google-agents-cli-eval
Evaluate ADK agents with metrics, evalsets, and the iterative eval-fix loop.
correctness-and-error-handling
MUST be used whenever fixing correctness and error handling issues in a Flows app. This skill finds AND fixes bugs, missing error states, unhandled rejections, and edge-case failures — it does not just report them. Triggers: correctness, error handling, bug fix, edge case, crash, unhandled, null, undefined, empty state, loading state, error boundary, try catch, async error, useEffect cleanup, type guard, runtime error, robustness.
Full skill instructions
Find and fix correctness issues and missing error handling in $ARGUMENTS (or the whole app if no argument is given). Work through every step below. Each step searches for problems and then fixes them in place. Only report issues that cannot be auto-fixed.
Read these files before checking anything:
src/main.tsx / src/App.tsx — top-level error boundaries and auth flow**/hooks/*.ts, **/contexts/*.tsx — shared async state**/api/*.ts, **/services/*.ts — CDF SDK call sitesFor each async data source, note:
# Find TODO/FIXME/HACK in critical code paths (not test files)
grep -rn --include="*.ts" --include="*.tsx" -E "(TODO|FIXME|HACK|XXX):" src/ | grep -v ".test." | grep -v ".spec."
# Find "fix" or "broken" or "workaround" markers
grep -rn --include="*.ts" --include="*.tsx" -i -E "(TODO.*fix|workaround|broken|known.?bug|temporary.?hack)" src/
For each match in a critical path (data fetching, rendering, auth, navigation):
Do not leave TODOs in critical paths. Every one must be resolved or converted to a safe fallback.
Every Flows app must have at least one React Error Boundary wrapping the main content so that an unexpected render-time exception shows a user-friendly message instead of a blank screen.
grep -rn --include="*.tsx" --include="*.ts" -E "ErrorBoundary|componentDidCatch|getDerivedStateFromError" src/
If no error boundary exists, create the ErrorFallback component and add the ErrorBoundary wrapper to App.tsx. Install react-error-boundary if not present:
pnpm add react-error-boundary
Then add to App.tsx:
import { ErrorBoundary } from "react-error-boundary";
function ErrorFallback({ error }: { error: Error }) {
return (
<div role="alert" className="p-8 text-center">
<p className="text-lg font-semibold">Something went wrong</p>
<pre className="mt-2 text-sm text-muted-foreground">{error.message}</pre>
</div>
);
}
// Wrap the main content:
<ErrorBoundary FallbackComponent={ErrorFallback}>
<MainContent />
</ErrorBoundary>
Write the updated App.tsx with the ErrorBoundary in place. Do not just suggest it — make the edit.
Search for every async function and Promise chain that does not have error handling:
# Find async functions
grep -rn --include="*.ts" --include="*.tsx" -E "async\s+function|async\s+\(" src/
# Find .then() without .catch()
grep -rn --include="*.ts" --include="*.tsx" -E "\.then\(" src/ | grep -v "\.catch\("
Fix each one:
async functions that lack try/catch: wrap the function body in try/catch. Log the error with context and re-throw so callers/query layers can handle it:async function fetchAssets(sdk: CogniteClient) {
try {
const result = await sdk.assets.list({ limit: 100 });
return result.items;
} catch (error) {
console.error("Failed to fetch assets:", error);
throw error;
}
}
.then() without .catch(): add .catch() to the chain:somePromise.then(handleResult).catch((error) => {
console.error("Operation failed:", error);
});
useQuery/useMutation) missing isError handling: add the error check and error UI to the component:const { data, isLoading, isError, error } = useQuery({
queryKey: ["assets"],
queryFn: () => fetchAssets(sdk),
});
if (isError) return <ErrorMessage error={error} />;
Read each file, make the edit, and write it back.
For each component that fetches data, it must have three distinct UI states:
| State | Required UI |
|---|---|
| Loading | Spinner, skeleton, or loading indicator |
| Error | User-readable message (not a raw error object or blank space) |
| Empty | "No results" / "Nothing here yet" message (not a blank list) |
Search for components that render data without checking loading state:
grep -rn --include="*.tsx" -E "\.(map|filter|find)\(" src/ | grep -v "isLoading\|isPending\|skeleton\|Skeleton"
For each hit, read the component and add the missing states directly:
if (isLoading) {
return <div className="flex items-center justify-center p-8"><Spinner /></div>;
}
if (isError) {
return (
<div role="alert" className="p-4 text-center text-destructive">
<p>Failed to load data. Please try again.</p>
</div>
);
}
.map():if (!data || data.length === 0) {
return (
<div className="p-8 text-center text-muted-foreground">
<p>No results found.</p>
</div>
);
}
Insert these checks in the correct order (loading, then error, then empty) above the existing data render. Write each fixed file.
External data (CDF responses, URL params, localStorage, JSON.parse) must be validated before use. TypeScript types alone are not runtime guarantees.
# Find JSON.parse without validation
grep -rn --include="*.ts" --include="*.tsx" -E "JSON\.parse\(" src/
# Find localStorage reads
grep -rn --include="*.ts" --include="*.tsx" -E "localStorage\.(get|set)Item" src/
# Find useSearchParams usage
grep -rn --include="*.ts" --include="*.tsx" -E "useSearchParams|searchParams\.get" src/
Fix each one:
JSON.parse(x) as T — replace with Zod safeParse:import { z } from "zod";
const MySchema = z.object({ /* fields */ });
const parseResult = MySchema.safeParse(JSON.parse(raw));
if (!parseResult.success) {
console.warn("Invalid stored data, using defaults:", parseResult.error);
return defaultValue;
}
const validated = parseResult.data;
searchParams.get("id") without null check — add nullish fallback:const id = searchParams.get("id") ?? defaultId;
localStorage.getItem(key) used directly — add type guard and fallback:const raw = localStorage.getItem(key);
if (raw === null) return defaultValue;
try {
const parsed = JSON.parse(raw);
// validate parsed shape
return isValidShape(parsed) ? parsed : defaultValue;
} catch {
return defaultValue;
}
Do not cast external data with as MyType — that bypasses runtime safety. Read, fix, and write each file.
Read every component that accesses properties of data returned from CDF or passed via props.
grep -rn --include="*.tsx" --include="*.ts" -E "\w+\[0\]\." src/
Fix each unsafe pattern found:
// Before: asset.properties.space.Asset.name
// After:
const name = asset.properties?.["my-space"]?.["Asset"]?.name ?? "Unknown";
.map() on possibly-undefined array — add nullish fallback:// Before: items.map(renderItem)
// After:
(items ?? []).map(renderItem)
.at() with optional chaining:// Before: items[0].name
// After:
const first = items.at(0)?.name ?? "—";
Read each file with a match, apply the fix, and write it back.
Every useEffect that sets up a subscription, timer, event listener, or async operation that can outlive the component must return a cleanup function.
grep -rn --include="*.tsx" --include="*.ts" -B 2 -A 15 "useEffect" src/
For each useEffect, check whether cleanup is needed and add the cleanup function if missing:
| Pattern | Fix to add |
|---|---|
addEventListener | Add return () => removeEventListener(...) |
setInterval / setTimeout | Add return () => clearInterval(id) / clearTimeout(id) |
| CDF streaming / SSE | Add return () => stream.close() |
fetch / CDF SDK call | Add AbortController: const controller = new AbortController() at the top, pass controller.signal to fetch, add return () => controller.abort(), and guard state updates with if (!controller.signal.aborted) |
| Zustand / event emitter subscription | Add return () => unsubscribe() |
Reference pattern for async effects:
useEffect(() => {
const controller = new AbortController();
async function load() {
try {
const data = await fetchWithSignal(controller.signal);
if (!controller.signal.aborted) setState(data);
} catch (err) {
if (err instanceof Error && err.name !== "AbortError") {
setError(err);
}
}
}
load();
return () => controller.abort();
}, [id]);
Read each effect, add the missing cleanup, and write the file.
For each feature, check and add guards for:
limit and there are more pages, ensure pagination is communicated to the user. Add a "Load more" or pagination indicator if missing.For Atlas tool execute functions, add argument validation at the top of every execute function:
execute: async (args) => {
if (!args.assetId || typeof args.assetId !== "string") {
return { output: "Missing or invalid assetId", details: null };
}
// ... safe to proceed
}
Search for execute functions, read each one, add the validation, and write the file.
Produce a structured report covering:
| Severity | File | Line | Issue | Status |
|---|---|---|---|---|
| HIGH | src/hooks/useAssets.ts | 34 | Unhandled promise rejection | FIXED — wrapped in try/catch |
| MEDIUM | src/components/AssetList.tsx | 12 | No empty state | FIXED — added empty state check |
| MEDIUM | src/auth/flow.ts | 45 | Auth error handling needs product decision | UNFIXED — requires team input |
If no issues are found in a step, state "No issues found" for that step. Do not skip steps silently.
Summarize what was fixed by severity. Flag any remaining HIGH issues that could cause data loss, crashes in production, or misleading UI states, and list them first for immediate attention.
Evaluate ADK agents with metrics, evalsets, and the iterative eval-fix loop.
Postgres performance optimization and best practices from Supabase. Use this skill when writing, reviewing, or optimizing Postgres queries, schema designs, or database configurations.
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
Ultra-compressed code review comments: location, problem, fix on one line each.
Structured debugging methodology that mandates root cause investigation before attempting any fixes.
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
Comprehensive guide for Firebase Crashlytics, including provisioning and SDK usage. Use this skill when the user needs help setting up Crashlytics, adding crash reporting, or using the Crashlytics SDK in their application.
Comprehensive reference guide for writing, debugging, and maintaining Playwright tests across all testing types and scenarios.
Use for Vercel cost and performance optimization on deployed projects, especially Next.js, SvelteKit, Nuxt, and limited Astro apps. Collect Vercel metrics, usage, project config, and code scan results first; investigate only metric-backed candidates; produce ranked recommendations grounded in verified files and version-aware Vercel/framework docs. Trigger for Vercel bill reduction, slow or expensive routes, caching opportunities, Function Invocations, Build Minutes, Fast Data Transfer, Core Web Vitals, Bot Management, Fluid compute, or cost breakdown requests.
Comprehensive competitor analysis from URLs, combining site scraping with SEO and market data into structured profiles.
Read, create, search, and manage Obsidian vault notes via CLI with built-in plugin development and debugging tools.
Comprehensive website auditing across 230+ rules in 21 categories including SEO, performance, security, and accessibility.
Opus.ai: Revolutionize Your Web Experience
Build a no-code AI app in minutes.
An IDE for code migration from legacy to modern frameworks through coding agents.
Automate CGI animation in live-action scenes
Branded artistic QR-code concepts
Launch a website in seconds with AI.
Streamline Your Coding Experience with AI Code Helper
Convert any screenshot or design to clean code.