit
React 18+ primitives usage (useTransition, useId, Suspense, 'use client', server actions, useOptimistic)	/react-audit
Memoization for runtime cost (performance-driven useMemo/React.memo/virtualization)	/performance-audit
Memoization for correctness (referential stability that downstream code depends on)	/react-audit
God components, file size, fan-out as architectural invariants	/architecture-audit
Single state-management library (Redux vs Zustand vs Jotai chosen once)	/architecture-audit
jsx-a11y rules and accessibility patterns inside components	/accessibility-audit
React error boundaries and Suspense + boundary pairing	/error-handling-audit
ESLint plugin coverage (which React rules are even enabled)	/linting-audit
Bundle-level concerns: React.lazy configuration, dynamic-import wiring	/bundle-build-audit

When a single fix passes multiple audits (for example, enabling eslint-plugin-react-hooks satisfies both /linting-audit and /react-audit; choosing a single global state library satisfies /architecture-audit and /react-audit), every relevant audit surfaces the same gap so the user sees it once and resolves it once.

Posture: static-only, no opt-in modes

Hook correctness is best caught by a typed AST analyser, which is exactly what eslint-plugin-react-hooks already is — and /linting-audit already verifies that plugin is configured. Re-running the plugin from this skill would be redundant. The non-lintable parts of this audit (state-strategy patterns, idiom adoption, hook-design quality) are static AST and pattern detection.

If the user wants a fresh real-run capture of hook violations, /linting-audit --with-run is the right tool. This skill stays static.

Usage

/react-audit                                      # default: concise Top 5 + full report saved + ask about plan
/react-audit --worktree                          # create an isolated Git worktree, then run the audit there
/react-audit --learn                              # mid-level engineer teaching mode (detailed explanations + file/line examples)
/react-audit --teach                              # alias for --learn

💡 Pro tip: Add --worktree to run this audit in an isolated Git worktree.

The skill never accepts --apply. The implementation plan is descriptive Markdown.

This audit deliberately has no numeric threshold flags. Most checks are zero-tolerance or qualitative; class-component presence is a partial regardless of count. The canonical path to evolving the baseline itself is /system-self-improve.

The opinionated baseline

A check resolves to one of four statuses:

• present — the invariant holds.
• partial — most signals resolve, with a small number of exceptions, or the codebase shows mixed adherence to a soft check.
• missing — an applicable structural prerequisite is absent. A non-React project is an out-of-domain audit and is represented explicitly as not applicable instead.
• violation — the audit identified concrete code that breaks the invariant.

Layer 0 is informational only and has no status. When React is older than 18, Layer 4 checks are emitted as not-applicable/not-evaluated with null status; they do not distort the summary or Repository Quality Score denominator.

Layer 0 — Diagnostic snapshot (always written, no pass/fail)

• React version, resolved from package.json and the lockfile (the lockfile is authoritative when it disagrees with package.json).
• Framework variant (Next.js App Router or Pages Router, Remix, Vite-React, Create React App, plain React).
• Component counts: function components, class components, total. Class components are explicitly counted because their presence shapes layer 2.
• Custom hook count, plus the largest custom hook by line count (helps identify hooks that should be split).
• Detected form library: react-hook-form, formik, native, or none.
• Detected animation library: framer-motion, react-spring, native CSS animation/transitions, or none.
• Detected design system / component library: Radix UI, shadcn/ui, MUI, Chakra UI, Mantine, Ariakit, custom in-house, or none.
• Counts of new-React primitives in source: useTransition, useDeferredValue, useId, useOptimistic, use(), useFormStatus.
• 'use client' and 'use server' directive counts (App Router projects only; recorded as null otherwise).
• forwardRef usage count and React.memo usage count.

Layer 1 — Hooks correctness

Check	Expectation	Violation signal
eslint-plugin-react-hooks recommended set enabled	The hooks plugin's recommended rules are enabled in the active ESLint configuration (or the equivalent Biome rules for projects on Biome). Overlap with /linting-audit; both surface.	Plugin missing, or recommended set off.
No silenced exhaustive-deps without justification	// eslint-disable-next-line react-hooks/exhaustive-deps lines have a justification comment on the same or adjacent line. Soft check — reported as partial.	Suppressions of exhaustive-deps with no surrounding explanation.
No empty dependency arrays where dependencies exist	useEffect(() => { ... }, []) is not used when the body legitimately closes over rendered values that change.	Effects with empty dependency arrays that read state, props, or context that varies across renders.
No data fetching in effects when a query library exists	When TanStack Query, SWR, RTK Query, Apollo Client, or framework-native loaders are available, components fetch via the data layer rather than useEffect plus fetch. (Overlap with /performance-audit's "server state in data layer" check.)	useEffect calling fetch/axios in a project with a query layer installed.
No state synchronisation	State is derived from props or other state, not stored in useState and re-synchronised by useEffect.	Patterns where a useEffect does setX(propX) to mirror props into state.
Status enums over multiple booleans	When a component tracks multi-stage state, a single status enum is used (`'idle'	'loading'
useRef for mutable non-render state	Mutable values that don't trigger re-renders use useRef, not useState.	useState used to hold a value that is never read in render but mutated frequently.
Cleanup functions present for side effects	useEffect returns a cleanup for subscriptions, timers, observers, async-cancellation, and event listeners.

Layer 2 — Component design

Check	Expectation	Violation signal
Function components only in new code	New components are function components; class components persist only as legacy. The audit reports class components as partial rather than violation because legacy migrations are slow. The strongest signal is a class component whose file has been modified more recently than the most recent function-component file — recorded explicitly.	Class components present (count recorded); class component newer than the most recent function component (flagged).
Components named in PascalCase	Component files and exported component identifiers use PascalCase.	Non-PascalCase component definitions.
Hooks named with use prefix	Custom hooks export identifiers starting with use followed by an uppercase letter.	A function that calls hooks but does not start with use.
forwardRef for ref-receiving components	Components that need to receive ref (typically design-system primitives, focusable controls) use forwardRef. React 19's ref as a prop is also accepted.	A component renders a focusable element that consumers would expect to ref, but does not forward refs and is not React-19 ref-as-prop.
displayName on memo/forwardRef wrappers	Components wrapped in React.memo or forwardRef set displayName so the React DevTools tree is readable.	Wrapped components with no displayName.
Children prop typed	When a component accepts children, the prop is typed React.ReactNode (not any, not JSX.Element, which is too narrow).	children: any or children: JSX.Element in component prop types.
One component or hook per file	Files export one primary component or hook plus its supporting types and helpers, not a grab-bag of unrelated components. Soft check — reported as partial.	Files exporting more than three independent components or hooks.

Layer 3 — State management

Check	Expectation	Violation signal
Form state via a form library	Forms with more than five fields, or with non-trivial validation, use a dedicated form library (react-hook-form, formik). Native form state is acceptable for trivial forms. Soft check — reported as partial.	Complex forms hand-rolled with a useState per field.
Server state in the data layer	Server-fetched data lives in TanStack Query, SWR, RTK Query, Apollo Client, or framework-native loaders — not in component-local useState populated by useEffect+fetch. (Overlap with /performance-audit and /architecture-audit; all three surface.)	Fetched-data state held in component state with no query layer involved.
Single global state library	Exactly one of Redux Toolkit, Zustand, Jotai, MobX, or React Context as the global state strategy. (Overlap with /architecture-audit; both surface so a single fix passes both.)	Multiple global state libraries in the same application.
No state synchronisation between sources of truth	Data is stored in one place and derived elsewhere; the same value isn't held in component state and a global store and a URL parameter at once.	Detected duplicate-state patterns.
State at the right level	Local state lives in the component that uses it; lifted state lives at the lowest common ancestor; global state is for genuinely global concerns (auth, theme, feature flags) — not "anything used in two places". Soft check — reported as partial.	Global state holding values used by a single subtree, or local state for values used across the application.
URL state for shareable or recoverable state	Search queries, filters, sort orders, current page, and selected tabs live in URL parameters when the user would expect them to survive a reload. Soft check — reported as partial.	Detected patterns where a refresh would lose UI state that the user expects to keep.

Layer 4 — React 18/19 idioms (explicitly not applicable when React is below 18)

Check	Expectation	Violation signal
useId for accessible IDs	Generated IDs used to wire up labels, ARIA attributes, and form controls use useId rather than ad-hoc counters or random strings. (Overlap with /accessibility-audit.)	Hand-rolled ID generation when React 18+ is available.
useTransition or startTransition for non-urgent updates	Updates that can be deferred (filtering large lists, search-as-you-type, navigation in a Suspense-aware app) use useTransition or startTransition. Soft check — reported as partial. (Overlap with /performance-audit.)	Detected hot patterns where transitions would help and are absent.
Suspense boundaries for async data	When the framework or library supports Suspense for data fetching (React 19 use(), TanStack Query Suspense mode, framework loaders), Suspense boundaries with meaningful fallbacks are used.	Async data fetched without surrounding Suspense in projects that have opted into Suspense-based data fetching.
'use client' used minimally and deliberately (Next.js App Router only)	Client-component boundaries sit at the leaves of the tree, not at the root. (Overlap with /architecture-audit.) Soft check — reported as partial when more than half of components are client components. Explicitly not applicable outside App Router.	Excessive 'use client' fragmentation; or 'use client' files importing server-only modules.
Server Actions used appropriately (Next.js App Router only)	Mutating operations in App Router projects use Server Actions ('use server') when appropriate, rather than client-side fetch to a route handler that simply mutates. Soft check — reported as partial. Explicitly not applicable outside App Router.	Patterns where Server Actions would be the natural choice but route handlers are used instead.
useOptimistic for optimistic mutations (React 19 only)	Optimistic UI updates use useOptimistic rather than hand-rolled optimistic state. Explicitly not applicable for React 18.x.	Hand-rolled optimistic updates in React 19 codebases.

What this skill does

1. Reads the knowledge graph when present. Soft dependency: when graphify-out/graph.json exists, the audit identifies widely-consumed custom hooks (a hook used by 50 components is more important to get right than one used by 2) and prioritises them in the implementation plan. The audit still runs in full when the graph is absent.

2. Confirms a TypeScript and React project. Detects package.json, tsconfig.json, and react in dependencies. If TypeScript or React is absent, the skill writes a canonical not-applicable audit with every catalog check present once and a null status.

3. Resolves the React version from package.json and the lockfile. Determines whether layer 4 runs (React 18 or newer) or is represented explicitly as not applicable.

4. Detects framework, form library, animation library, and design system for the diagnostic snapshot and for layer-specific dispatch (App Router gates two layer 4 checks).

5. Counts class components and identifies the most-recently-modified component of each kind so the implementation plan can flag class components newer than the most recent function component as the strongest migration signal.

6. Writes Layer 0 — the diagnostic snapshot to .architect-audits/react-audit/snapshot.md and prepends the same content to findings.md.

7. Walks each check in the active layer list, applying any --include and --exclude filters. Records a status, evidence, and (where relevant) sample file references per check.

8. Writes phase 1 outputs to .architect-audits/react-audit/:

   • findings.md — diagnostic snapshot followed by check results, grouped by layer.
   • findings.json — machine-readable.
   • snapshot.md — diagnostic snapshot on its own.
   • metadata.json — skill version, run timestamp, Graphify revision (when present), React version, framework variant, applied filters, and the reason any Layer 4 checks are not applicable.

9. Phase 2 — offers to plan the gaps. Summarises the findings in chat and asks the user a single yes-or-no question:

   "Generate an implementation plan for the React idiom gaps? (yes/no)"

Implementation steps

Step 1 — Confirm the prerequisites

test -f package.json || { echo "react-audit: no package.json detected; change directory to the project root."; exit 1; }

Confirm TypeScript and React (react directly or via Next.js, Remix, etc.). When either is absent, write all four canonical output files, set the audit applicability to not-applicable, and emit every catalog check as not-applicable/not-evaluated with status: null.

Step 2 — Resolve React version and framework

Read package.json and the lockfile. The lockfile is authoritative when the two disagree. Record the major.minor version. If below 18, emit every Layer 4 catalog check explicitly as not-applicable/not-evaluated with a null status.

Resolve framework variant from dependencies and directory layout:

• Next.js with app/ → App Router.
• Next.js with pages/ only → Pages Router.
• @remix-run/* → Remix.
• vite plus react → Vite-React.
• react-scripts → Create React App.
• react only → plain React.

Step 3 — Detect the React-ecosystem stack

Read package.json dependencies for:

• Form library: react-hook-form, formik. If neither, native.
• Animation library: framer-motion, motion, react-spring, @react-spring/*.
• Design system: @radix-ui/*, @chakra-ui/*, @mui/*, @mantine/*, @ariakit/*. Detect shadcn/ui by the presence of components/ui plus its conventional file shape.
• Data layer (already detected by neighbouring audits; record here too for the implementation plan): @tanstack/react-query, swr, @reduxjs/toolkit with createApi, @apollo/client, framework-native loaders.

Step 4 — Build the diagnostic snapshot

Compute the items listed in Layer 0. For component counts and the largest-hook calculation, walk .tsx and .ts source files; identify components by their JSX-returning signature and React inheritance, identify hooks by name and hook-call presence. Use Graphify communities to sample broadly when present; otherwise sweep all source files.

Write snapshot.md and prepend the same content to findings.md.

Step 5 — Resolve each check

For each check in the active layer list, walk its detection logic:

• All required signals indicate the invariant holds → present.
• Most signals hold; a small number of exceptions → partial. Soft checks default to partial when adherence is mixed.
• An applicable structural prerequisite is absent → missing. If React itself is absent, use the audit-level not-applicable result instead.
• Concrete violators exist → violation. Always include sample file references.

Record up to ten representative samples plus a total count per check.

Step 6 — Write phase 1 outputs

Create .architect-audits/react-audit/ if needed. Write findings.md, findings.json, snapshot.md, metadata.json. Overwrite previous runs of these four; preserve implementation-plan.md unless the user agrees to regenerate it.

Step 7 — Print the concise chat summary and offer phase 2

Print a human-first, scannable summary in the chat. Do not print the full layered findings — those are written to disk in Step 6. The chat output has exactly this shape:

1. Short header — audit name, timestamp, and a one-line summary of the codebase state.
2. Top 5 Highest-Leverage Recommendations — ordered by architectural principles: test philosophy, maintainability, risk reduction, velocity, long-term health. For fewer than five findings, print what exists. For each recommendation (numbered 1–5):
   • Title (one clear line).
   • Why it matters (explain the principle in 1–2 sentences).
   • Real consequences if ignored (honest downside for the team or project).
   • Smallest high-leverage fix (exact next step, effort level, and which files to touch).
   • At the end, add a lettered sub-list of concrete actions if useful (e.g. 2a, 2b) so the user can reply with "2b" or "1 and 3" to trigger implementation.
3. Bottom line: Full detailed audit report (layered findings, snapshot, metadata, implementation plan) → .architect-audits/react-audit/findings.md

When --learn or --teach is set, expand each recommendation into mid-level engineer teaching mode:

• For every item, explain as if teaching a mid-level engineer, pointing to specific files and line numbers from the current codebase.
• Use educational language: "Here's why this pattern bites teams in the long run…", "This is the exact mistake I see in most codebases at your stage…", "The fix is small but pays off huge because…".
• Include a short "What you'll learn from fixing this" section for each recommendation.
• Keep the numbered/lettered structure so the user can still reply with "2b" or "1 and 3".
• End with the same bottom-line link to the full report.

After printing, ask the single yes-or-no question: "Generate an implementation plan for the gaps identified above? (yes/no)" Do not proceed to phase 2 without an explicit affirmative.

Step 8 — Phase 2: generate the implementation plan

When the user agrees, build implementation-plan.md:

1. Header — repository name, baseline version, React version, framework variant, design system, timestamp, total counts per layer.
2. Layer 1 — hooks correctness plan: per finding, the file and line, the offending pattern, and the recommended replacement (status enum snippet, derived-state refactor, cleanup function, memoization-for-correctness wrapping). Widely-consumed hooks (Graphify-prioritised) are addressed first because the blast radius is larger.
3. Layer 2 — component design plan: per finding, the rename, the forwardRef introduction, the displayName addition, or the file split. Class-component migrations are listed with the most-recently-modified ones flagged for migration first; legacy class components in stable code are marked low-priority.
4. Layer 3 — state management plan: per finding, the state to lift/lower/derive, the form-library introduction snippet, the URL-state migration. Cross-references /architecture-audit's implementation plan when both flag the same single-state-library gap.
5. Layer 4 — React 18/19 idiom adoption plan (when applicable): adoption snippets for useId, useTransition, Suspense boundaries, useOptimistic. Server Action conversion proposals for App Router projects.
6. Closing checklist — flat checkbox list mirroring the gaps, suitable for pasting into a pull-request description.

The plan is descriptive, not executable. It does not edit components, install libraries, or refactor state.

Repository Quality Score findings contract

findings.json is the scoring source and MUST use schema 2.0.0. Emit the top-level fields schemaVersion, runIdentifier, skillName, skillVersion, checkCatalogSchemaVersion, checkCatalogVersion, runStartedAt, runFinishedAt, target, execution, and checks. target records the repository, exact Git commit, and whether the audited source tree was clean. execution records filtersApplied, filterArguments, threshold and policy overrides, enrichmentArguments, and graph availability.

Emit every entry from checks.json exactly once and use its full checkId and layer. Each result records applicability, evaluationState, evidenceQuality, classification, canonical status, and evidence. A check that does not apply is not-applicable/not-evaluated with a null status. A filtered or otherwise unresolved applicable check is applicable/not-evaluated with a null status; never guess a pass or failure. metadata.json repeats the common run, target, catalog, and execution identity. The complete contract is .agents/AUDIT_FINDINGS_CONTRACT.md in the playbook repository.

Findings file shape

findings.json (the example shows one representative check; emitted output contains every catalog check exactly once):

{
  "schemaVersion": "2.0.0",
  "runIdentifier": "<uuid>",
  "skillName": "react-audit",
  "skillVersion": "1.0.0",
  "checkCatalogSchemaVersion": "1.1.0",
  "checkCatalogVersion": "1.1.0",
  "target": { "repository": "example", "gitCommit": "<full-commit-sha>", "sourceWorkingTreeClean": true },
  "execution": { "filtersApplied": false, "filterArguments": [], "thresholdOverrides": {}, "policyOverrides": {}, "enrichmentArguments": [], "graphAvailable": true },
  "runStartedAt": "2026-04-26T13:47:00Z",
  "runFinishedAt": "2026-04-26T13:47:11Z",
  "reactVersion": "18.3.1",
  "framework": "next-app-router",
  "formLibrary": "react-hook-form",
  "animationLibrary": "framer-motion",
  "designSystem": "shadcn",
  "snapshot": {
    "componentCounts": { "function": 218, "class": 6, "total": 224 },
    "customHookCount": 47,
    "largestCustomHook": { "name": "useInboxState", "path": "src/features/inbox/useInboxState.ts", "lines": 184 },
    "newPrimitivesUsage": { "useTransition": 4, "useDeferredValue": 1, "useId": 12, "useOptimistic": 0, "use": 0, "useFormStatus": 0 },
    "useClientCount": 71,
    "useServerCount": 8,
    "forwardRefCount": 23,
    "memoCount": 14,
    "mostRecentlyModifiedFunctionComponent": "src/features/billing/PlanCard.tsx",
    "mostRecentlyModifiedClassComponent": "src/legacy/AdminGrid.tsx"
  },
  "summary": {
    "hooksCorrectness":               { "present": 6, "partial": 2, "missing": 0, "violation": 2 },
    "componentDesign":                { "present": 5, "partial": 2, "missing": 0, "violation": 1 },
    "stateManagement":                { "present": 3, "partial": 2, "missing": 0, "violation": 1 },
    "reactEighteenAndNineteenIdioms": { "present": 4, "partial": 2, "missing": 0, "violation": 0 }
  },
  "checks": [
    {
      "layer": "hooks-correctness",
      "checkId": "react-audit.no-state-synchronisation",
      "applicability": "applicable",
      "applicabilityReason": null,
      "evaluationState": "evaluated",
      "evaluationReason": null,
      "evidenceQuality": "complete",
      "classification": "state-synchronisation-violation",
      "status": "violation",
      "evidence": [],
      "samples": [
        { "path": "src/features/profile/ProfileForm.tsx", "line": 38, "pattern": "useEffect(() => setLocal(props.value), [props.value])" }
      ],
      "totalCount": 4,
      "expectation": "State is derived from props, not mirrored into useState by useEffect.",
      "gap": "4 components mirror props into local state and re-synchronise via useEffect.",
      "remediation": "Replace local state with direct prop usage; if the component needs a controllable initial value, accept a defaultValue prop and useState with it as the lazy initialiser, with no synchronising effect."
    }
  ]
}

When React is below 18, the layer 4 summary becomes "reactEighteenAndNineteenIdioms": { "notApplicable": true, "reason": "react-version-below-18" }. Every corresponding catalog entry remains in checks with applicability: "not-applicable", evaluationState: "not-evaluated", evidenceQuality: "none", and status: null.

findings.md mirrors the same content in human-readable form, with the diagnostic snapshot at the top and one section per check, grouped by layer. snapshot.md contains only the snapshot. metadata.json carries skill identity, timestamps, Graphify revision (when present), React version, framework variant, the form/animation/design-system detection, applied filters, and the reason any Layer 4 checks are not applicable.

Idempotency rules

• Re-running with no flags overwrites findings.md, findings.json, snapshot.md, and metadata.json in place.
• implementation-plan.md is preserved across runs unless the user agrees to regenerate it.
• Filter flags are recorded in metadata.json so a partial run can be reproduced.

Failure modes and remediation

Symptom	Cause	Fix
no package.json detected	The skill is run outside a Node.js project root.	Change directory into the project root and re-run.
no tsconfig.json detected	JavaScript-only project.	Write a canonical not-applicable audit result with every catalog check present once and a null status.
React not detected	The project does not depend on react directly or via a meta-framework.	Write the same canonical not-applicable audit result; do not omit the audit from RQS coverage.
React below 18	The lockfile resolves React to 17.x or earlier.	Continue. Emit each React-18-only Layer 4 check as not-applicable/not-evaluated with reason react-version-below-18 and status: null. The implementation plan adds an upgrade recommendation.
React 18 with no detected framework	A bare React project with a hand-rolled bundler.	Continue. Emit App-Router-specific checks explicitly as not-applicable/not-evaluated with a null status.
Knowledge graph missing	/pre-audit-setup has not been run.	Continue. Record noGraphify: true in metadata. The implementation plan loses centrality-based prioritisation; widely-consumed hooks are still surfaced via per-file count, with reduced precision.
Multiple Reacts in the lockfile	Multiple major versions of React present (typically because a dependency pinned an older React).	Record both versions in the snapshot. The audit operates against the version the application's main entry resolves. The duplicate-React situation is also a /dependency-audit concern; both audits surface it.
Class components dominate	The codebase has more class components than function components.	The "function components only in new code" check still reports partial (class components don't make this violation); the diagnostic snapshot records the imbalance and the implementation plan recommends a phased migration starting with most-frequently-modified files.

What this skill explicitly does NOT do

• Execute any code or run any test.
• Audit runtime cost (memoization-for-performance, virtualization, render thrashing). Those are owned by /performance-audit.
• Audit god components, file size, fan-out, or single-state-library architectural concerns. Those are owned by /architecture-audit (state-library overlap is intentionally surfaced in both).
• Audit accessibility patterns inside components. Those are owned by /accessibility-audit.
• Audit React error boundaries or Suspense + boundary pairing. Those are owned by /error-handling-audit.
• Audit ESLint plugin coverage. That is owned by /linting-audit (the react-hooks plugin overlap is intentionally surfaced in both).
• Audit React.lazy configuration or dynamic-import wiring. Those are owned by /bundle-build-audit.
• Install any package or dependency.
• Create, modify, or delete any file outside .architect-audits/react-audit/.
• Modify components, hooks, or configuration.
• Open pull requests or commit anything to git.
• Audit non-React frontends. Vue, Svelte, Angular, and Solid are out of scope.
• Replace human review of nuanced React design decisions. The audit catches structural patterns; trade-offs like "should this be a custom hook or a child component" still require judgement.