@inspectr/mcplab 1.28.5 → 1.30.0

package/dist/app-server/evaluation-journal-projection.d.ts

@@ -0,0 +1,9 @@
+import type { ResultsJson } from '@inspectr/mcplab-core';
+export declare function projectEvaluationJournal(params: {
+    runsDir: string;
+    evaluationRunId: string;
+    evaluationName?: string;
+    runNote?: string;
+    executionStatus?: 'stopped';
+}): ResultsJson | null;
+//# sourceMappingURL=evaluation-journal-projection.d.ts.map

package/dist/app-server/evaluation-journal-projection.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-journal-projection.d.ts","sourceRoot":"","sources":["../../src/app-server/evaluation-journal-projection.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,uBAAuB,CAAC;AAKzD,wBAAgB,wBAAwB,CAAC,MAAM,EAAE;IAC/C,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,eAAe,CAAC,EAAE,SAAS,CAAC;CAC7B,GAAG,WAAW,GAAG,IAAI,CA2ErB"}

package/dist/app-server/evaluation-journal-projection.js

@@ -0,0 +1,76 @@
+import { join } from 'node:path';
+import { projectEvaluationResult } from './evaluation-result-projection.js';
+import { persistAppRunArtifacts } from './app-run-artifacts.js';
+import { readExecutionEvents } from './execution-journal.js';
+export function projectEvaluationJournal(params) {
+    const events = readExecutionEvents(join(params.runsDir, params.evaluationRunId));
+    const executions = events
+        .filter((event) => event.type === 'execution_completed' && event.results)
+        .reduce((results, event) => {
+        const executionId = event.executionId ?? event.results.metadata.run_id;
+        if (results.some((result) => result.metadata.run_id === executionId))
+            return results;
+        results.push(event.results);
+        return results;
+    }, []);
+    if (executions.length === 0)
+        return null;
+    const failedExecutions = new Set(events
+        .filter((event) => event.type === 'execution_failed')
+        .map((event) => event.executionId ?? event.eventId));
+    const stoppedExecutions = new Set(events
+        .filter((event) => event.type === 'execution_stopped' || event.type === 'evaluation_stopped')
+        .map((event) => event.executionId ?? event.eventId));
+    const executionTraceRecords = events
+        .filter((event) => event.type === 'execution_completed' && Array.isArray(event.traceRecords))
+        .flatMap((event) => event.traceRecords);
+    const roverTraceEvents = events.filter((event) => event.type === 'rover_event');
+    const traceRecords = [...executionTraceRecords, ...roverTraceEvents].sort((a, b) => {
+        const timestamp = (record) => typeof record.ts === 'string'
+            ? Date.parse(record.ts)
+            : typeof record.ts_start === 'string'
+                ? Date.parse(record.ts_start)
+                : 0;
+        return timestamp(a) - timestamp(b);
+    });
+    const results = projectEvaluationResult({
+        evaluationRunId: params.evaluationRunId,
+        runId: params.evaluationRunId,
+        evaluationName: params.evaluationName,
+        runNote: params.runNote,
+        executions,
+        failedExecutions: failedExecutions.size,
+        stoppedExecutions: stoppedExecutions.size
+    });
+    if (params.executionStatus || stoppedExecutions.size > 0)
+        results.metadata.execution_status = 'stopped';
+    const start = events.find((event) => event.type === 'evaluation_started');
+    if (typeof start?.newConversationBeforeStart === 'boolean')
+        results.metadata.rerun_new_conversation_before_start = start.newConversationBeforeStart;
+    if (typeof start?.newConversationBetweenScenarios === 'boolean')
+        results.metadata.rerun_new_conversation_between_scenarios =
+            start.newConversationBetweenScenarios;
+    persistAppRunArtifacts({
+        runDir: join(params.runsDir, params.evaluationRunId),
+        results,
+        resolvedConfig: {
+            evaluation_run_id: params.evaluationRunId,
+            evaluation_name: params.evaluationName
+        },
+        ...(traceRecords.length > 0
+            ? {
+                traceRecords: [
+                    {
+                        type: 'trace_meta',
+                        trace_version: 3,
+                        run_id: params.evaluationRunId,
+                        ts: new Date().toISOString()
+                    },
+                    ...traceRecords
+                ]
+            }
+            : {})
+    });
+    return results;
+}
+//# sourceMappingURL=evaluation-journal-projection.js.map

package/dist/app-server/evaluation-journal-projection.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-journal-projection.js","sourceRoot":"","sources":["../../src/app-server/evaluation-journal-projection.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,OAAO,EAAE,uBAAuB,EAAE,MAAM,mCAAmC,CAAC;AAC5E,OAAO,EAAE,sBAAsB,EAAE,MAAM,wBAAwB,CAAC;AAChE,OAAO,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAE7D,MAAM,UAAU,wBAAwB,CAAC,MAMxC;IACC,MAAM,MAAM,GAAG,mBAAmB,CAAC,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,eAAe,CAAC,CAAC,CAAC;IACjF,MAAM,UAAU,GAAG,MAAM;SACtB,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,qBAAqB,IAAI,KAAK,CAAC,OAAO,CAAC;SACxE,MAAM,CAAgB,CAAC,OAAO,EAAE,KAAK,EAAE,EAAE;QACxC,MAAM,WAAW,GAAG,KAAK,CAAC,WAAW,IAAK,KAAK,CAAC,OAAuB,CAAC,QAAQ,CAAC,MAAM,CAAC;QACxF,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,WAAW,CAAC;YAAE,OAAO,OAAO,CAAC;QACrF,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,OAAsB,CAAC,CAAC;QAC3C,OAAO,OAAO,CAAC;IACjB,CAAC,EAAE,EAAE,CAAC,CAAC;IACT,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACzC,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAC9B,MAAM;SACH,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,kBAAkB,CAAC;SACpD,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,OAAO,CAAC,CACtD,CAAC;IACF,MAAM,iBAAiB,GAAG,IAAI,GAAG,CAC/B,MAAM;SACH,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,mBAAmB,IAAI,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC;SAC5F,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,WAAW,IAAI,KAAK,CAAC,OAAO,CAAC,CACtD,CAAC;IACF,MAAM,qBAAqB,GAAG,MAAM;SACjC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,qBAAqB,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;SAC5F,OAAO,CACN,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,YAAwE,CAC1F,CAAC;IACJ,MAAM,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,aAAa,CAAC,CAAC;IAChF,MAAM,YAAY,GAAG,CAAC,GAAG,qBAAqB,EAAE,GAAG,gBAAgB,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACjF,MAAM,SAAS,GAAG,CAAC,MAA4C,EAAE,EAAE,CACjE,OAAO,MAAM,CAAC,EAAE,KAAK,QAAQ;YAC3B,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,CAAC;YACvB,CAAC,CAAC,OAAO,MAAM,CAAC,QAAQ,KAAK,QAAQ;gBACnC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC;gBAC7B,CAAC,CAAC,CAAC,CAAC;QACV,OAAO,SAAS,CAAC,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;IACrC,CAAC,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,uBAAuB,CAAC;QACtC,eAAe,EAAE,MAAM,CAAC,eAAe;QACvC,KAAK,EAAE,MAAM,CAAC,eAAe;QAC7B,cAAc,EAAE,MAAM,CAAC,cAAc;QACrC,OAAO,EAAE,MAAM,CAAC,OAAO;QACvB,UAAU;QACV,gBAAgB,EAAE,gBAAgB,CAAC,IAAI;QACvC,iBAAiB,EAAE,iBAAiB,CAAC,IAAI;KAC1C,CAAC,CAAC;IACH,IAAI,MAAM,CAAC,eAAe,IAAI,iBAAiB,CAAC,IAAI,GAAG,CAAC;QACtD,OAAO,CAAC,QAAQ,CAAC,gBAAgB,GAAG,SAAS,CAAC;IAChD,MAAM,KAAK,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,oBAAoB,CAAC,CAAC;IAC1E,IAAI,OAAO,KAAK,EAAE,0BAA0B,KAAK,SAAS;QACxD,OAAO,CAAC,QAAQ,CAAC,mCAAmC,GAAG,KAAK,CAAC,0BAA0B,CAAC;IAC1F,IAAI,OAAO,KAAK,EAAE,+BAA+B,KAAK,SAAS;QAC7D,OAAO,CAAC,QAAQ,CAAC,wCAAwC;YACvD,KAAK,CAAC,+BAA+B,CAAC;IAC1C,sBAAsB,CAAC;QACrB,MAAM,EAAE,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,eAAe,CAAC;QACpD,OAAO;QACP,cAAc,EAAE;YACd,iBAAiB,EAAE,MAAM,CAAC,eAAe;YACzC,eAAe,EAAE,MAAM,CAAC,cAAc;SACvC;QACD,GAAG,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC;YACzB,CAAC,CAAC;gBACE,YAAY,EAAE;oBACZ;wBACE,IAAI,EAAE,YAAqB;wBAC3B,aAAa,EAAE,CAAU;wBACzB,MAAM,EAAE,MAAM,CAAC,eAAe;wBAC9B,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;qBAC7B;oBACD,GAAG,YAAY;iBAChB;aACF;YACH,CAAC,CAAC,EAAE,CAAC;KACR,CAAC,CAAC;IACH,OAAO,OAAO,CAAC;AACjB,CAAC"}

package/dist/app-server/evaluation-journal-writer.d.ts

@@ -0,0 +1,23 @@
+import type { ResultsJson, ScenarioRunTraceRecord } from '@inspectr/mcplab-core';
+import { appendExecutionEvent } from './execution-journal.js';
+type JournalWriterOptions = {
+    runsDir: string;
+    evaluationRunId: string;
+    evaluationName?: string;
+    runNote?: string;
+    appendEvent?: typeof appendExecutionEvent;
+};
+export declare function recordEvaluationExecution(options: JournalWriterOptions & {
+    executionId: string;
+    results: ResultsJson;
+    traceRecords?: ScenarioRunTraceRecord[];
+    executionSource: 'mcplab' | 'rover';
+    eventId?: string;
+}): ResultsJson | null;
+export declare function recordEvaluationTerminalExecution(options: JournalWriterOptions & {
+    executionId: string;
+    status: 'error' | 'stopped';
+    reason: string;
+}): ResultsJson | null;
+export {};
+//# sourceMappingURL=evaluation-journal-writer.d.ts.map

package/dist/app-server/evaluation-journal-writer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-journal-writer.d.ts","sourceRoot":"","sources":["../../src/app-server/evaluation-journal-writer.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,WAAW,EAAE,sBAAsB,EAAE,MAAM,uBAAuB,CAAC;AACjF,OAAO,EAAE,oBAAoB,EAA8B,MAAM,wBAAwB,CAAC;AAG1F,KAAK,oBAAoB,GAAG;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,eAAe,EAAE,MAAM,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,WAAW,CAAC,EAAE,OAAO,oBAAoB,CAAC;CAC3C,CAAC;AAEF,wBAAgB,yBAAyB,CACvC,OAAO,EAAE,oBAAoB,GAAG;IAC9B,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,WAAW,CAAC;IACrB,YAAY,CAAC,EAAE,sBAAsB,EAAE,CAAC;IACxC,eAAe,EAAE,QAAQ,GAAG,OAAO,CAAC;IACpC,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,GACA,WAAW,GAAG,IAAI,CAqBpB;AAED,wBAAgB,iCAAiC,CAC/C,OAAO,EAAE,oBAAoB,GAAG;IAC9B,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,OAAO,GAAG,SAAS,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC;CAChB,GACA,WAAW,GAAG,IAAI,CAoBpB"}

package/dist/app-server/evaluation-journal-writer.js

@@ -0,0 +1,41 @@
+import { join } from 'node:path';
+import { appendExecutionEvent } from './execution-journal.js';
+import { projectEvaluationJournal } from './evaluation-journal-projection.js';
+export function recordEvaluationExecution(options) {
+    const event = {
+        eventId: options.eventId ?? `${options.executionSource}-result-${options.executionId}`,
+        type: 'execution_completed',
+        ts: new Date().toISOString(),
+        evaluationRunId: options.evaluationRunId,
+        executionId: options.executionId,
+        executionSource: options.executionSource,
+        results: options.results,
+        ...(options.traceRecords ? { traceRecords: options.traceRecords } : {})
+    };
+    (options.appendEvent ?? appendExecutionEvent)(join(options.runsDir, options.evaluationRunId), event);
+    return projectEvaluationJournal({
+        runsDir: options.runsDir,
+        evaluationRunId: options.evaluationRunId,
+        evaluationName: options.evaluationName,
+        runNote: options.runNote
+    });
+}
+export function recordEvaluationTerminalExecution(options) {
+    const event = {
+        eventId: `execution-${options.status}-${options.executionId}`,
+        type: options.status === 'error' ? 'execution_failed' : 'execution_stopped',
+        ts: new Date().toISOString(),
+        executionId: options.executionId,
+        evaluationRunId: options.evaluationRunId,
+        reason: options.reason
+    };
+    (options.appendEvent ?? appendExecutionEvent)(join(options.runsDir, options.evaluationRunId), event);
+    return projectEvaluationJournal({
+        runsDir: options.runsDir,
+        evaluationRunId: options.evaluationRunId,
+        evaluationName: options.evaluationName,
+        runNote: options.runNote,
+        ...(options.status === 'stopped' ? { executionStatus: 'stopped' } : {})
+    });
+}
+//# sourceMappingURL=evaluation-journal-writer.js.map

package/dist/app-server/evaluation-journal-writer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-journal-writer.js","sourceRoot":"","sources":["../../src/app-server/evaluation-journal-writer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAEjC,OAAO,EAAE,oBAAoB,EAA8B,MAAM,wBAAwB,CAAC;AAC1F,OAAO,EAAE,wBAAwB,EAAE,MAAM,oCAAoC,CAAC;AAU9E,MAAM,UAAU,yBAAyB,CACvC,OAMC;IAED,MAAM,KAAK,GAA0B;QACnC,OAAO,EAAE,OAAO,CAAC,OAAO,IAAI,GAAG,OAAO,CAAC,eAAe,WAAW,OAAO,CAAC,WAAW,EAAE;QACtF,IAAI,EAAE,qBAAqB;QAC3B,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QAC5B,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,GAAG,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,OAAO,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACxE,CAAC;IACF,CAAC,OAAO,CAAC,WAAW,IAAI,oBAAoB,CAAC,CAC3C,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,eAAe,CAAC,EAC9C,KAAK,CACN,CAAC;IACF,OAAO,wBAAwB,CAAC;QAC9B,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,cAAc,EAAE,OAAO,CAAC,cAAc;QACtC,OAAO,EAAE,OAAO,CAAC,OAAO;KACzB,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,iCAAiC,CAC/C,OAIC;IAED,MAAM,KAAK,GAA0B;QACnC,OAAO,EAAE,aAAa,OAAO,CAAC,MAAM,IAAI,OAAO,CAAC,WAAW,EAAE;QAC7D,IAAI,EAAE,OAAO,CAAC,MAAM,KAAK,OAAO,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,mBAAmB;QAC3E,EAAE,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QAC5B,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,MAAM,EAAE,OAAO,CAAC,MAAM;KACvB,CAAC;IACF,CAAC,OAAO,CAAC,WAAW,IAAI,oBAAoB,CAAC,CAC3C,IAAI,CAAC,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,eAAe,CAAC,EAC9C,KAAK,CACN,CAAC;IACF,OAAO,wBAAwB,CAAC;QAC9B,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,eAAe,EAAE,OAAO,CAAC,eAAe;QACxC,cAAc,EAAE,OAAO,CAAC,cAAc;QACtC,OAAO,EAAE,OAAO,CAAC,OAAO;QACxB,GAAG,CAAC,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,SAAkB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACjF,CAAC,CAAC;AACL,CAAC"}

package/dist/app-server/evaluation-result-projection.d.ts

@@ -0,0 +1,12 @@
+import type { ResultsJson } from '@inspectr/mcplab-core';
+/** Project completed execution snapshots into the single canonical evaluation result. */
+export declare function projectEvaluationResult(params: {
+    evaluationRunId: string;
+    runId: string;
+    executions: ResultsJson[];
+    evaluationName?: string;
+    runNote?: string;
+    failedExecutions?: number;
+    stoppedExecutions?: number;
+}): ResultsJson;
+//# sourceMappingURL=evaluation-result-projection.d.ts.map

package/dist/app-server/evaluation-result-projection.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-result-projection.d.ts","sourceRoot":"","sources":["../../src/app-server/evaluation-result-projection.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,WAAW,EAAc,MAAM,uBAAuB,CAAC;AAErE,yFAAyF;AACzF,wBAAgB,uBAAuB,CAAC,MAAM,EAAE;IAC9C,eAAe,EAAE,MAAM,CAAC;IACxB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,WAAW,EAAE,CAAC;IAC1B,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,iBAAiB,CAAC,EAAE,MAAM,CAAC;CAC5B,GAAG,WAAW,CAmId"}

package/dist/app-server/evaluation-result-projection.js

@@ -0,0 +1,107 @@
+import { calculatePassRate } from '@inspectr/mcplab-core';
+/** Project completed execution snapshots into the single canonical evaluation result. */
+export function projectEvaluationResult(params) {
+    const executions = params.executions;
+    const failedExecutions = params.failedExecutions ?? 0;
+    const stoppedExecutions = params.stoppedExecutions ?? 0;
+    const totalRuns = executions.reduce((sum, result) => sum + result.summary.total_runs, 0) +
+        failedExecutions +
+        stoppedExecutions;
+    const totalScenarios = executions.reduce((sum, result) => sum + result.summary.total_scenarios, 0);
+    const weighted = (field) => {
+        if (totalRuns === 0)
+            return null;
+        const values = executions.map((result) => result.summary[field]);
+        if (field === 'avg_tool_latency_ms' && values.some((value) => value === null))
+            return null;
+        return (executions.reduce((sum, result) => sum + (result.summary[field] ?? 0) * result.summary.total_runs, 0) / totalRuns);
+    };
+    const outcomes = {
+        passed: 0,
+        failed: failedExecutions,
+        incomplete: 0,
+        error: 0,
+        stopped: stoppedExecutions
+    };
+    for (const result of executions) {
+        for (const outcome of Object.keys(outcomes)) {
+            outcomes[outcome] = (outcomes[outcome] ?? 0) + (result.summary.outcomes?.[outcome] ?? 0);
+        }
+    }
+    if (outcomes.stopped === 0)
+        delete outcomes.stopped;
+    const sourceMetadata = executions[0]?.metadata;
+    const configPaths = Array.from(new Set(executions.map((result) => result.metadata.config_path).filter(Boolean)));
+    const rerunAgents = Array.from(new Set(executions.flatMap((result) => result.metadata.rerun_agents ?? [])));
+    const rerunScenarioIds = Array.from(new Set(executions.flatMap((result) => result.metadata.rerun_scenario_ids ?? [])));
+    // Children share evaluation overrides, but Rover snapshots may omit them and
+    // MCP snapshots contain only the overrides for their selected scenarios.
+    const serverOverrideAll = executions.find((result) => (result.metadata.rerun_server_override_all?.length ?? 0) > 0)?.metadata.rerun_server_override_all;
+    const scenarioServerOverrides = Object.fromEntries(executions.flatMap((result) => Object.entries(result.metadata.rerun_scenario_server_overrides ?? {})));
+    const executionSources = new Set(executions.map((result) => result.metadata.execution_source).filter(Boolean));
+    const executionClients = new Set(executions.map((result) => result.metadata.execution_client).filter(Boolean));
+    const aggregateNumericMetadata = (key) => {
+        const values = executions.map((result) => result.metadata[key]);
+        const numericValues = values.filter((value) => typeof value === 'number' && Number.isFinite(value));
+        if (values.length === 0 || numericValues.length !== values.length) {
+            return undefined;
+        }
+        return numericValues.reduce((sum, value) => sum + value, 0);
+    };
+    const mcpServerVersions = new Map();
+    for (const execution of executions) {
+        for (const [serverId, version] of Object.entries(execution.metadata.mcp_server_versions ?? {})) {
+            const previous = mcpServerVersions.get(serverId);
+            if (!mcpServerVersions.has(serverId))
+                mcpServerVersions.set(serverId, version);
+            else if (previous !== version)
+                mcpServerVersions.set(serverId, null);
+        }
+    }
+    const toolTokensTotal = aggregateNumericMetadata('tool_tokens_total');
+    const totalDurationMs = aggregateNumericMetadata('total_duration_ms');
+    const totalToolDurationMs = aggregateNumericMetadata('total_tool_duration_ms');
+    const executionClient = executionClients.size === 1
+        ? Array.from(executionClients)[0]
+        : executionClients.size > 1
+            ? 'mixed'
+            : undefined;
+    return {
+        metadata: {
+            run_id: params.runId,
+            timestamp: new Date().toISOString(),
+            config_hash: sourceMetadata?.config_hash ?? params.evaluationRunId,
+            config_path: configPaths.length === 1 ? configPaths[0] : undefined,
+            cli_version: sourceMetadata?.cli_version ?? 'unknown',
+            mcp_server_versions: Object.fromEntries(mcpServerVersions),
+            ...(toolTokensTotal !== undefined ? { tool_tokens_total: toolTokensTotal } : {}),
+            ...(totalDurationMs !== undefined ? { total_duration_ms: totalDurationMs } : {}),
+            ...(totalToolDurationMs !== undefined ? { total_tool_duration_ms: totalToolDurationMs } : {}),
+            ...(executionClient ? { execution_client: executionClient } : {}),
+            ...(executionSources.size === 1
+                ? { execution_source: executions[0]?.metadata.execution_source }
+                : {}),
+            config_name: params.evaluationName ||
+                sourceMetadata?.config_name ||
+                `Evaluation ${params.evaluationRunId}`,
+            ...(params.runNote?.trim() ? { run_note: params.runNote.trim() } : {}),
+            ...(rerunAgents.length > 0 ? { rerun_agents: rerunAgents } : {}),
+            ...(rerunScenarioIds.length > 0 ? { rerun_scenario_ids: rerunScenarioIds } : {}),
+            ...(serverOverrideAll ? { rerun_server_override_all: [...serverOverrideAll] } : {}),
+            ...(Object.keys(scenarioServerOverrides).length > 0
+                ? { rerun_scenario_server_overrides: scenarioServerOverrides }
+                : {}),
+            evaluation_run_id: params.runId
+        },
+        summary: {
+            total_scenarios: totalScenarios,
+            total_runs: totalRuns,
+            pass_rate: calculatePassRate(outcomes.passed ?? 0, totalRuns, outcomes.stopped),
+            avg_tool_calls_per_run: weighted('avg_tool_calls_per_run') ?? 0,
+            avg_tool_latency_ms: weighted('avg_tool_latency_ms'),
+            outcomes
+        },
+        scenarios: executions.flatMap((result) => result.scenarios)
+    };
+}
+//# sourceMappingURL=evaluation-result-projection.js.map

package/dist/app-server/evaluation-result-projection.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evaluation-result-projection.js","sourceRoot":"","sources":["../../src/app-server/evaluation-result-projection.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,uBAAuB,CAAC;AAG1D,yFAAyF;AACzF,MAAM,UAAU,uBAAuB,CAAC,MAQvC;IACC,MAAM,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC;IACrC,MAAM,gBAAgB,GAAG,MAAM,CAAC,gBAAgB,IAAI,CAAC,CAAC;IACtD,MAAM,iBAAiB,GAAG,MAAM,CAAC,iBAAiB,IAAI,CAAC,CAAC;IACxD,MAAM,SAAS,GACb,UAAU,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC;QACtE,gBAAgB;QAChB,iBAAiB,CAAC;IACpB,MAAM,cAAc,GAAG,UAAU,CAAC,MAAM,CACtC,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,MAAM,CAAC,OAAO,CAAC,eAAe,EACrD,CAAC,CACF,CAAC;IACF,MAAM,QAAQ,GAAG,CAAC,KAAuD,EAAiB,EAAE;QAC1F,IAAI,SAAS,KAAK,CAAC;YAAE,OAAO,IAAI,CAAC;QACjC,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC;QACjE,IAAI,KAAK,KAAK,qBAAqB,IAAI,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,IAAI,CAAC;YAAE,OAAO,IAAI,CAAC;QAC3F,OAAO,CACL,UAAU,CAAC,MAAM,CACf,CAAC,GAAG,EAAE,MAAM,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,UAAU,EAC/E,CAAC,CACF,GAAG,SAAS,CACd,CAAC;IACJ,CAAC,CAAC;IACF,MAAM,QAAQ,GAAwC;QACpD,MAAM,EAAE,CAAC;QACT,MAAM,EAAE,gBAAgB;QACxB,UAAU,EAAE,CAAC;QACb,KAAK,EAAE,CAAC;QACR,OAAO,EAAE,iBAAiB;KAC3B,CAAC;IACF,KAAK,MAAM,MAAM,IAAI,UAAU,EAAE,CAAC;QAChC,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAiB,EAAE,CAAC;YAC5D,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC;QAC3F,CAAC;IACH,CAAC;IACD,IAAI,QAAQ,CAAC,OAAO,KAAK,CAAC;QAAE,OAAO,QAAQ,CAAC,OAAO,CAAC;IACpD,MAAM,cAAc,GAAG,UAAU,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC;IAC/C,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAC5B,IAAI,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,WAAW,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CACjF,CAAC;IACF,MAAM,WAAW,GAAG,KAAK,CAAC,IAAI,CAC5B,IAAI,GAAG,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,YAAY,IAAI,EAAE,CAAC,CAAC,CAC5E,CAAC;IACF,MAAM,gBAAgB,GAAG,KAAK,CAAC,IAAI,CACjC,IAAI,GAAG,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,kBAAkB,IAAI,EAAE,CAAC,CAAC,CAClF,CAAC;IACF,6EAA6E;IAC7E,yEAAyE;IACzE,MAAM,iBAAiB,GAAG,UAAU,CAAC,IAAI,CACvC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,yBAAyB,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC,CACzE,EAAE,QAAQ,CAAC,yBAAyB,CAAC;IACtC,MAAM,uBAAuB,GAAG,MAAM,CAAC,WAAW,CAChD,UAAU,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAC5B,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,+BAA+B,IAAI,EAAE,CAAC,CACtE,CACF,CAAC;IACF,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAC9B,UAAU,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAC7E,CAAC;IACF,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAC9B,UAAU,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,gBAAgB,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAC7E,CAAC;IACF,MAAM,wBAAwB,GAAG,CAC/B,GAAyE,EACrD,EAAE;QACtB,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC;QAChE,MAAM,aAAa,GAAG,MAAM,CAAC,MAAM,CACjC,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAChF,CAAC;QACF,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,IAAI,aAAa,CAAC,MAAM,KAAK,MAAM,CAAC,MAAM,EAAE,CAAC;YAClE,OAAO,SAAS,CAAC;QACnB,CAAC;QACD,OAAO,aAAa,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,CAAC,GAAG,GAAG,KAAK,EAAE,CAAC,CAAC,CAAC;IAC9D,CAAC,CAAC;IACF,MAAM,iBAAiB,GAAG,IAAI,GAAG,EAAyB,CAAC;IAC3D,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACnC,KAAK,MAAM,CAAC,QAAQ,EAAE,OAAO,CAAC,IAAI,MAAM,CAAC,OAAO,CAC9C,SAAS,CAAC,QAAQ,CAAC,mBAAmB,IAAI,EAAE,CAC7C,EAAE,CAAC;YACF,MAAM,QAAQ,GAAG,iBAAiB,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;YACjD,IAAI,CAAC,iBAAiB,CAAC,GAAG,CAAC,QAAQ,CAAC;gBAAE,iBAAiB,CAAC,GAAG,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;iBAC1E,IAAI,QAAQ,KAAK,OAAO;gBAAE,iBAAiB,CAAC,GAAG,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QACvE,CAAC;IACH,CAAC;IACD,MAAM,eAAe,GAAG,wBAAwB,CAAC,mBAAmB,CAAC,CAAC;IACtE,MAAM,eAAe,GAAG,wBAAwB,CAAC,mBAAmB,CAAC,CAAC;IACtE,MAAM,mBAAmB,GAAG,wBAAwB,CAAC,wBAAwB,CAAC,CAAC;IAC/E,MAAM,eAAe,GACnB,gBAAgB,CAAC,IAAI,KAAK,CAAC;QACzB,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,CAAC,CAAC;QACjC,CAAC,CAAC,gBAAgB,CAAC,IAAI,GAAG,CAAC;YACzB,CAAC,CAAC,OAAO;YACT,CAAC,CAAC,SAAS,CAAC;IAClB,OAAO;QACL,QAAQ,EAAE;YACR,MAAM,EAAE,MAAM,CAAC,KAAK;YACpB,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,WAAW,EAAE,cAAc,EAAE,WAAW,IAAI,MAAM,CAAC,eAAe;YAClE,WAAW,EAAE,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS;YAClE,WAAW,EAAE,cAAc,EAAE,WAAW,IAAI,SAAS;YACrD,mBAAmB,EAAE,MAAM,CAAC,WAAW,CAAC,iBAAiB,CAAC;YAC1D,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAChF,GAAG,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAChF,GAAG,CAAC,mBAAmB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,sBAAsB,EAAE,mBAAmB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC7F,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACjE,GAAG,CAAC,gBAAgB,CAAC,IAAI,KAAK,CAAC;gBAC7B,CAAC,CAAC,EAAE,gBAAgB,EAAE,UAAU,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,gBAAgB,EAAE;gBAChE,CAAC,CAAC,EAAE,CAAC;YACP,WAAW,EACT,MAAM,CAAC,cAAc;gBACrB,cAAc,EAAE,WAAW;gBAC3B,cAAc,MAAM,CAAC,eAAe,EAAE;YACxC,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,MAAM,CAAC,OAAO,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACtE,GAAG,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAChE,GAAG,CAAC,gBAAgB,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAChF,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,yBAAyB,EAAE,CAAC,GAAG,iBAAiB,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACnF,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,uBAAuB,CAAC,CAAC,MAAM,GAAG,CAAC;gBACjD,CAAC,CAAC,EAAE,+BAA+B,EAAE,uBAAuB,EAAE;gBAC9D,CAAC,CAAC,EAAE,CAAC;YACP,iBAAiB,EAAE,MAAM,CAAC,KAAK;SAChC;QACD,OAAO,EAAE;YACP,eAAe,EAAE,cAAc;YAC/B,UAAU,EAAE,SAAS;YACrB,SAAS,EAAE,iBAAiB,CAAC,QAAQ,CAAC,MAAM,IAAI,CAAC,EAAE,SAAS,EAAE,QAAQ,CAAC,OAAO,CAAC;YAC/E,sBAAsB,EAAE,QAAQ,CAAC,wBAAwB,CAAC,IAAI,CAAC;YAC/D,mBAAmB,EAAE,QAAQ,CAAC,qBAAqB,CAAC;YACpD,QAAQ;SACT;QACD,SAAS,EAAE,UAAU,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC;KAC5D,CAAC;AACJ,CAAC"}

package/dist/app-server/execution-journal.d.ts

@@ -0,0 +1,13 @@
+export interface ExecutionJournalEvent {
+    eventId: string;
+    type: string;
+    ts: string;
+    executionId?: string;
+    [key: string]: unknown;
+}
+export declare function journalPath(runDir: string): string;
+export declare function appendExecutionEvent(runDir: string, event: ExecutionJournalEvent): void;
+export declare function readExecutionEvents(runDir: string): ExecutionJournalEvent[];
+export declare function readLatestResultSnapshot(runDir: string): ExecutionJournalEvent | null;
+export declare function writeAtomicText(path: string, content: string): void;
+//# sourceMappingURL=execution-journal.d.ts.map

package/dist/app-server/execution-journal.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"execution-journal.d.ts","sourceRoot":"","sources":["../../src/app-server/execution-journal.ts"],"names":[],"mappings":"AAUA,MAAM,WAAW,qBAAqB;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED,wBAAgB,WAAW,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAElD;AAED,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,qBAAqB,GAAG,IAAI,CAGvF;AAED,wBAAgB,mBAAmB,CAAC,MAAM,EAAE,MAAM,GAAG,qBAAqB,EAAE,CAiB3E;AAED,wBAAgB,wBAAwB,CAAC,MAAM,EAAE,MAAM,GAAG,qBAAqB,GAAG,IAAI,CAGrF;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,IAAI,CAInE"}

package/dist/app-server/execution-journal.js

@@ -0,0 +1,39 @@
+import { appendFileSync, existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from 'node:fs';
+import { join } from 'node:path';
+export function journalPath(runDir) {
+    return join(runDir, 'execution-events.jsonl');
+}
+export function appendExecutionEvent(runDir, event) {
+    mkdirSync(runDir, { recursive: true });
+    appendFileSync(journalPath(runDir), `${JSON.stringify(event)}\n`, 'utf8');
+}
+export function readExecutionEvents(runDir) {
+    const path = journalPath(runDir);
+    if (!existsSync(path))
+        return [];
+    return readFileSync(path, 'utf8')
+        .split('\n')
+        .map((line) => line.trim())
+        .filter(Boolean)
+        .flatMap((line) => {
+        try {
+            const value = JSON.parse(line);
+            return value && typeof value.eventId === 'string' && typeof value.type === 'string'
+                ? [value]
+                : [];
+        }
+        catch {
+            return [];
+        }
+    });
+}
+export function readLatestResultSnapshot(runDir) {
+    const snapshots = readExecutionEvents(runDir).filter((event) => event.type === 'result_snapshot');
+    return snapshots[snapshots.length - 1] ?? null;
+}
+export function writeAtomicText(path, content) {
+    const temporaryPath = `${path}.tmp-${process.pid}-${Date.now()}`;
+    writeFileSync(temporaryPath, content, 'utf8');
+    renameSync(temporaryPath, path);
+}
+//# sourceMappingURL=execution-journal.js.map

package/dist/app-server/execution-journal.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"execution-journal.js","sourceRoot":"","sources":["../../src/app-server/execution-journal.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,cAAc,EACd,UAAU,EACV,SAAS,EACT,YAAY,EACZ,UAAU,EACV,aAAa,EACd,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAUjC,MAAM,UAAU,WAAW,CAAC,MAAc;IACxC,OAAO,IAAI,CAAC,MAAM,EAAE,wBAAwB,CAAC,CAAC;AAChD,CAAC;AAED,MAAM,UAAU,oBAAoB,CAAC,MAAc,EAAE,KAA4B;IAC/E,SAAS,CAAC,MAAM,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACvC,cAAc,CAAC,WAAW,CAAC,MAAM,CAAC,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;AAC5E,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,MAAc;IAChD,MAAM,IAAI,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC;IACjC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,EAAE,CAAC;IACjC,OAAO,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC;SAC9B,KAAK,CAAC,IAAI,CAAC;SACX,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;SAC1B,MAAM,CAAC,OAAO,CAAC;SACf,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE;QAChB,IAAI,CAAC;YACH,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAA0B,CAAC;YACxD,OAAO,KAAK,IAAI,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,IAAI,OAAO,KAAK,CAAC,IAAI,KAAK,QAAQ;gBACjF,CAAC,CAAC,CAAC,KAAK,CAAC;gBACT,CAAC,CAAC,EAAE,CAAC;QACT,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC,CAAC,CAAC;AACP,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,MAAc;IACrD,MAAM,SAAS,GAAG,mBAAmB,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,KAAK,iBAAiB,CAAC,CAAC;IAClG,OAAO,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,IAAI,IAAI,CAAC;AACjD,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,IAAY,EAAE,OAAe;IAC3D,MAAM,aAAa,GAAG,GAAG,IAAI,QAAQ,OAAO,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;IACjE,aAAa,CAAC,aAAa,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IAC9C,UAAU,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC;AAClC,CAAC"}

package/dist/app-server/http.d.ts

@@ -3,4 +3,4 @@ export declare function asJson(res: ServerResponse, code: number, body: unknown)
 export declare function asHtml(res: ServerResponse, code: number, body: string): void;
 export declare function asText(res: ServerResponse, code: number, body: string): void;
 export declare function parseBody(req: IncomingMessage): Promise<any>;
-//# sourceMappingURL=http.d.ts.map
+//# sourceMappingURL=http.d.ts.map

package/dist/app-server/index.d.ts

@@ -1,3 +1,3 @@
 export type { AppServerOptions } from './types.js';
 export { startAppServer } from './router.js';
-//# sourceMappingURL=index.d.ts.map
+//# sourceMappingURL=index.d.ts.map

package/dist/app-server/inspectr-session-manager.d.ts

@@ -0,0 +1,45 @@
+import type { McpRequestHeaderContext } from '@inspectr/mcplab-core';
+export interface InspectrProxySession {
+  upstreamOrigin: string;
+  proxyOrigin: string;
+  dashboardUrl: string;
+}
+export declare function buildInspectrRequestHeaders(
+  context: McpRequestHeaderContext
+): Record<string, string>;
+interface InspectrChild {
+  kill(signal?: NodeJS.Signals): boolean;
+  once(event: string, listener: (...args: any[]) => void): this;
+}
+interface InspectrSessionManagerOptions {
+  storageRoot: string;
+  commandPath?: string;
+  log?: (message: string) => void;
+  allocatePort?: () => Promise<number>;
+  spawn?: (
+    command: string,
+    args: string[],
+    options: {
+      cwd: string;
+      stdio: ['ignore', 'ignore', 'ignore'];
+    }
+  ) => InspectrChild;
+  waitForReady?: (session: InspectrProxySession, child: InspectrChild) => Promise<void>;
+}
+export declare class InspectrSessionManager {
+  private readonly storageRoot;
+  private readonly commandPath;
+  private readonly log;
+  private readonly allocatePort;
+  private readonly spawn;
+  private readonly waitForReady;
+  private readonly sessions;
+  private readonly starting;
+  constructor(options: InspectrSessionManagerOptions);
+  acquire(serverUrl: string): Promise<InspectrProxySession>;
+  closeAll(): Promise<void>;
+  private start;
+}
+export declare function rewriteUrlThroughInspectr(serverUrl: string, proxyOrigin: string): string;
+export {};
+//# sourceMappingURL=inspectr-session-manager.d.ts.map

package/dist/app-server/inspectr-session-manager.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"inspectr-session-manager.d.ts","sourceRoot":"","sources":["../../src/app-server/inspectr-session-manager.ts"],"names":[],"mappings":"AAMA,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,uBAAuB,CAAC;AAErE,MAAM,WAAW,oBAAoB;IACnC,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;CACtB;AAED,wBAAgB,2BAA2B,CACzC,OAAO,EAAE,uBAAuB,GAC/B,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAYxB;AAED,UAAU,aAAa;IACrB,IAAI,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC,OAAO,GAAG,OAAO,CAAC;IACvC,IAAI,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC,GAAG,IAAI,EAAE,GAAG,EAAE,KAAK,IAAI,GAAG,IAAI,CAAC;CAC/D;AAED,UAAU,6BAA6B;IACrC,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;IAChC,YAAY,CAAC,EAAE,MAAM,OAAO,CAAC,MAAM,CAAC,CAAC;IACrC,KAAK,CAAC,EAAE,CACN,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,QAAQ,EAAE,QAAQ,CAAC,CAAA;KAAE,KAC5D,aAAa,CAAC;IACnB,YAAY,CAAC,EAAE,CAAC,OAAO,EAAE,oBAAoB,EAAE,KAAK,EAAE,aAAa,KAAK,OAAO,CAAC,IAAI,CAAC,CAAC;CACvF;AAID,qBAAa,sBAAsB;IACjC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAS;IACrC,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAS;IACrC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAA4B;IAChD,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAwB;IACrD,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAsD;IAC5E,OAAO,CAAC,QAAQ,CAAC,YAAY,CAA6D;IAC1F,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAqC;IAC9D,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA8C;gBAE3D,OAAO,EAAE,6BAA6B;IAS5C,OAAO,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,oBAAoB,CAAC;IAczD,QAAQ,IAAI,OAAO,CAAC,IAAI,CAAC;YAOjB,KAAK;CAiDpB;AA2BD,wBAAgB,yBAAyB,CAAC,SAAS,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,MAAM,CAGxF"}

package/dist/app-server/inspectr-session-manager.js

@@ -0,0 +1,203 @@
+import { createHash, randomBytes } from 'node:crypto';
+import { existsSync, mkdirSync, readFileSync } from 'node:fs';
+import { spawn as defaultSpawn } from 'node:child_process';
+import { createConnection, createServer } from 'node:net';
+import { dirname, join } from 'node:path';
+import { createRequire } from 'node:module';
+export function buildInspectrRequestHeaders(context) {
+    const tags = [
+        'mcplab',
+        `mcplab-run:${tagPart(context.runId)}`,
+        ...(context.scenarioId ? [`mcplab-scenario:${tagPart(context.scenarioId)}`] : []),
+        ...(context.agentName ? [`mcplab-agent:${tagPart(context.agentName)}`] : []),
+        `mcplab-server:${tagPart(context.serverName)}`
+    ];
+    return {
+        'inspectr-tag': tags.join(', '),
+        'inspectr-trace-id': context.requestId ?? `mcplab-run:${tagPart(context.runId)}`
+    };
+}
+export class InspectrSessionManager {
+    storageRoot;
+    commandPath;
+    log;
+    allocatePort;
+    spawn;
+    waitForReady;
+    sessions = new Map();
+    starting = new Map();
+    constructor(options) {
+        this.storageRoot = options.storageRoot;
+        this.commandPath = options.commandPath ?? resolveInspectrCommand();
+        this.log = options.log ?? console.log;
+        this.allocatePort = options.allocatePort ?? allocateFreePort;
+        this.spawn = options.spawn ?? ((command, args, spawnOptions) => defaultSpawn(command, args, spawnOptions));
+        this.waitForReady = options.waitForReady ?? waitForPorts;
+    }
+    async acquire(serverUrl) {
+        const upstreamOrigin = normalizeOrigin(serverUrl);
+        const existing = this.sessions.get(upstreamOrigin);
+        if (existing)
+            return existing;
+        const inFlight = this.starting.get(upstreamOrigin);
+        if (inFlight)
+            return inFlight;
+        const promise = this.start(upstreamOrigin).finally(() => {
+            this.starting.delete(upstreamOrigin);
+        });
+        this.starting.set(upstreamOrigin, promise);
+        return promise;
+    }
+    async closeAll() {
+        await Promise.allSettled(this.starting.values());
+        const sessions = [...this.sessions.values()];
+        this.sessions.clear();
+        await Promise.all(sessions.map((session) => stopChild(session.child)));
+    }
+    async start(upstreamOrigin) {
+        const proxyPort = await this.allocatePort();
+        const dashboardPort = await this.allocatePort();
+        const cwd = join(this.storageRoot, originKey(upstreamOrigin));
+        mkdirSync(cwd, { recursive: true });
+        const channel = `mcplab-${originKey(upstreamOrigin)}`;
+        const channelCode = randomBytes(16).toString('hex');
+        const dashboardUrl = new URL(`http://127.0.0.1:${dashboardPort}`);
+        dashboardUrl.searchParams.set('channel', channel);
+        dashboardUrl.searchParams.set('channelCode', channelCode);
+        const session = {
+            upstreamOrigin,
+            proxyOrigin: `http://127.0.0.1:${proxyPort}`,
+            dashboardUrl: dashboardUrl.toString()
+        };
+        const child = this.spawn(this.commandPath, [
+            `--listen=127.0.0.1:${proxyPort}`,
+            `--app-port=${dashboardPort}`,
+            `--backend=${upstreamOrigin}`,
+            `--channel=${channel}`,
+            `--channel-code=${channelCode}`,
+            '--print=false'
+        ], { cwd, stdio: ['ignore', 'ignore', 'ignore'] });
+        try {
+            await this.waitForReady(session, child);
+        }
+        catch (error) {
+            await stopChild(child);
+            throw new Error(`Inspectr failed to start for ${upstreamOrigin}: ${error instanceof Error ? error.message : String(error)}`);
+        }
+        this.log(`[mcplab-app] Inspectr started for ${session.upstreamOrigin}, proxy: ${session.proxyOrigin}, dashboard: ${session.dashboardUrl}`);
+        const managed = session;
+        Object.defineProperties(managed, {
+            child: { value: child, enumerable: false },
+            cwd: { value: cwd, enumerable: false }
+        });
+        this.sessions.set(upstreamOrigin, managed);
+        child.once('exit', () => {
+            if (this.sessions.get(upstreamOrigin) === managed)
+                this.sessions.delete(upstreamOrigin);
+        });
+        return managed;
+    }
+}
+function resolveInspectrCommand() {
+    const require = createRequire(import.meta.url);
+    const packageJsonPath = require.resolve('@inspectr/inspectr/package.json');
+    const manifest = JSON.parse(readFileSync(packageJsonPath, 'utf8'));
+    const bin = typeof manifest.bin === 'string' ? manifest.bin : manifest.bin?.inspectr;
+    if (!bin)
+        throw new Error('Installed @inspectr/inspectr package has no inspectr executable');
+    const packageBin = join(dirname(packageJsonPath), bin);
+    return existsSync(packageBin) ? packageBin : 'inspectr';
+}
+function normalizeOrigin(serverUrl) {
+    let parsed;
+    try {
+        parsed = new URL(serverUrl);
+    }
+    catch {
+        throw new Error(`Inspectr requires an absolute MCP server URL: ${serverUrl}`);
+    }
+    if (parsed.protocol !== 'http:' && parsed.protocol !== 'https:') {
+        throw new Error(`Inspectr supports HTTP MCP server URLs only: ${serverUrl}`);
+    }
+    return parsed.origin;
+}
+export function rewriteUrlThroughInspectr(serverUrl, proxyOrigin) {
+    const original = new URL(serverUrl);
+    return new URL(`${original.pathname}${original.search}`, proxyOrigin).toString();
+}
+function originKey(origin) {
+    return createHash('sha256').update(origin).digest('hex').slice(0, 16);
+}
+function tagPart(value) {
+    return value.trim().replace(/[\u0000-\u001f\u007f,]+/g, '_').slice(0, 120) || 'unknown';
+}
+async function allocateFreePort() {
+    const server = createServer();
+    await new Promise((resolve, reject) => {
+        server.once('error', reject);
+        server.listen(0, '127.0.0.1', () => resolve());
+    });
+    const address = server.address();
+    const port = typeof address === 'object' && address ? address.port : 0;
+    await new Promise((resolve) => server.close(() => resolve()));
+    if (!port)
+        throw new Error('Could not allocate a local Inspectr port');
+    return port;
+}
+async function waitForPorts(session, child) {
+    const ports = [new URL(session.proxyOrigin).port, new URL(session.dashboardUrl).port].map(Number);
+    let childExited = false;
+    const markExited = () => {
+        childExited = true;
+    };
+    child.once('error', markExited);
+    child.once('exit', markExited);
+    const deadline = Date.now() + 15_000;
+    let stableChecks = 0;
+    while (Date.now() < deadline) {
+        if (childExited)
+            throw new Error('Inspectr process exited during startup');
+        if (await Promise.all(ports.map(canConnect))) {
+            stableChecks += 1;
+            if (stableChecks >= 2 && (await hasDashboardConfig(session.dashboardUrl)))
+                return;
+        }
+        else {
+            stableChecks = 0;
+        }
+        await new Promise((resolve) => setTimeout(resolve, 50));
+    }
+    throw new Error('timed out waiting for Inspectr proxy and dashboard');
+}
+async function hasDashboardConfig(dashboardUrl) {
+    try {
+        const configUrl = new URL('/app/config', dashboardUrl);
+        const response = await fetch(configUrl, {
+            signal: AbortSignal.timeout(1_000)
+        });
+        if (!response.ok)
+            return false;
+        const config = (await response.json());
+        return Boolean(config.token && config.channel_code && config.sse_endpoint);
+    }
+    catch {
+        return false;
+    }
+}
+function canConnect(port) {
+    return new Promise((resolve) => {
+        const socket = createConnection({ host: '127.0.0.1', port });
+        socket.once('connect', () => {
+            socket.destroy();
+            resolve(true);
+        });
+        socket.once('error', () => {
+            socket.destroy();
+            resolve(false);
+        });
+    });
+}
+async function stopChild(child) {
+    child.kill('SIGTERM');
+}
+//# sourceMappingURL=inspectr-session-manager.js.map