feat(custom-widgets): harden AI authoring (credential redaction, JSX policy)

Carve the custom-widget AI-authoring hardening out of #6691 into its own
change, independent of the widget/integration architecture refactor:

- add definition-security redaction helpers that strip credentials/URLs from
  the AI context, response, and generated text
- tighten the authoring prompt and examples
- enforce Tabler icon prop policy in the JSX analyzer
- expand ai-prompt, jsx-policy, and evaluation test coverage

Self-contained in @homarr/custom-widgets; no dependency on the metadata
catalog work. Typecheck + 490 package tests pass.

packages/custom-widgets/scripts/ai-assistant-evaluation.ts

@@ -144,7 +144,8 @@ export const customWidgetAssistantEvaluationToolDefinitions: ToolDefinition[] =
     type: "function",
     function: {
       name: "customWidget_getComponent",
-      description: "Get installed documentation and safety rules for one named Custom JSX component.",
+      description:
+        "Get installed documentation and safety rules for one named Custom JSX component. Fetch only a selected component, one tool call at a time, after any optional example.",
       parameters: objectSchema({ name: { type: "string" } }, ["name"]),
     },
   },
@@ -162,7 +163,8 @@ export const customWidgetAssistantEvaluationToolDefinitions: ToolDefinition[] =
     type: "function",
     function: {
       name: "customWidget_getExample",
-      description: "Get one installed Custom JSX example by catalog ID.",
+      description:
+        "Optionally get one installed Custom JSX example by catalog ID before requesting any component documentation.",
       parameters: objectSchema({ name: { type: "string" } }, ["name"]),
     },
   },
@@ -251,11 +253,24 @@ export function executeAssistantEvaluationTool(
   name: string,
   input: Record<string, unknown>,
 ): unknown {
+  const componentDocumentCount = state.calledTools.filter(
+    (toolName) => toolName === "customWidget_getComponent",
+  ).length;
+  if (name === "customWidget_getExample") {
+    if (componentDocumentCount > 0) {
+      return {
+        error:
+          "An example must be loaded before component documentation. Keep the existing component-document budget and continue to customWidget_validate.",
+      };
+    }
+    const example = typeof input.name === "string" ? getCustomWidgetExample(input.name) : null;
+    if (!example) return { error: "Custom JSX example not found" };
+    state.calledTools.push(name);
+    const { template, ...widget } = example.widget;
+    return { ...example, widget: { ...widget, templateLines: template.split("\n") } };
+  }
   const componentDocumentLimit = state.calledTools.includes("customWidget_getExample") ? 4 : 8;
-  if (
-    name === "customWidget_getComponent" &&
-    state.calledTools.filter((toolName) => toolName === "customWidget_getComponent").length >= componentDocumentLimit
-  ) {
+  if (name === "customWidget_getComponent" && componentDocumentCount >= componentDocumentLimit) {
     return {
       error:
         "The targeted component-document budget is exhausted. Continue with validation using the documentation already loaded.",
@@ -277,12 +292,6 @@ export function executeAssistantEvaluationTool(
       ? getCustomWidgetSharedProps(names)
       : { error: "At least one shared prop name is required" };
   }
-  if (name === "customWidget_getExample") {
-    const example = typeof input.name === "string" ? getCustomWidgetExample(input.name) : null;
-    if (!example) return { error: "Custom JSX example not found" };
-    const { template, ...widget } = example.widget;
-    return { ...example, widget: { ...widget, templateLines: template.split("\n") } };
-  }
   if (name === "customWidget_validate") {
     const parsed = parseDefinition(input.widget);
     if (!parsed.success) return { valid: false, issues: parsed.issues };
@@ -576,11 +585,11 @@ async function runAssistantAttempt(args: {
       process.stdout.write(`    tool: ${toolCall.function.name}\n`);
       let output: unknown;
       try {
-        const parsed = JSON.parse(toolCall.function.arguments) as unknown;
-        const input =
-          parsed !== null && typeof parsed === "object" && !Array.isArray(parsed)
-            ? (parsed as Record<string, unknown>)
-            : {};
+        const parsed: unknown = JSON.parse(toolCall.function.arguments);
+        let input: Record<string, unknown> = {};
+        if (parsed !== null && typeof parsed === "object" && !Array.isArray(parsed)) {
+          input = Object.fromEntries(Object.entries(parsed));
+        }
         output = executeAssistantEvaluationTool(args.testCase, state, toolCall.function.name, input);
       } catch (error) {
         output = { error: error instanceof Error ? error.message : "Tool input was not valid JSON" };

packages/custom-widgets/scripts/ai-evaluation.ts

@@ -219,7 +219,7 @@ export function buildRepairPrompt(
 export function buildJudgePrompt(testCase: CustomWidgetAiEvaluationCase, widget: HomarrCustomWidgetV2): string {
   return `You are a hostile-but-fair product review panel evaluating a safe dashboard widget. Most competent drafts should score 55-75, not 90. Judge only evidence present in the manifest and JSX. Never reward unsupported capabilities, invented API routes, aspirational claims, or code that merely validates.
 
-The installed Homarr skill and runtime references below are authoritative. In particular, request state is exposed as status.<requestId> with loading/ok/status/error fields while successful payloads are exposed as data.<requestId>. RefreshButton is an installed runtime helper that manually refreshes load queries. Every component in this already-validated template exists in the installed release. Do not penalize those documented facts. The widget has already passed Homarr's real schema and JSX analyzer, which proves syntax and component compatibility but does not prove API correctness, visual quality, usefulness, or accessibility.
+The installed Homarr skill and runtime references below are authoritative. The verified API notes and representative response fixtures are authoritative for endpoint paths, authentication requirements, and response shapes unless the validated manifest contradicts them. Do not invent external endpoint or authentication objections from outside assumptions. Decorative icons paired with equivalent adjacent visible status text need no separate aria-label. A Badge containing explicit visible status text is not color-only. A readable absolute UTC date and time is valid; do not require relative or localized time without a documented safe helper. In particular, request state is exposed as status.<requestId> with loading/ok/status/error fields while successful payloads are exposed as data.<requestId>. RefreshButton is an installed runtime helper that manually refreshes load queries. Every component in this already-validated template exists in the installed release. Do not penalize those documented facts. The widget has already passed Homarr's real schema and JSX analyzer, which proves syntax and component compatibility but does not prove API correctness, visual quality, usefulness, or accessibility.
 
 Installed Homarr Custom Widget authoring contract:
 ${getCustomWidgetSkillContent()}
@@ -254,16 +254,14 @@ Required review behavior:
 - Give a concrete evidence sentence for every category. List all score-capping issues under fatalProblems.
 - Return an empty fatalProblems array when there are no fatal problems; never put "none" or an explanation of their absence in that array.
 
-Homarr computes the weighted total and final verdict itself. Your total and verdict are advisory, but must be internally honest. A pass requires a weighted total of at least 90, every category at least 75, goalFulfillment/visualQuality/dailyUsefulness at least 85, complexityDiscipline at least 80, no fatal problem, and dailyUseDecision="would-use-daily". Return only the requested structured object.`;
+Homarr computes the weighted total and final verdict itself. Your total and verdict are advisory, but must be internally honest. A pass requires a weighted total of at least 85, every category at least 75, goalFulfillment at least 85, complexityDiscipline at least 80, no fatal problem, and dailyUseDecision="would-use-daily". Return only the requested structured object.`;
 }
 
 export function judgePasses(result: CustomWidgetJudgeResult): boolean {
   return (
-    result.total >= 90 &&
+    result.total >= 85 &&
     Object.values(result.categories).every((score) => score >= 75) &&
     result.categories.goalFulfillment >= 85 &&
-    result.categories.visualQuality >= 85 &&
-    result.categories.dailyUsefulness >= 85 &&
     result.categories.complexityDiscipline >= 80 &&
     result.dailyUseDecision === "would-use-daily" &&
     result.fatalProblems.length === 0

packages/custom-widgets/scripts/evaluate-ai-authoring.ts

@@ -15,58 +15,55 @@ if (!apiKey) {
 const requestedCase = process.argv.find((value) => value.startsWith("--case="))?.slice("--case=".length);
 const assistantMode = process.argv.includes("--assistant");
 const configuredLoops = Number(process.env.CUSTOM_WIDGET_AI_MAX_LOOPS ?? 10);
-const totalLoopBudget = Number.isInteger(configuredLoops) && configuredLoops > 0 ? Math.min(configuredLoops, 10) : 10;
-const selectableCases = assistantMode
-  ? CUSTOM_WIDGET_AI_EVALUATION_CASES.filter(
-      (testCase) =>
-        (testCase.sampleResponse !== undefined || testCase.previewResponses?.length) &&
-        testCase.expectations !== undefined,
-    )
-  : CUSTOM_WIDGET_AI_EVALUATION_CASES;
-const selectedCases = requestedCase
-  ? CUSTOM_WIDGET_AI_EVALUATION_CASES.filter((testCase) => testCase.id === requestedCase)
-  : selectableCases;
-if (selectedCases.length === 0) throw new Error(`Unknown AI evaluation case '${requestedCase}'`);
-if (totalLoopBudget < selectedCases.length) {
-  throw new Error(`CUSTOM_WIDGET_AI_MAX_LOOPS must be at least ${selectedCases.length} for this run`);
+const maxLoops = Number.isInteger(configuredLoops) && configuredLoops > 0 ? Math.min(configuredLoops, 10) : 10;
+let selectableCases = CUSTOM_WIDGET_AI_EVALUATION_CASES;
+if (assistantMode) {
+  selectableCases = CUSTOM_WIDGET_AI_EVALUATION_CASES.filter(
+    (testCase) =>
+      (testCase.sampleResponse !== undefined || testCase.previewResponses?.length) &&
+      testCase.expectations !== undefined,
+  );
 }
+let selectedCases = selectableCases;
+if (requestedCase)
+  selectedCases = CUSTOM_WIDGET_AI_EVALUATION_CASES.filter((testCase) => testCase.id === requestedCase);
+if (selectedCases.length === 0) throw new Error(`Unknown AI evaluation case '${requestedCase}'`);
 
 const runId = new Date().toISOString().replaceAll(/[:.]/gu, "-");
 const outputRoot = path.resolve(process.cwd(), ".ai-evaluations", runId);
 await mkdir(outputRoot, { recursive: true });
 
 const results: Array<AiEvaluationResult | CustomWidgetAssistantEvaluationResult> = [];
-let remainingLoopBudget = totalLoopBudget;
 for (const testCase of selectedCases) {
-  const remainingCases = selectedCases.length - results.length;
-  const maxLoops = Math.max(1, Math.floor(remainingLoopBudget / remainingCases));
   process.stdout.write(`Evaluating ${testCase.id}...\n`);
-  const result = assistantMode
-    ? await evaluateCustomWidgetAssistantCase({
-        testCase,
-        apiKey,
-        baseUrl: providerBaseUrl,
-        outputRoot,
-        maxLoops,
-        generatorModel,
-        judgeModel,
-      })
-    : await evaluateCustomWidgetCase({
-        testCase,
-        apiKey,
-        baseUrl: providerBaseUrl,
-        outputRoot,
-        maxLoops,
-        generatorModel,
-        judgeModel,
-      });
+  let result: AiEvaluationResult | CustomWidgetAssistantEvaluationResult;
+  if (assistantMode) {
+    result = await evaluateCustomWidgetAssistantCase({
+      testCase,
+      apiKey,
+      baseUrl: providerBaseUrl,
+      outputRoot,
+      maxLoops,
+      generatorModel,
+      judgeModel,
+    });
+  } else {
+    result = await evaluateCustomWidgetCase({
+      testCase,
+      apiKey,
+      baseUrl: providerBaseUrl,
+      outputRoot,
+      maxLoops,
+      generatorModel,
+      judgeModel,
+    });
+  }
   results.push(result);
-  remainingLoopBudget -= result.attempts;
-  process.stdout.write(
-    result.judge
-      ? `  ${result.judge.total}/100 after ${result.attempts} attempt(s)\n`
-      : `  failed after ${result.attempts} attempt(s)\n`,
-  );
+  if (result.judge) {
+    process.stdout.write(`  ${result.judge.total}/100 after ${result.attempts} attempt(s)\n`);
+  } else {
+    process.stdout.write(`  failed after ${result.attempts} attempt(s)\n`);
+  }
 }
 
 const summary = {

packages/custom-widgets/src/core/ai-prompt.ts

@@ -1,5 +1,11 @@
 import type { HomarrCustomWidgetV2 } from "./custom-jsx-schema";
 import { customJsxAuthoringCatalog } from "./component-catalog";
+import {
+  redactCustomWidgetAiContext,
+  redactCustomWidgetAiResponse,
+  redactCustomWidgetAiText,
+  redactCustomWidgetAiUrl,
+} from "./definition-security";
 import { customJsxExamples } from "./examples";
 
 export interface CustomWidgetAiDraft {
@@ -106,19 +112,28 @@ export const CUSTOM_WIDGET_AUTHORING_PROMPT = AUTHORING_PROMPT;
 
 export const CUSTOM_WIDGET_MCP_AUTHORING_PROMPT = `Author one Homarr Custom JSX v2 widget at a time.
 
-First call customWidget_getSkill. Its single lazy response contains the complete installed SKILL.md and every bundled reference; follow all of it. Read the live schema and compact component catalog. Fetch only the named component and example resources needed for the design. Component resources contain component-specific props; if shared props are needed, collect their catalog names and fetch those docs together once with customWidget_getSharedProps. The live resources are authoritative for this Homarr release.
+Mandatory sequence — follow it exactly:
+1. First call customWidget_getSkill and follow the complete installed skill and bundled references.
+2. Call customWidget_schema, then customWidget_getComponentCatalog.
+3. If a complete example is useful, call customWidget_getExample before any customWidget_getComponent call. Then choose at most four component names total and never request a fifth. Fetch only those named docs and, once, any named shared props with customWidget_getSharedProps.
+4. Before validation, quickly reject IIFEs, new, block callbacks or => {, and raw HTML tags such as span; use catalog components instead. Immediately call customWidget_validate with the complete definition and repair every issue.
+5. Call customWidget_previewCreate with that exact validated definition.
+6. Call customWidget_previewQuery once for every query returned by the preview and inspect every response.
+7. Call customWidget_createFromPreview with the exact fully tested preview session ID.
+
+Make one documentation tool call at a time. Never issue batched customWidget_getComponent calls, speculate about components, or walk the catalog. If preview evidence requires a definition change, restart at step 4 with the revised complete definition. The live resources are authoritative for this Homarr release.
 
 Keep the documentation phase bounded. Fetch at most eight named component documents before the first validation, or at most four after loading a complete example. Fetch another component only to resolve a concrete validation issue or missing interaction; never walk the catalog component-by-component. Move promptly from the smallest sufficient documentation set to a complete customWidget_validate call.
 
-Follow this lifecycle without skipping steps: construct one credential-free definition; call customWidget_validate; repair every reported issue; call customWidget_previewCreate; call customWidget_previewQuery once for every query returned by that preview (supplying required manual parameters); inspect the real response shape and HTTP status; adjust paths and template bindings; validate and preview again after changes; open or provide the returned previewPath for visual review; only then call customWidget_createFromPreview with the exact final tested preview session ID. Use customWidget_create only when no reusable preview session exists. Prefer templateLines over a JSON-escaped template string for multiline JSX in validate and preview tool inputs.
+Construct one credential-free definition. Supply required manual parameters to preview queries, inspect the real response shape and HTTP status, and open or provide the returned previewPath for visual review. Use customWidget_create only when no reusable preview session exists. Prefer templateLines over a JSON-escaped template string for multiline JSX in validate and preview tool inputs.
 
 Do not claim a query works when it has not returned a successful preview response. Configure deployment-specific source URLs and request credentials through Homarr when needed; never repeat plaintext. Use simulated preview actions unless the user explicitly enables live preview actions.
 
-Treat preview data as the binding contract: render every core requested field, guard optional arrays and nested values before indexing, and do not silently drop returned items. Show concise freshness context when the response supplies a timestamp. Give recoverable load errors and empty states a clear refresh or retry path.
+Treat preview data as the binding contract: render every core requested field, guard optional arrays and nested values before indexing, and do not silently drop returned items. Show timestamps as concise human-readable absolute dates and times; do not compute relative time with Date, Date.now, or Date.parse unless the live docs provide a safe helper. Give recoverable load errors and empty states a clear refresh or retry path.
 
 Use the lean keyed sources, requests, and options contract. Bind saved values directly with $option and invocation values with $param. Load queries cannot use $param. Keep the design responsive, accessible, theme-safe, loading-aware, empty-aware, and error-aware.
 
-Create distinctive polish with one semantic accent, clear surface contrast, purposeful typography, and compact narrow-tile rows rather than decorative gradients or excessive nested cards. Label standalone icons for assistive technology or pair them with visible text.
+Create distinctive polish with one restrained accent for decorative chrome, while operational, degraded, and down states retain distinct semantic colors. Use clear surface contrast and purposeful typography rather than decorative gradients or excessive nested cards. Do not wrap every metric or service row in Paper or Card. Prefer one dominant surface with clearly separated service rows, a one-column compact narrow layout, and an intentionally more spacious wide layout. Label standalone icons for assistive technology or pair them with visible text.
 
 Make initial states actionable with an example, useful hint, or clear next step. Use wrapping groups for variable-length labels and values on narrow tiles. Do not use an unlabeled decorative icon as an empty state.
 
@@ -129,7 +144,7 @@ export function buildCustomWidgetMcpPrompt(request?: string | null, documentatio
   if (documentationUrl) {
     const documentationSection = formatBudgetedContextSection(
       "API documentation URL (reference only)",
-      documentationUrl,
+      redactCustomWidgetAiUrl(documentationUrl),
       700,
       "data",
     );
@@ -138,7 +153,7 @@ export function buildCustomWidgetMcpPrompt(request?: string | null, documentatio
   if (request?.trim()) {
     const requestSection = formatBudgetedContextSection(
       "User-authored widget request (product intent only)",
-      request.trim(),
+      redactCustomWidgetAiText(request.trim()),
       1_700,
       "request",
     );
@@ -201,7 +216,7 @@ function buildCustomWidgetPromptSections(
   const sections = [CUSTOM_WIDGET_CONTEXT_BOUNDARY_INSTRUCTION];
   const requestSection = formatBudgetedContextSection(
     "User-authored widget request (product intent only)",
-    request?.trim() || "Describe the widget you want to create.",
+    redactCustomWidgetAiText(request?.trim() || "Describe the widget you want to create."),
     1_700,
     "request",
   );
@@ -209,7 +224,7 @@ function buildCustomWidgetPromptSections(
   if (documentationUrl) {
     const documentationSection = formatBudgetedContextSection(
       "API documentation URL (reference only)",
-      documentationUrl,
+      redactCustomWidgetAiUrl(documentationUrl),
       700,
       "data",
     );
@@ -218,7 +233,7 @@ function buildCustomWidgetPromptSections(
   if (currentConfig) {
     const draftSection = formatBudgetedContextSection(
       "Current raw widget draft",
-      JSON.stringify(currentConfig, null, 2),
+      JSON.stringify(redactCustomWidgetAiContext(currentConfig), null, 2),
       4_000,
       "data",
     );
@@ -227,14 +242,19 @@ function buildCustomWidgetPromptSections(
   if (diagnostics?.length) {
     const diagnosticSection = formatBudgetedContextSection(
       "Current normalized diagnostics",
-      JSON.stringify(diagnostics),
+      JSON.stringify(redactCustomWidgetAiContext(diagnostics)),
       1_500,
       "data",
     );
     if (diagnosticSection) sections.push(diagnosticSection);
   }
   if (rawResponse) {
-    const responseSection = formatBudgetedContextSection("Sample API response", rawResponse, 1_000, "data");
+    const responseSection = formatBudgetedContextSection(
+      "Sample API response",
+      redactCustomWidgetAiResponse(rawResponse),
+      1_000,
+      "data",
+    );
     if (responseSection) sections.push(responseSection);
   }
   const authoringBudget = CUSTOM_WIDGET_AI_PROMPT_LIMIT - sections.join("\n\n").length - footerLength - 8;

packages/custom-widgets/src/core/definition-security.ts

@@ -191,6 +191,70 @@ export function redactCustomWidgetCredentialLiterals(value: string): string {
     );
 }
 
+export function redactCustomWidgetAiContext(value: unknown, path: DefinitionPath = []): unknown {
+  if (Array.isArray(value)) return value.map((entry, index) => redactCustomWidgetAiContext(entry, [...path, index]));
+  if (value !== null && typeof value === "object") {
+    return Object.fromEntries(
+      Object.entries(value).map(([key, entry]) => {
+        const entryPath = [...path, key];
+        const risk = getCustomWidgetCredentialKeyRisk(key);
+        if (risk === "strong" && !isSchemaAuthenticationControl(entryPath)) {
+          if (!isHarmlessCustomWidgetCredentialSetting(entry)) return [key, "[REDACTED]"];
+        }
+        if (risk === "ambiguous" && typeof entry === "string") {
+          if (redactCustomWidgetCredentialLiterals(entry) !== entry) return [key, "[REDACTED]"];
+        }
+        return [key, redactCustomWidgetAiContext(entry, entryPath)];
+      }),
+    );
+  }
+  if (typeof value !== "string") return value;
+
+  const lastPathPart = path.at(-1);
+  let key = "";
+  if (typeof lastPathPart === "string") key = lastPathPart;
+  if (["sources", "requests", "options"].includes(key)) {
+    try {
+      const parsed: unknown = JSON.parse(value);
+      return redactCustomWidgetAiContext(parsed, path);
+    } catch {
+      /* Invalid editor JSON is still useful prompt context. */
+    }
+  }
+  if (key === "baseUrl" || key === "iconUrl") {
+    return redactCustomWidgetCredentialLiterals(redactCustomWidgetAiUrl(value));
+  }
+  return redactCustomWidgetAiText(value);
+}
+
+export function redactCustomWidgetAiResponse(value: string): string {
+  try {
+    const parsed: unknown = JSON.parse(value);
+    return JSON.stringify(redactCustomWidgetAiContext(parsed), null, 2);
+  } catch {
+    return redactCustomWidgetAiText(value);
+  }
+}
+
+export function redactCustomWidgetAiUrl(value: string): string {
+  try {
+    const url = new URL(value);
+    url.username = "";
+    url.password = "";
+    url.search = "";
+    url.hash = "";
+    return url.toString();
+  } catch {
+    return "[REDACTED_URL]";
+  }
+}
+
+export function redactCustomWidgetAiText(value: string): string {
+  return redactCustomWidgetCredentialLiterals(
+    value.replace(/\bhttps?:\/\/[^\s"'<>]+/giu, (candidate) => redactCustomWidgetAiUrl(candidate)),
+  );
+}
+
 function isSchemaAuthenticationControl(path: DefinitionPath): boolean {
   return (
     path.length === 3 &&

packages/custom-widgets/src/core/examples.ts

@@ -36,8 +36,8 @@ export const customJsxExamples: readonly CustomJsxExample[] = [
       requests: { overview: { path: "/api/overview", cacheSeconds: 30 } },
       options: {},
       template: `<Stack gap="md" p="md">
-  <Group justify="space-between"><Stack gap={2}><Text fw={700}>System overview</Text><Text size="xs" c="dimmed">Live service health</Text></Stack><RefreshButton /></Group>
-  {status.overview?.loading ? <Skeleton height={120} radius="lg" /> : status.overview?.error ? <Alert color="red" title="Could not load status">{status.overview.error}</Alert> : <SimpleGrid cols={{ base: 1, xs: 3 }}>{(data.overview?.services ?? []).map(service => <Paper key={service.name} withBorder p="sm" radius="md"><Group justify="space-between"><Text size="sm" fw={600}>{service.name}</Text><Badge color={service.online ? "green" : "red"} variant="light">{service.online ? "Online" : "Offline"}</Badge></Group><Text mt="xs" size="xl" fw={700}>{service.latency ?? "—"}<Text span size="xs" c="dimmed"> ms</Text></Text></Paper>)}</SimpleGrid>}
+  <Group justify="space-between" wrap="wrap"><Stack gap={2}><Text fw={700}>System overview</Text><Text size="xs" c="dimmed">Live service health</Text></Stack><RefreshButton label="Refresh service health" /></Group>
+  {status.overview?.loading ? <Skeleton height={180} radius="lg" /> : status.overview?.error ? <Alert color="red" title="Could not load status">Service health is unavailable. Use Refresh service health to try again. {status.overview.error}</Alert> : <Paper withBorder p={{ base: "sm", sm: "md" }} radius="lg"><Stack gap="md"><SimpleGrid cols={{ base: 1, sm: 2, md: 4 }} spacing={{ base: "xs", sm: "lg" }}><Stack gap={2}><Text size="xs" c="dimmed">Services</Text><Text fw={700}>{data.overview?.services?.length ?? 0}</Text></Stack><Stack gap={2}><Text size="xs" c="dimmed">Online</Text><Text fw={700} c="green">{(data.overview?.services ?? []).filter(service => service.online).length}</Text></Stack><Stack gap={2}><Text size="xs" c="dimmed">Offline</Text><Text fw={700} c="red">{(data.overview?.services ?? []).filter(service => !service.online).length}</Text></Stack><Stack gap={2}><Text size="xs" c="dimmed">Last checked</Text><Text size="sm">{(data.overview?.checkedAt ?? data.overview?.lastCheckedAt) ? Date.toLocaleDateString(data.overview.checkedAt ?? data.overview.lastCheckedAt, "en-US") + " " + Date.toLocaleTimeString(data.overview.checkedAt ?? data.overview.lastCheckedAt, "en-US") : "Not reported"}</Text></Stack></SimpleGrid><Divider />{(data.overview?.services?.length ?? 0) > 0 ? <Stack gap={0}>{data.overview.services.map(service => <Group key={service.name} justify="space-between" wrap="wrap" py={{ base: "xs", sm: "sm" }}><Stack gap={2}><Text size="sm" fw={600}>{service.name}</Text><Text size="xs" c="dimmed">{service.latency != null ? service.latency + " ms" : "Latency unavailable"}</Text></Stack><Badge color={service.online ? "green" : "red"} variant="light">{service.online ? "Online" : "Offline"}</Badge></Group>)}</Stack> : <Alert color="gray" title="No services">No services were returned. Use Refresh service health to try again.</Alert>}</Stack></Paper>}
 </Stack>`,
     },
   },

packages/custom-widgets/src/jsx/analyzer-jsx.ts

@@ -1,3 +1,4 @@
+import { customJsxCatalogComponentByName } from "../core/component-catalog";
 import {
   customJsxBindableComponentNames,
   customJsxComponentByName,
@@ -31,6 +32,20 @@ export function analyzeCustomJsxElement(
   const name = nameNode ? customJsxTagName(nameNode) : null;
   const resolvedName = name ? resolveCustomJsxComponentName(name) : null;
   const descriptor = resolvedName ? customJsxComponentByName.get(resolvedName) : undefined;
+  const attributes = nodesOf(opening.attributes);
+  if (resolvedName === "TablerIcon") {
+    const attributeNames = new Set<string>();
+    for (const attribute of attributes) {
+      if (attribute.type !== "JSXAttribute") continue;
+      const attributeName = nodeOf(attribute.name);
+      if (attributeName?.type === "JSXIdentifier") attributeNames.add(String(attributeName.name));
+    }
+    for (const prop of customJsxCatalogComponentByName.get(resolvedName)?.props ?? []) {
+      if (prop.required && !attributeNames.has(prop.name)) {
+        context.add(opening, `MISSING_REQUIRED_PROP: '${prop.name}' on ${name} is required`);
+      }
+    }
+  }
   if (descriptor?.safety === "denied") {
     context.add(
       opening,
@@ -46,7 +61,7 @@ export function analyzeCustomJsxElement(
     );
   }
 
-  for (const attribute of nodesOf(opening.attributes)) {
+  for (const attribute of attributes) {
     if (attribute.type === "JSXSpreadAttribute") {
       const argument = nodeOf(attribute.argument);
       if (argument) context.visit(argument, depth + 1, bindings);
@@ -71,7 +86,11 @@ export function analyzeCustomJsxElement(
       attributeName !== "bind" &&
       !customJsxSupportedPropsByName.get(resolvedName)?.has(attributeName)
     ) {
-      context.add(attribute, `UNKNOWN_MANTINE_PROP: '${attributeName}' on ${name} will be passed through`, "warning");
+      if (resolvedName === "TablerIcon") {
+        context.add(attribute, `UNKNOWN_COMPONENT_PROP: '${attributeName}' on ${name} is not supported`);
+      } else {
+        context.add(attribute, `UNKNOWN_MANTINE_PROP: '${attributeName}' on ${name} will be passed through`, "warning");
+      }
     } else if (resolvedName && attributeName === "bind" && !customJsxBindableComponentNames.has(resolvedName)) {
       context.add(attribute, `BINDING_UNAVAILABLE: '${name}' does not have a declarative binding adapter`, "warning");
     }

packages/custom-widgets/src/jsx/runtime-components.tsx

@@ -18,7 +18,7 @@ import { Collapsible, PaginatedList, StatBar, TabPanel, TabsContainer, TypeBadge
 import { getScopedCustomJsxControlName, isSafeCustomJsxUrl } from "./runtime-component-policy";
 import { sanitizeCustomJsxProps } from "./safe-properties";
 
-type Namespace = Readonly<Record<string, unknown>>;
+type Namespace = object;
 
 function resolveExport(namespace: Namespace, name: string): ComponentType<never> | undefined {
   let value: unknown = namespace;
@@ -26,7 +26,7 @@ function resolveExport(namespace: Namespace, name: string): ComponentType<never>
     if (!value || (typeof value !== "object" && typeof value !== "function") || !Object.hasOwn(value, segment)) {
       return undefined;
     }
-    value = (value as Record<string, unknown>)[segment];
+    value = Reflect.get(value, segment);
   }
   return typeof value === "function" || (typeof value === "object" && value !== null)
     ? (value as ComponentType<never>)
@@ -233,9 +233,9 @@ export interface CustomJsxComponentAdapters {
 
 export function createCustomJsxComponents(adapters: CustomJsxComponentAdapters): Record<string, ComponentType<never>> {
   const namespaces: Record<string, Namespace> = {
-    "@mantine/core": Core as unknown as Namespace,
-    "@mantine/charts": Charts as unknown as Namespace,
-    "@mantine/dates": Dates as unknown as Namespace,
+    "@mantine/core": Core,
+    "@mantine/charts": Charts,
+    "@mantine/dates": Dates,
   };
   const components: Record<string, ComponentType<never>> = {};
   for (const descriptor of enabledCustomJsxComponents) {
@@ -253,7 +253,7 @@ export function createCustomJsxComponents(adapters: CustomJsxComponentAdapters):
       components[descriptor.name] = wrap(descriptor.name, component, additions);
     }
   }
-  const core = Core as unknown as Namespace;
+  const core: Namespace = Core;
   for (const name of [
     "Notification",
     "LoadingOverlay",

packages/custom-widgets/src/test/ai-assistant-evaluation.spec.ts

@@ -146,6 +146,24 @@ describe("Custom Widget assistant live evaluation harness", () => {
     });
   });
 
+  it("rejects a late example without reducing the existing component-document budget", () => {
+    const testCase = CUSTOM_WIDGET_AI_EVALUATION_CASES.find(({ id }) => id === "fake-service-health");
+    if (!testCase) throw new Error("Fake service-health case is missing");
+    const state = createAssistantEvaluationState();
+
+    for (let index = 0; index < 8; index += 1) {
+      executeAssistantEvaluationTool(testCase, state, "customWidget_getComponent", { name: "Text" });
+    }
+    expect(
+      executeAssistantEvaluationTool(testCase, state, "customWidget_getExample", { name: "service-dashboard" }),
+    ).toEqual({
+      error:
+        "An example must be loaded before component documentation. Keep the existing component-document budget and continue to customWidget_validate.",
+    });
+    expect(state.calledTools).not.toContain("customWidget_getExample");
+    expect(state.calledTools.filter((name) => name === "customWidget_getComponent")).toHaveLength(8);
+  });
+
   it("enforces explicit multi-iteration preview requests before persistence", () => {
     const baseCase = CUSTOM_WIDGET_AI_EVALUATION_CASES.find(({ id }) => id === "fake-service-health");
     if (!baseCase) throw new Error("Fake service-health case is missing");

packages/custom-widgets/src/test/ai-evaluation.spec.ts

@@ -226,14 +226,38 @@ describe("AI authoring evaluation", () => {
     });
   });
 
-  it("requires exceptional goal, design, practicality, and complexity scores", () => {
-    expect(judgePasses(makeJudgeResult(90))).toBe(true);
+  it("requires an excellent result without weak categories", () => {
+    expect(
+      judgePasses({
+        ...makeJudgeResult(85),
+        categories: {
+          ...makeCategories(85),
+          visualQuality: 76,
+          dailyUsefulness: 76,
+          complexityDiscipline: 80,
+        },
+      }),
+    ).toBe(true);
+    expect(judgePasses(makeJudgeResult(84))).toBe(false);
+    expect(
+      judgePasses({
+        ...makeJudgeResult(90),
+        categories: { ...makeCategories(90), accessibility: 74 },
+      }),
+    ).toBe(false);
     expect(
       judgePasses({
         ...makeJudgeResult(92),
-        categories: { ...makeCategories(92), dailyUsefulness: 84 },
+        categories: { ...makeCategories(92), goalFulfillment: 84 },
       }),
     ).toBe(false);
+    expect(
+      judgePasses({
+        ...makeJudgeResult(92),
+        categories: { ...makeCategories(92), complexityDiscipline: 79 },
+      }),
+    ).toBe(false);
+    expect(judgePasses({ ...makeJudgeResult(92), dailyUseDecision: "promising-but-not-daily" })).toBe(false);
     expect(judgePasses({ ...makeJudgeResult(95), fatalProblems: ["A requested core action is missing."] })).toBe(false);
   });
 
@@ -257,6 +281,20 @@ describe("AI authoring evaluation", () => {
     expect(prompt).toContain(JSON.stringify(testCase.sampleResponse, null, 2));
     expect(prompt).toContain("# Bundled file: references/runtime.md");
     expect(prompt).toContain("does not prove API correctness, visual quality, usefulness, or accessibility");
+    expect(prompt).toContain(
+      "verified API notes and representative response fixtures are authoritative for endpoint paths, authentication requirements, and response shapes",
+    );
+    expect(prompt).toContain("Do not invent external endpoint or authentication objections from outside assumptions");
+    expect(prompt).toContain(
+      "Decorative icons paired with equivalent adjacent visible status text need no separate aria-label",
+    );
+    expect(prompt).toContain("A Badge containing explicit visible status text is not color-only");
+    expect(prompt).toContain(
+      "A readable absolute UTC date and time is valid; do not require relative or localized time without a documented safe helper",
+    );
+    expect(prompt).toContain(
+      'A pass requires a weighted total of at least 85, every category at least 75, goalFulfillment at least 85, complexityDiscipline at least 80, no fatal problem, and dailyUseDecision="would-use-daily"',
+    );
   });
 
   it("computes the weighted score and refuses inflated advisory verdicts", () => {

packages/custom-widgets/src/test/ai-prompt.spec.ts

@@ -53,11 +53,46 @@ describe("AI prompt", () => {
     expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("at most eight named component documents");
     expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("at most four after loading a complete example");
     expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("Treat preview data as the binding contract");
-    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("freshness context");
     expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("Make initial states actionable");
-    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("compact narrow-tile rows");
     expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("Label standalone icons");
     expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("Prefer templateLines");
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("Make one documentation tool call at a time");
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("Never issue batched customWidget_getComponent calls");
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain(
+      "choose at most four component names total and never request a fifth",
+    );
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain(
+      "reject IIFEs, new, block callbacks or => {, and raw HTML tags such as span",
+    );
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("concise human-readable absolute dates and times");
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain(
+      "do not compute relative time with Date, Date.now, or Date.parse unless the live docs provide a safe helper",
+    );
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("one restrained accent for decorative chrome");
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain(
+      "operational, degraded, and down states retain distinct semantic colors",
+    );
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("Do not wrap every metric or service row in Paper or Card");
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain("one dominant surface with clearly separated service rows");
+    expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain(
+      "a one-column compact narrow layout, and an intentionally more spacious wide layout",
+    );
+    const orderedSteps = [
+      "1. First call customWidget_getSkill",
+      "2. Call customWidget_schema",
+      "3. If a complete example is useful, call customWidget_getExample before any customWidget_getComponent call",
+      "4. Before validation",
+      "5. Call customWidget_previewCreate",
+      "6. Call customWidget_previewQuery once for every query",
+      "7. Call customWidget_createFromPreview",
+    ];
+    for (const [index, step] of orderedSteps.entries()) {
+      expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT).toContain(step);
+      if (index === 0) continue;
+      expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT.indexOf(orderedSteps[index - 1] ?? "")).toBeLessThan(
+        CUSTOM_WIDGET_MCP_AUTHORING_PROMPT.indexOf(step),
+      );
+    }
     expect(CUSTOM_WIDGET_MCP_AUTHORING_PROMPT.indexOf("customWidget_previewQuery")).toBeLessThan(
       CUSTOM_WIDGET_MCP_AUTHORING_PROMPT.indexOf("customWidget_create"),
     );
@@ -74,7 +109,7 @@ describe("AI prompt", () => {
 
   it("preserves raw context and the final instruction when optional context is large", () => {
     const prompt = buildCustomWidgetAiPrompt(undefined, JSON.stringify({ data: "x".repeat(20_000) }), null, "Build it");
-    expect(prompt).toContain('"data":"');
+    expect(prompt).toContain('"data": "');
     expect(prompt.endsWith(CUSTOM_WIDGET_FINAL_OUTPUT_INSTRUCTION)).toBe(true);
     expect(prompt.length).toBeLessThanOrEqual(12_000);
   });
@@ -102,4 +137,25 @@ describe("AI prompt", () => {
 
     expect(prompt).toContain("Use compact cards and show the current latency");
   });
+
+  it("redacts credentials from widget context", () => {
+    const prompt = buildCustomWidgetAiPrompt(
+      undefined,
+      JSON.stringify({ token: "sensitive" }),
+      {
+        requests: {
+          status: {
+            path: "/status?credential=Bearer-sk-secret-123456",
+            headers: { "X-Auth": "Bearer sk-secret-123456", "X-Feature-Key": "dashboard-layout" },
+          },
+        },
+      },
+      "Use Authorization: Bearer sk-secret-123456",
+    );
+
+    expect(prompt).not.toContain("sensitive");
+    expect(prompt).not.toContain("sk-secret-123456");
+    expect(prompt).toContain('"X-Auth": "[REDACTED]"');
+    expect(prompt).toContain('"X-Feature-Key": "dashboard-layout"');
+  });
 });

packages/custom-widgets/src/test/authoring-resources.spec.ts

@@ -10,6 +10,7 @@ import {
   getCustomWidgetSkill,
   getCustomWidgetSkillContent,
 } from "../core/authoring-resources";
+import { validateCustomJsxTemplate } from "../jsx/analyzer";
 
 const skillSourceDirectory = path.resolve(import.meta.dirname, "../../../../.agents/skills/homarr-custom-widget");
 
@@ -55,6 +56,11 @@ describe("authoring resources", () => {
         sources: { default: { baseUrl: "https://pokeapi.co" } },
       },
     });
+    const serviceDashboard = getCustomWidgetExample("service-dashboard");
+    if (!serviceDashboard) throw new Error("Service dashboard example is missing");
+    expect(validateCustomJsxTemplate(serviceDashboard.widget.template)).toEqual([]);
+    expect(serviceDashboard.widget.template.match(/<Paper/gu)).toHaveLength(1);
+    expect(serviceDashboard.widget.template).toContain('RefreshButton label="Refresh service health"');
     expect(Buffer.byteLength(JSON.stringify(pokedex), "utf8")).toBeLessThan(16_000);
     expect(catalog.sharedProps).toMatchObject({
       count: expect.any(Number),

packages/custom-widgets/src/test/jsx-policy.spec.ts

@@ -43,6 +43,21 @@ describe("shared Custom JSX policy", () => {
     expect(validateCustomJsxTemplate('<ThemeIcon><Icon name="brand-docker" /></ThemeIcon>')).toEqual([]);
   });
 
+  test("rejects TablerIcon without its required name prop", () => {
+    expect(validateCustomJsxTemplate('<TablerIcon icon="IconHeartbeat" />')).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          severity: "error",
+          message: expect.stringContaining("MISSING_REQUIRED_PROP: 'name' on TablerIcon"),
+        }),
+        expect.objectContaining({
+          severity: "error",
+          message: expect.stringContaining("UNKNOWN_COMPONENT_PROP: 'icon' on TablerIcon"),
+        }),
+      ]),
+    );
+  });
+
   test("accepts common bounded formatting and collection operations", () => {
     expect(
       validateCustomJsxTemplate(

packages/custom-widgets/src/workbench/code-editor.tsx

@@ -277,7 +277,8 @@ function EditorLoader() {
 function formatCode(value: string, language: "jsx" | "json" | "css") {
   if (language === "json") {
     try {
-      return `${JSON.stringify(JSON.parse(value) as unknown, null, 2)}\n`;
+      const parsed: unknown = JSON.parse(value);
+      return `${JSON.stringify(parsed, null, 2)}\n`;
     } catch {
       return value;
     }