fix(report): redesign evaluation card, add run duration, fix suite labeling

core/src/browser.ts

@@ -70,3 +70,12 @@ export {
 export type { LlmConfig, ProviderName } from "./config/types.js";
 
 export { getAdapter } from "./telemetry/adapter.js";
+
+export { renderReport } from "./report/render.js";
+export type {
+  ReportViewModel,
+  EvaluatorViewModel,
+  ResultViewModel,
+  TurnViewModel,
+  DetailCard,
+} from "./report/types.js";

core/src/execute/aggregate.ts

@@ -103,6 +103,8 @@ export interface ReportMeta {
   generatedAt: string;
   targetName: string;
   targetKind: "agent" | "mcp";
+  /** See {@link UnifiedRunReport.suiteId}. */
+  suiteId: string;
   effort: Effort;
   attackModel: string;
   judgeModel: string;
@@ -130,6 +132,7 @@ export function buildUnifiedReport(
     generatedAt: meta.generatedAt,
     targetName: meta.targetName,
     targetKind: meta.targetKind,
+    suiteId: meta.suiteId,
     effort: meta.effort,
     attackModel: meta.attackModel,
     judgeModel: meta.judgeModel,

core/src/execute/runAll.ts

@@ -51,6 +51,8 @@ export async function runAll(
   config: RunConfig,
   options?: RunAllOptions
 ): Promise<UnifiedRunReport> {
+  const runStartedAt = Date.now();
+
   // Fan every progress event to the legacy onProgress callback (backward-compat)
   // and to any registered lifecycle listeners.
   const listeners = options?.listeners ?? [];
@@ -137,12 +139,13 @@ export async function runAll(
     evaluatorResults.push(...loop.evaluatorResults);
     const stopReason = loop.stopReason;
 
-    // Build report (partial or complete) with stop reason and token usage.
+    // Build report (partial or complete) with stop reason, token usage, and duration.
     const report = buildReport(config, evaluatorResults);
     const usage = tokenTracker.totals;
     if (usage.totalTokens > 0) {
       report.summary.tokenUsage = usage;
     }
+    report.summary.durationMs = Date.now() - runStartedAt;
     if (stopReason) {
       (report as UnifiedRunReport & { stopReason?: string }).stopReason = stopReason;
     }
@@ -307,6 +310,16 @@ async function curateTracesIfConfigured(
   }
 }
 
+/**
+ * Label for the "Evaluation Suite" report field. Only a `mode: "suite"` selection
+ * runs a whole named suite verbatim — explicit evaluator lists and preloaded specs
+ * are, by definition, a hand-picked subset, so they're labeled "Custom Suite"
+ * rather than borrowing a suite name that would overstate what actually ran.
+ */
+function suiteLabel(selection: RunConfig["selection"]): string {
+  return selection.mode === "suite" ? selection.suite : "Custom Suite";
+}
+
 /** Assemble a {@link UnifiedRunReport} from Node-side run config and evaluator results. */
 function buildReport(config: RunConfig, evaluators: EvaluatorResult[]): UnifiedRunReport {
   const { attackModel, judgeModel } = modelLabel(config.attackerLlm, config.judgeLlm);
@@ -316,6 +329,7 @@ function buildReport(config: RunConfig, evaluators: EvaluatorResult[]): UnifiedR
       generatedAt: new Date().toISOString(),
       targetName: config.target.name,
       targetKind: config.target.kind,
+      suiteId: suiteLabel(config.selection),
       effort: config.effort,
       attackModel,
       judgeModel,

core/src/execute/runAllBrowser.ts

@@ -69,6 +69,7 @@ export async function runAllBrowser(
   agentTarget: AgentTarget,
   options?: BrowserRunAllOptions
 ): Promise<UnifiedRunReport> {
+  const runStartedAt = Date.now();
   const notify = options?.onProgress ?? (() => {});
   const attackModel = createModel(config.attackerLlm);
   const judgeModel = createModel(config.judgeLlm ?? config.attackerLlm);
@@ -237,6 +238,7 @@ export async function runAllBrowser(
   if (usage.totalTokens > 0) {
     report.summary.tokenUsage = usage;
   }
+  report.summary.durationMs = Date.now() - runStartedAt;
   return report;
 }
 
@@ -254,6 +256,10 @@ function buildBrowserReport(
         generatedAt: new Date().toISOString(),
         targetName: config.targetName ?? "target",
         targetKind: "agent",
+        // The browser path takes preloaded evaluator specs with no suite concept —
+        // the extension builds its own report/suite label independently of this
+        // return value (see popup.js's buildFinalReport).
+        suiteId: "Custom Suite",
         effort: config.effort,
         attackModel,
         judgeModel,

core/src/execute/types.ts

@@ -256,6 +256,12 @@ export interface UnifiedRunReport {
   generatedAt: string;
   targetName: string;
   targetKind: "agent" | "mcp";
+  /**
+   * The evaluation suite that was run, or "Custom Suite" when the selection
+   * doesn't correspond to one whole named suite (an explicit evaluator list,
+   * preloaded specs, or a hand-picked subset of a suite's evaluators).
+   */
+  suiteId: string;
   effort: Effort;
   attackModel: string;
   judgeModel: string;
@@ -267,6 +273,8 @@ export interface UnifiedRunReport {
     safetyScore: number;
     attackSuccessRate: number;
     tokenUsage?: { inputTokens: number; outputTokens: number; totalTokens: number };
+    /** Wall-clock time for the whole run, from the first evaluator to the last. */
+    durationMs?: number;
   };
   evaluators: EvaluatorResult[];
   /** Set when the run was stopped early due to a non-retryable LLM error. */

core/src/report/buildReport.ts

@@ -69,7 +69,7 @@ function toReportViewModel(report: UnifiedRunReport): ReportViewModel {
     generatedAt: report.generatedAt,
     generatorModel: report.attackModel,
     judgeModel: report.judgeModel,
-    target: { name: report.targetName },
+    target: { name: report.targetName, suiteId: report.suiteId },
     summary: report.summary,
     evaluators: report.evaluators.map(toEvaluatorViewModel),
     stopReason: report.stopReason,