@tonyclaw/agent-inspector 2.0.28 → 2.0.30

package/.output/server/_tanstack-start-manifest_v-C0fdSWnA.mjs

@@ -0,0 +1,4 @@
+const tsrStartManifest = () => ({ "routes": { "__root__": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/__root.tsx", "children": ["/", "/api/config", "/api/groups", "/api/health", "/api/logs", "/api/mcp", "/api/models", "/api/providers", "/api/runs", "/api/sessions", "/proxy/$", "/session/$sessionId", "/api/knowledge/candidates", "/api/knowledge/project-context", "/api/knowledge/search", "/api/knowledge/sessions/$sessionId/candidates"], "preloads": ["/assets/main-XHHZsOBK.js"], "assets": [] }, "/": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/index.tsx", "assets": [], "preloads": ["/assets/index-Cv5vPgpr.js", "/assets/ProxyViewerContainer-NtOD8Ex_.js"] }, "/api/config": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/config.ts", "children": ["/api/config/paths"] }, "/api/groups": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/groups.ts", "children": ["/api/groups/$groupId"] }, "/api/health": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/health.ts" }, "/api/logs": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/logs.ts", "children": ["/api/logs/$id", "/api/logs/stream"] }, "/api/mcp": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/mcp.ts" }, "/api/models": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/models.ts" }, "/api/providers": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.ts", "children": ["/api/providers/$providerId", "/api/providers/export", "/api/providers/import", "/api/providers/scan"] }, "/api/runs": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/runs.ts", "children": ["/api/runs/$runId"] }, "/api/sessions": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/sessions.ts" }, "/proxy/$": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/proxy/$.ts" }, "/session/$sessionId": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/session/$sessionId.tsx", "assets": [], "preloads": ["/assets/_sessionId-zimzVrJE.js", "/assets/ProxyViewerContainer-NtOD8Ex_.js"] }, "/api/config/paths": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/config.paths.ts" }, "/api/groups/$groupId": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/groups.$groupId.ts", "children": ["/api/groups/$groupId/evidence", "/api/groups/$groupId/sessions"] }, "/api/knowledge/candidates": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/knowledge.candidates.ts", "children": ["/api/knowledge/candidates/$candidateId"] }, "/api/knowledge/project-context": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/knowledge.project-context.ts" }, "/api/knowledge/search": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/knowledge.search.ts" }, "/api/logs/$id": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/logs.$id.ts", "children": ["/api/logs/$id/chunks", "/api/logs/$id/replay"] }, "/api/logs/stream": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/logs.stream.ts" }, "/api/providers/$providerId": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.$providerId.ts", "children": ["/api/providers/$providerId/model-metadata", "/api/providers/$providerId/test"] }, "/api/providers/export": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.export.ts" }, "/api/providers/import": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.import.ts" }, "/api/providers/scan": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.scan.ts" }, "/api/runs/$runId": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/runs.$runId.ts", "children": ["/api/runs/$runId/evidence"] }, "/api/groups/$groupId/evidence": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/groups.$groupId.evidence.ts" }, "/api/groups/$groupId/sessions": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/groups.$groupId.sessions.ts" }, "/api/knowledge/candidates/$candidateId": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/knowledge.candidates.$candidateId.ts", "children": ["/api/knowledge/candidates/$candidateId/promote"] }, "/api/logs/$id/chunks": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/logs.$id.chunks.ts" }, "/api/logs/$id/replay": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/logs.$id.replay.ts" }, "/api/providers/$providerId/model-metadata": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.$providerId.model-metadata.ts" }, "/api/providers/$providerId/test": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.$providerId.test.ts", "children": ["/api/providers/$providerId/test/log"] }, "/api/runs/$runId/evidence": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/runs.$runId.evidence.ts" }, "/api/knowledge/candidates/$candidateId/promote": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/knowledge.candidates.$candidateId.promote.ts" }, "/api/knowledge/sessions/$sessionId/candidates": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/knowledge.sessions.$sessionId.candidates.ts" }, "/api/providers/$providerId/test/log": { "filePath": "C:/Users/claw/workspace/agent-inspector/src/routes/api/providers.$providerId.test.log.ts" } }, "clientEntry": "/assets/main-XHHZsOBK.js" });
+export {
+  tsrStartManifest
+};

package/.output/server/index.mjs

@@ -35,103 +35,103 @@ const headers = ((m) => function headersRouteRule(event) {
   }
 });
 const assets = {
-  "/assets/index-B9_VaAWl.js": {
+  "/assets/index-Cv5vPgpr.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"74-wegwMejcMaCe2qDEma7y6qUh/6U"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "etag": '"74-qPu6sF/kGmsOicFgVQR8W+cHfLQ"',
+    "mtime": "2026-07-01T01:34:44.486Z",
     "size": 116,
-    "path": "../public/assets/index-B9_VaAWl.js"
+    "path": "../public/assets/index-Cv5vPgpr.js"
+  },
+  "/assets/index-CwlHPmgL.css": {
+    "type": "text/css; charset=utf-8",
+    "etag": '"1a3f0-E2f0kOdSBi1RhjoK38+CSij/apg"',
+    "mtime": "2026-07-01T01:34:44.486Z",
+    "size": 107504,
+    "path": "../public/assets/index-CwlHPmgL.css"
   },
   "/assets/alibaba-TTwafVwX.svg": {
     "type": "image/svg+xml",
     "etag": '"171b-6dyV5K8QjiaY35sN9qNprh9zDIs"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "mtime": "2026-07-01T01:34:44.485Z",
     "size": 5915,
     "path": "../public/assets/alibaba-TTwafVwX.svg"
   },
-  "/assets/CompareDrawer-CicCP3Hb.js": {
+  "/assets/CompareDrawer-BDzfRSt1.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"49fe-CmOMF3QSwAcX6NWjQdoZrIdyoIw"',
-    "mtime": "2026-06-30T03:21:19.570Z",
+    "etag": '"49fe-SgBlfrYSy5DyWozx9HIYT63pan8"',
+    "mtime": "2026-07-01T01:34:44.486Z",
     "size": 18942,
-    "path": "../public/assets/CompareDrawer-CicCP3Hb.js"
+    "path": "../public/assets/CompareDrawer-BDzfRSt1.js"
   },
-  "/assets/index-DsiKfWCp.css": {
-    "type": "text/css; charset=utf-8",
-    "etag": '"1a3d5-pnLJEAiyUUKlRw5OUnvwf1dCQeI"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 107477,
-    "path": "../public/assets/index-DsiKfWCp.css"
+  "/assets/main-XHHZsOBK.js": {
+    "type": "text/javascript; charset=utf-8",
+    "etag": '"51726-1yTScfInPZoti54sKap6zxOxpvY"',
+    "mtime": "2026-07-01T01:34:44.486Z",
+    "size": 333606,
+    "path": "../public/assets/main-XHHZsOBK.js"
+  },
+  "/assets/qwen-CONDcHqt.png": {
+    "type": "image/png",
+    "etag": '"572c3-cdJAPaHdOvFCGzuaQjagdgOu6XE"',
+    "mtime": "2026-07-01T01:34:44.485Z",
+    "size": 357059,
+    "path": "../public/assets/qwen-CONDcHqt.png"
   },
   "/assets/minimax-BPMzvuL-.jpeg": {
     "type": "image/jpeg",
     "etag": '"1b06-IwivU89ko5UTMUM1/t7hn4sQK9A"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "mtime": "2026-07-01T01:34:44.486Z",
     "size": 6918,
     "path": "../public/assets/minimax-BPMzvuL-.jpeg"
   },
-  "/assets/qwen-CONDcHqt.png": {
-    "type": "image/png",
-    "etag": '"572c3-cdJAPaHdOvFCGzuaQjagdgOu6XE"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 357059,
-    "path": "../public/assets/qwen-CONDcHqt.png"
+  "/assets/RequestAnatomy-G0dV8-D_.js": {
+    "type": "text/javascript; charset=utf-8",
+    "etag": '"4485-dLrPuUurtUkiFHTiud70zTzhUi8"',
+    "mtime": "2026-07-01T01:34:44.486Z",
+    "size": 17541,
+    "path": "../public/assets/RequestAnatomy-G0dV8-D_.js"
   },
-  "/assets/ReplayDialog-Bc8q3ujm.js": {
+  "/assets/ReplayDialog-p-5TKg75.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"318c-QS16LlsGhExlRZzuuDmYzqWLBN0"',
-    "mtime": "2026-06-30T03:21:19.570Z",
+    "etag": '"318c-2IdqYt4/fxajfqXLAVb8U+TSsfM"',
+    "mtime": "2026-07-01T01:34:44.486Z",
     "size": 12684,
-    "path": "../public/assets/ReplayDialog-Bc8q3ujm.js"
+    "path": "../public/assets/ReplayDialog-p-5TKg75.js"
+  },
+  "/assets/ResponseView-D2wE6lD-.js": {
+    "type": "text/javascript; charset=utf-8",
+    "etag": '"30d0-5TfShOzs9M0k3ghgcC/Uc1CZi5U"',
+    "mtime": "2026-07-01T01:34:44.486Z",
+    "size": 12496,
+    "path": "../public/assets/ResponseView-D2wE6lD-.js"
+  },
+  "/assets/StreamingChunkSequence-D8JRK-j_.js": {
+    "type": "text/javascript; charset=utf-8",
+    "etag": '"d57-jGvUqgjv7HbrCBsM6bKKdb9Ijro"',
+    "mtime": "2026-07-01T01:34:44.486Z",
+    "size": 3415,
+    "path": "../public/assets/StreamingChunkSequence-D8JRK-j_.js"
   },
   "/assets/zhipuai-BPNAnxo-.svg": {
     "type": "image/svg+xml",
     "etag": '"2bf8-hNaLCTi89nOFCsIIfWpP/jrfo0s"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "mtime": "2026-07-01T01:34:44.485Z",
     "size": 11256,
     "path": "../public/assets/zhipuai-BPNAnxo-.svg"
   },
-  "/assets/main-Brj0Gn91.js": {
-    "type": "text/javascript; charset=utf-8",
-    "etag": '"51726-mwq2VuKeQG//deWEjQSFZIL9ZWo"',
-    "mtime": "2026-06-30T03:21:19.568Z",
-    "size": 333606,
-    "path": "../public/assets/main-Brj0Gn91.js"
-  },
-  "/assets/RequestAnatomy-Coyy3zcH.js": {
+  "/assets/_sessionId-zimzVrJE.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"4485-KFNCPYVKcx+a4jW1uo7q7Az4T5c"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 17541,
-    "path": "../public/assets/RequestAnatomy-Coyy3zcH.js"
-  },
-  "/assets/_sessionId-9rEF0uSE.js": {
-    "type": "text/javascript; charset=utf-8",
-    "etag": '"d2-+KgPaqhF/ImO1Zb9tZi0sMB8OPI"',
-    "mtime": "2026-06-30T03:21:19.568Z",
+    "etag": '"d2-XtmPH9ZczbI8wqjiKa5PV3BIn0Y"',
+    "mtime": "2026-07-01T01:34:44.486Z",
     "size": 210,
-    "path": "../public/assets/_sessionId-9rEF0uSE.js"
-  },
-  "/assets/ResponseView-BWJzgh0c.js": {
-    "type": "text/javascript; charset=utf-8",
-    "etag": '"30d0-XQuX3LcRTXyoZYFeRDp9J2gMpd0"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 12496,
-    "path": "../public/assets/ResponseView-BWJzgh0c.js"
-  },
-  "/assets/StreamingChunkSequence-BSC3uuE5.js": {
-    "type": "text/javascript; charset=utf-8",
-    "etag": '"d57-1GnBSzcha+n1obLlcv4lwyBhPf4"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 3415,
-    "path": "../public/assets/StreamingChunkSequence-BSC3uuE5.js"
+    "path": "../public/assets/_sessionId-zimzVrJE.js"
   },
-  "/assets/ProxyViewerContainer-BccuA6p5.js": {
+  "/assets/ProxyViewerContainer-NtOD8Ex_.js": {
     "type": "text/javascript; charset=utf-8",
-    "etag": '"af111-eqyso0QJHFS7+IdUoP/1PcawtQE"',
-    "mtime": "2026-06-30T03:21:19.570Z",
-    "size": 717073,
-    "path": "../public/assets/ProxyViewerContainer-BccuA6p5.js"
+    "etag": '"af8bd-AtJl2HP2L1UcDbcZviRHY1WMV8U"',
+    "mtime": "2026-07-01T01:34:44.486Z",
+    "size": 719037,
+    "path": "../public/assets/ProxyViewerContainer-NtOD8Ex_.js"
   }
 };
 function readAsset(id) {

package/README.md

@@ -154,7 +154,7 @@ http://localhost:25947/api/mcp
 Coding agents can use the MCP surface in three ways:
 
 - Tools: perform actions such as listing logs, reading bounded log details, testing providers,
-  creating evaluation runs, and exporting evidence packs.
+  creating evaluation runs/groups, attaching sessions, and exporting evidence packs.
 - Resources: read stable evidence URIs without invoking a tool call.
 - Prompts: start common analysis/reporting tasks from reusable templates.
 
@@ -176,6 +176,24 @@ The evaluation-oriented run flow is:
 <dataDir>/evidence/<runId>/evidence.html
 ```
 
+For benchmark or CI systems that spawn multiple coding-agent processes, use a group as the
+outer evaluation artifact:
+
+1. Call `inspector_create_group` once for the evaluation batch.
+2. Start each OpenCode/Codex process through the proxy with its own model/provider.
+3. As each process discovers or reports its session id, call `inspector_add_group_session` with
+   `groupId`, `sessionId`, and optional `runId`, `provider`, `model`, `agent`, `label`, and
+   metadata.
+4. Call `inspector_update_group` to mark the whole batch `running`, `failed`, `completed`, or
+   `cancelled`.
+5. Call `inspector_export_group_evidence` to write an aggregate provider/model matrix under:
+
+```text
+<dataDir>/evidence/groups/<groupId>/evidence.json
+<dataDir>/evidence/groups/<groupId>/evidence.md
+<dataDir>/evidence/groups/<groupId>/evidence.html
+```
+
 Useful MCP resource URIs:
 
 | Resource | Purpose |
@@ -187,12 +205,17 @@ Useful MCP resource URIs:
 | `inspector://failures/recent` | Read recent failed/cancelled runs and non-success evidence classifications. |
 | `inspector://runs/{runId}` | Read one declared run and its evidence paths. |
 | `inspector://evidence/{runId}` | Read the exported Markdown evidence pack for a run. |
+| `inspector://groups` | List declared evaluation groups. |
+| `inspector://groups/{groupId}` | Read one group, including attached session/run members and evidence paths. |
+| `inspector://groups/{groupId}/sessions` | Read only the attached group member list. |
+| `inspector://groups/{groupId}/evidence` | Read the exported Markdown evidence pack for a group. |
 
 Built-in prompt templates help turn captured evidence into user-facing output:
 
 - `inspector_analyze_session_failure`: explain why a session failed and cite log evidence.
 - `inspector_triage_recent_failures`: prioritize recent failed runs and suggest next actions.
 - `inspector_generate_jenkins_report`: produce a Jenkins-ready Markdown report for a run.
+- `inspector_generate_group_report`: produce a provider/model evaluation report for a group.
 - `inspector_extract_repro_steps`: extract concise reproduction steps from a session.
 
 ## Jenkins And GitCode

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tonyclaw/agent-inspector",
-  "version": "2.0.28",
+  "version": "2.0.30",
   "type": "module",
   "description": "Agent observability and knowledge capture layer for AI coding tools.",
   "license": "MIT",

package/src/components/ProxyViewer.tsx

@@ -783,6 +783,9 @@ export function ProxyViewer({
               <Plus className="size-4 shrink-0 text-muted-foreground/70" aria-hidden="true" />
               <McpLogo className="size-10 shrink-0" />
             </span>
+            <span className="max-w-[calc(100vw-7rem)] text-sm font-semibold text-muted-foreground">
+              Local First. Evidence First.
+            </span>
             <McpReadyBadge />
           </h1>
           <div className="justify-self-end">

package/src/components/providers/ProviderCard.tsx

@@ -87,6 +87,16 @@ function hasHeaderDetails(headers: Record<string, string> | undefined): boolean
   return formatHeaders(headers) !== undefined;
 }
 
+function formatElapsed(ms: number): string {
+  if (ms < 1000) return `${ms}ms`;
+  return `${(ms / 1000).toFixed(1)}s`;
+}
+
+function formatTokenRate(tokensPerSecond: number): string {
+  if (tokensPerSecond >= 10) return `${tokensPerSecond.toFixed(0)}/s`;
+  return `${tokensPerSecond.toFixed(1)}/s`;
+}
+
 function hasDiagnosticDetails(result: TestResult): boolean {
   const debug = result.debug;
   return (
@@ -255,6 +265,7 @@ function TestStatus({ result }: { result: ProviderTestState }): JSX.Element {
   }
   if (result.success) {
     const tokenParts: ReactNode[] = [];
+    const timingParts: ReactNode[] = [];
     if (result.inputTokens !== undefined) {
       tokenParts.push(<span key="in">{result.inputTokens} in</span>);
     }
@@ -275,11 +286,29 @@ function TestStatus({ result }: { result: ProviderTestState }): JSX.Element {
         </span>,
       );
     }
+    if (result.firstChunkMs !== undefined && result.firstChunkMs !== null) {
+      timingParts.push(<span key="first">first {formatElapsed(result.firstChunkMs)}</span>);
+    }
+    if (result.totalStreamMs !== undefined && result.totalStreamMs !== null) {
+      timingParts.push(<span key="total">total {formatElapsed(result.totalStreamMs)}</span>);
+    }
+    if (
+      result.tokensPerSecond !== undefined &&
+      result.tokensPerSecond !== null &&
+      result.tokensPerSecond > 0
+    ) {
+      timingParts.push(<span key="rate">{formatTokenRate(result.tokensPerSecond)} tok/s</span>);
+    }
     const displayTokens: ReactNode[] = [];
     for (let i = 0; i < tokenParts.length; i++) {
       if (i > 0) displayTokens.push(", ");
       displayTokens.push(tokenParts[i]);
     }
+    const displayTiming: ReactNode[] = [];
+    for (let i = 0; i < timingParts.length; i++) {
+      if (i > 0) displayTiming.push(", ");
+      displayTiming.push(timingParts[i]);
+    }
     return (
       <TooltipProvider>
         <Tooltip>
@@ -290,6 +319,9 @@ function TestStatus({ result }: { result: ProviderTestState }): JSX.Element {
               {tokenParts.length > 0 && (
                 <span className="text-muted-foreground">({displayTokens})</span>
               )}
+              {timingParts.length > 0 && (
+                <span className="hidden text-muted-foreground lg:inline">({displayTiming})</span>
+              )}
             </div>
           </TooltipTrigger>
           <TooltipContent>Connection test passed</TooltipContent>

package/src/components/proxy-viewer/LogEntryHeader.tsx

@@ -11,6 +11,7 @@ import {
   Copy,
   FileDiff,
   FileTerminal,
+  Gauge,
   Globe,
   History,
   Loader2,
@@ -45,6 +46,11 @@ function formatElapsed(ms: number): string {
   return `${(ms / 1000).toFixed(1)}s`;
 }
 
+function formatTokenRate(tokensPerSecond: number): string {
+  if (tokensPerSecond >= 10) return `${tokensPerSecond.toFixed(0)}/s`;
+  return `${tokensPerSecond.toFixed(1)}/s`;
+}
+
 /**
  * Inline trend indicator: small arrow (green up / red down) plus the absolute
  * delta in compact form. Returns `null` when there is no trend to display.
@@ -148,6 +154,11 @@ export const LogEntryHeader = memo(function ({
     log.elapsedMs > slowResponseThresholdSeconds * 1000;
 
   const hasTokens = log.inputTokens !== null || log.outputTokens !== null;
+  const firstChunkMs = log.firstChunkMs ?? null;
+  const tokensPerSecond = log.tokensPerSecond ?? null;
+  const firstChunkLabel = firstChunkMs === null ? null : formatElapsed(firstChunkMs);
+  const tokenRateLabel =
+    tokensPerSecond === null || tokensPerSecond <= 0 ? null : formatTokenRate(tokensPerSecond);
 
   return (
     <TooltipProvider>
@@ -238,11 +249,37 @@ export const LogEntryHeader = memo(function ({
                 ? `Slow response: ${formatElapsed(log.elapsedMs)} exceeds ${formatElapsed(
                     slowResponseThresholdSeconds * 1000,
                   )}`
-                : "Elapsed response time"}
+                : log.streaming
+                  ? "Total streaming response time"
+                  : "Elapsed response time"}
             </TooltipContent>
           </Tooltip>
         )}
 
+        {/* Streaming timing */}
+        {log.streaming && firstChunkLabel !== null && (
+          <Tooltip>
+            <TooltipTrigger asChild>
+              <span className="flex items-center gap-1 text-xs text-cyan-400 shrink-0">
+                <Radio className="size-3" />
+                <span className="font-mono tabular-nums">first {firstChunkLabel}</span>
+              </span>
+            </TooltipTrigger>
+            <TooltipContent>Time to first streaming chunk</TooltipContent>
+          </Tooltip>
+        )}
+        {log.streaming && tokenRateLabel !== null && (
+          <Tooltip>
+            <TooltipTrigger asChild>
+              <span className="flex items-center gap-1 text-xs text-emerald-400 shrink-0">
+                <Gauge className="size-3" />
+                <span className="font-mono tabular-nums">{tokenRateLabel}</span>
+              </span>
+            </TooltipTrigger>
+            <TooltipContent>Output token throughput after first chunk</TooltipContent>
+          </Tooltip>
+        )}
+
         {/* Token counts */}
         {hasTokens && (
           <span className="flex items-center gap-1 text-xs shrink-0">

package/src/lib/groupContract.ts

@@ -0,0 +1,139 @@
+import { z } from "zod";
+import { JsonValueSchema } from "../proxy/schemas";
+import { InspectorRunSchema, InspectorRunStatusSchema, JenkinsReportSchema } from "./runContract";
+import { SessionInfoSchema, SessionTokenUsageSchema } from "./sessionInfoContract";
+
+export const InspectorGroupStatusSchema = z.enum([
+  "created",
+  "running",
+  "completed",
+  "failed",
+  "cancelled",
+]);
+
+export const InspectorGroupKindSchema = z.enum(["evaluation", "batch", "manual"]);
+
+export const InspectorGroupEvidenceSchema = z.object({
+  jsonPath: z.string(),
+  markdownPath: z.string(),
+  htmlPath: z.string(),
+  exportedAt: z.string(),
+});
+
+export const InspectorGroupMemberSchema = z.object({
+  id: z.string(),
+  groupId: z.string(),
+  sessionId: z.string(),
+  runId: z.string().nullable(),
+  label: z.string().nullable(),
+  agent: z.string().nullable(),
+  provider: z.string().nullable(),
+  model: z.string().nullable(),
+  status: InspectorRunStatusSchema.nullable(),
+  metadata: z.record(z.string(), JsonValueSchema),
+  addedAt: z.string(),
+  updatedAt: z.string(),
+});
+
+export const InspectorGroupSchema = z.object({
+  id: z.string(),
+  title: z.string(),
+  kind: InspectorGroupKindSchema,
+  task: z.string().nullable(),
+  project: z.string().nullable(),
+  status: InspectorGroupStatusSchema,
+  tags: z.array(z.string()),
+  metadata: z.record(z.string(), JsonValueSchema),
+  evidence: InspectorGroupEvidenceSchema.nullable(),
+  members: z.array(InspectorGroupMemberSchema),
+  createdAt: z.string(),
+  updatedAt: z.string(),
+});
+
+export const CreateInspectorGroupInputSchema = z
+  .object({
+    groupId: z.string().min(1).optional(),
+    title: z.string().min(1).optional(),
+    kind: InspectorGroupKindSchema.optional(),
+    task: z.string().nullable().optional(),
+    project: z.string().nullable().optional(),
+    status: InspectorGroupStatusSchema.optional(),
+    tags: z.array(z.string()).optional(),
+    metadata: z.record(z.string(), JsonValueSchema).optional(),
+  })
+  .optional();
+
+export const UpdateInspectorGroupInputSchema = z.object({
+  title: z.string().min(1).optional(),
+  kind: InspectorGroupKindSchema.optional(),
+  task: z.string().nullable().optional(),
+  project: z.string().nullable().optional(),
+  status: InspectorGroupStatusSchema.optional(),
+  tags: z.array(z.string()).optional(),
+  metadata: z.record(z.string(), JsonValueSchema).optional(),
+});
+
+export const AddInspectorGroupSessionInputSchema = z.object({
+  memberId: z.string().min(1).optional(),
+  sessionId: z.string().min(1),
+  runId: z.string().min(1).nullable().optional(),
+  label: z.string().min(1).nullable().optional(),
+  agent: z.string().min(1).nullable().optional(),
+  provider: z.string().min(1).nullable().optional(),
+  model: z.string().min(1).nullable().optional(),
+  status: InspectorRunStatusSchema.nullable().optional(),
+  metadata: z.record(z.string(), JsonValueSchema).optional(),
+});
+
+export const GroupEvidenceExportOptionsSchema = z
+  .object({
+    includeHistory: z.boolean().optional(),
+    latestLogLimit: z.number().int().positive().max(50).optional(),
+  })
+  .optional();
+
+export const InspectorGroupExportMemberSchema = z.object({
+  member: InspectorGroupMemberSchema,
+  run: InspectorRunSchema.nullable(),
+  session: SessionInfoSchema.nullable(),
+});
+
+export const InspectorGroupEvidenceSummarySchema = z.object({
+  memberCount: z.number().int().nonnegative(),
+  sessionsWithData: z.number().int().nonnegative(),
+  missingSessions: z.number().int().nonnegative(),
+  requestCount: z.number().int().nonnegative(),
+  errorCount: z.number().int().nonnegative(),
+  tokenUsage: SessionTokenUsageSchema,
+  models: z.array(z.string()),
+  providers: z.array(z.string()),
+  completedMembers: z.number().int().nonnegative(),
+  failedMembers: z.number().int().nonnegative(),
+  runningMembers: z.number().int().nonnegative(),
+});
+
+export const GroupEvidenceExportResultSchema = z.object({
+  group: InspectorGroupSchema,
+  evidence: InspectorGroupEvidenceSchema,
+  members: z.array(InspectorGroupExportMemberSchema),
+  summary: InspectorGroupEvidenceSummarySchema,
+  jenkinsReport: JenkinsReportSchema,
+});
+
+export const InspectorGroupsListResponseSchema = z.object({
+  groups: z.array(InspectorGroupSchema),
+});
+
+export type InspectorGroupStatus = z.infer<typeof InspectorGroupStatusSchema>;
+export type InspectorGroupKind = z.infer<typeof InspectorGroupKindSchema>;
+export type InspectorGroupEvidence = z.infer<typeof InspectorGroupEvidenceSchema>;
+export type InspectorGroupMember = z.infer<typeof InspectorGroupMemberSchema>;
+export type InspectorGroup = z.infer<typeof InspectorGroupSchema>;
+export type CreateInspectorGroupInput = z.infer<typeof CreateInspectorGroupInputSchema>;
+export type UpdateInspectorGroupInput = z.infer<typeof UpdateInspectorGroupInputSchema>;
+export type AddInspectorGroupSessionInput = z.infer<typeof AddInspectorGroupSessionInputSchema>;
+export type GroupEvidenceExportOptions = z.infer<typeof GroupEvidenceExportOptionsSchema>;
+export type InspectorGroupExportMember = z.infer<typeof InspectorGroupExportMemberSchema>;
+export type InspectorGroupEvidenceSummary = z.infer<typeof InspectorGroupEvidenceSummarySchema>;
+export type GroupEvidenceExportResult = z.infer<typeof GroupEvidenceExportResultSchema>;
+export type InspectorGroupsListResponse = z.infer<typeof InspectorGroupsListResponseSchema>;

package/src/lib/providerTestContract.ts

@@ -45,6 +45,9 @@ export const ProviderTestResultSchema = z.object({
   cacheCreationInputTokens: z.number().optional(),
   cacheReadInputTokens: z.number().optional(),
   latencyMs: z.number().optional(),
+  firstChunkMs: z.number().nullable().optional(),
+  totalStreamMs: z.number().nullable().optional(),
+  tokensPerSecond: z.number().nullable().optional(),
   content: z.array(ProviderTestContentSchema).optional(),
   rawResponse: z.string().optional(),
   streaming: z.boolean().optional(),

package/src/lib/sessionInfoContract.ts

@@ -21,6 +21,9 @@ export const SessionLogSummarySchema = z.object({
   hasError: z.boolean(),
   error: z.string().nullable(),
   latencyMs: z.number().nullable(),
+  firstChunkMs: z.number().nullable().optional(),
+  totalStreamMs: z.number().nullable().optional(),
+  tokensPerSecond: z.number().nullable().optional(),
   tokens: z.object({
     input: z.number().nullable(),
     output: z.number().nullable(),