@monoes/cli 1.0.0

package/dist/src/eval/dataset-manager.js

@@ -0,0 +1,97 @@
+/**
+ * DatasetManager - JSONL-based eval dataset management (Task 33)
+ */
+import { randomUUID } from 'crypto';
+import { appendFileSync, readFileSync, writeFileSync, existsSync } from 'fs';
+export class DatasetManager {
+    datasetsPath;
+    entriesPath;
+    constructor(datasetsPath, entriesPath) {
+        this.datasetsPath = datasetsPath;
+        this.entriesPath = entriesPath;
+    }
+    /**
+     * Create a dataset from a set of filtered traces.
+     */
+    createFromTraces(opts) {
+        const now = new Date().toISOString();
+        const agentSlugs = opts.agentSlugs ?? [...new Set(opts.traces.map((t) => t.agentSlug))];
+        const dataset = {
+            datasetId: randomUUID(),
+            name: opts.name,
+            description: opts.description,
+            agentSlugs,
+            createdAt: now,
+            updatedAt: now,
+            entryCount: opts.traces.length,
+        };
+        appendFileSync(this.datasetsPath, JSON.stringify(dataset) + '\n', 'utf-8');
+        for (const trace of opts.traces) {
+            const entry = {
+                entryId: randomUUID(),
+                datasetId: dataset.datasetId,
+                traceId: trace.traceId,
+                addedAt: now,
+            };
+            appendFileSync(this.entriesPath, JSON.stringify(entry) + '\n', 'utf-8');
+        }
+        return dataset;
+    }
+    /**
+     * List all datasets.
+     */
+    listDatasets() {
+        if (!existsSync(this.datasetsPath))
+            return [];
+        const content = readFileSync(this.datasetsPath, 'utf-8').trim();
+        if (!content)
+            return [];
+        return content.split('\n').map((line) => JSON.parse(line));
+    }
+    /**
+     * Get entries for a specific dataset.
+     */
+    getEntries(datasetId) {
+        if (!existsSync(this.entriesPath))
+            return [];
+        const content = readFileSync(this.entriesPath, 'utf-8').trim();
+        if (!content)
+            return [];
+        return content
+            .split('\n')
+            .map((line) => JSON.parse(line))
+            .filter((e) => e.datasetId === datasetId);
+    }
+    /**
+     * Add a single trace to an existing dataset.
+     */
+    addTraceToDataset(datasetId, traceId) {
+        const entry = {
+            entryId: randomUUID(),
+            datasetId,
+            traceId,
+            addedAt: new Date().toISOString(),
+        };
+        appendFileSync(this.entriesPath, JSON.stringify(entry) + '\n', 'utf-8');
+        // Update dataset entryCount
+        const datasets = this.listDatasets();
+        const updated = datasets.map((d) => {
+            if (d.datasetId === datasetId) {
+                return { ...d, entryCount: d.entryCount + 1, updatedAt: new Date().toISOString() };
+            }
+            return d;
+        });
+        writeFileSync(this.datasetsPath, updated.map((d) => JSON.stringify(d)).join('\n') + '\n', 'utf-8');
+        return entry;
+    }
+    /**
+     * Export a dataset to a JSON file.
+     */
+    exportToFile(datasetId, outputPath) {
+        const datasets = this.listDatasets();
+        const dataset = datasets.find((d) => d.datasetId === datasetId);
+        const entries = this.getEntries(datasetId);
+        writeFileSync(outputPath, JSON.stringify({ dataset, entries }, null, 2), 'utf-8');
+    }
+}
+//# sourceMappingURL=dataset-manager.js.map

package/dist/src/eval/dataset-manager.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"dataset-manager.js","sourceRoot":"","sources":["../../../src/eval/dataset-manager.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AACpC,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,aAAa,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AAU7E,MAAM,OAAO,cAAc;IACjB,YAAY,CAAS;IACrB,WAAW,CAAS;IAE5B,YAAY,YAAoB,EAAE,WAAmB;QACnD,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;IACjC,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,IAA0B;QACzC,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QACrC,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;QACxF,MAAM,OAAO,GAAgB;YAC3B,SAAS,EAAE,UAAU,EAAE;YACvB,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU;YACV,SAAS,EAAE,GAAG;YACd,SAAS,EAAE,GAAG;YACd,UAAU,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM;SAC/B,CAAC;QAEF,cAAc,CAAC,IAAI,CAAC,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAE3E,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChC,MAAM,KAAK,GAAqB;gBAC9B,OAAO,EAAE,UAAU,EAAE;gBACrB,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,OAAO,EAAE,GAAG;aACb,CAAC;YACF,cAAc,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAC1E,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,YAAY;QACV,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,YAAY,CAAC;YAAE,OAAO,EAAE,CAAC;QAC9C,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;QAChE,IAAI,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;QACxB,OAAO,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAgB,CAAC,CAAC;IAC5E,CAAC;IAED;;OAEG;IACH,UAAU,CAAC,SAAiB;QAC1B,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;YAAE,OAAO,EAAE,CAAC;QAC7C,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;QAC/D,IAAI,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;QACxB,OAAO,OAAO;aACX,KAAK,CAAC,IAAI,CAAC;aACX,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAqB,CAAC;aACnD,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC;IAC9C,CAAC;IAED;;OAEG;IACH,iBAAiB,CAAC,SAAiB,EAAE,OAAe;QAClD,MAAM,KAAK,GAAqB;YAC9B,OAAO,EAAE,UAAU,EAAE;YACrB,SAAS;YACT,OAAO;YACP,OAAO,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SAClC,CAAC;QACF,cAAc,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAExE,4BAA4B;QAC5B,MAAM,QAAQ,GAAG,IAAI,CAAC,YAAY,EAAE,CAAC;QACrC,MAAM,OAAO,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YACjC,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;gBAC9B,OAAO,EAAE,GAAG,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,GAAG,CAAC,EAAE,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC;YACrF,CAAC;YACD,OAAO,CAAC,CAAC;QACX,CAAC,CAAC,CAAC;QACH,aAAa,CAAC,IAAI,CAAC,YAAY,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAEnG,OAAO,KAAK,CAAC;IACf,CAAC;IAED;;OAEG;IACH,YAAY,CAAC,SAAiB,EAAE,UAAkB;QAChD,MAAM,QAAQ,GAAG,IAAI,CAAC,YAAY,EAAE,CAAC;QACrC,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC;QAChE,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC3C,aAAa,CAAC,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;IACpF,CAAC;CACF"}

package/dist/src/eval/dataset-runner.d.ts

@@ -0,0 +1,23 @@
+import type { EvalRunResult, EvalTrace } from '../../../shared/src/types/eval.js';
+export interface AgentRunnerResult {
+    agentOutput: string;
+    outcome: 'success' | 'failure' | 'timeout';
+    qualityScore: number;
+    latencyMs: number;
+}
+export interface DatasetRunOpts {
+    datasetId: string;
+    agentVersion: string;
+    traces: EvalTrace[];
+    agentRunner: (trace: EvalTrace) => Promise<AgentRunnerResult>;
+    baselineResult?: EvalRunResult;
+    regressionThreshold?: number;
+}
+export declare class DatasetRunner {
+    /**
+     * Run all traces through the agent runner and compute stats.
+     * Optionally compare against a baseline to detect regressions.
+     */
+    run(opts: DatasetRunOpts): Promise<EvalRunResult>;
+}
+//# sourceMappingURL=dataset-runner.d.ts.map

package/dist/src/eval/dataset-runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"dataset-runner.d.ts","sourceRoot":"","sources":["../../../src/eval/dataset-runner.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,aAAa,EAAE,SAAS,EAAoB,MAAM,mCAAmC,CAAC;AAEpG,MAAM,WAAW,iBAAiB;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,SAAS,GAAG,SAAS,GAAG,SAAS,CAAC;IAC3C,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,cAAc;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,SAAS,EAAE,CAAC;IACpB,WAAW,EAAE,CAAC,KAAK,EAAE,SAAS,KAAK,OAAO,CAAC,iBAAiB,CAAC,CAAC;IAC9D,cAAc,CAAC,EAAE,aAAa,CAAC;IAC/B,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC9B;AAED,qBAAa,aAAa;IACxB;;;OAGG;IACG,GAAG,CAAC,IAAI,EAAE,cAAc,GAAG,OAAO,CAAC,aAAa,CAAC;CA6DxD"}

package/dist/src/eval/dataset-runner.js

@@ -0,0 +1,59 @@
+/**
+ * DatasetRunner - Run eval datasets against agents and detect regressions (Task 33)
+ */
+import { randomUUID } from 'crypto';
+export class DatasetRunner {
+    /**
+     * Run all traces through the agent runner and compute stats.
+     * Optionally compare against a baseline to detect regressions.
+     */
+    async run(opts) {
+        const { datasetId, agentVersion, traces, agentRunner, baselineResult, regressionThreshold = 0.1, } = opts;
+        const results = [];
+        for (const trace of traces) {
+            const result = await agentRunner(trace);
+            results.push({ trace, result });
+        }
+        const passCount = results.filter((r) => r.result.outcome === 'success').length;
+        const failCount = results.length - passCount;
+        const totalQuality = results.reduce((sum, r) => sum + r.result.qualityScore, 0);
+        const totalLatency = results.reduce((sum, r) => sum + r.result.latencyMs, 0);
+        const avgQualityScore = results.length > 0 ? totalQuality / results.length : 0;
+        const avgLatencyMs = results.length > 0 ? totalLatency / results.length : 0;
+        // Regression detection
+        const regressionDetails = [];
+        let regressionDetected = false;
+        if (baselineResult) {
+            const delta = baselineResult.avgQualityScore - avgQualityScore;
+            if (delta > regressionThreshold) {
+                regressionDetected = true;
+                // Report per-trace regressions for traces with quality below baseline average
+                for (const { trace, result } of results) {
+                    if (result.qualityScore < baselineResult.avgQualityScore) {
+                        regressionDetails.push({
+                            traceId: trace.traceId,
+                            agentSlug: trace.agentSlug,
+                            baselineScore: baselineResult.avgQualityScore,
+                            currentScore: result.qualityScore,
+                            delta: baselineResult.avgQualityScore - result.qualityScore,
+                        });
+                    }
+                }
+            }
+        }
+        return {
+            runId: randomUUID(),
+            datasetId,
+            runAt: new Date().toISOString(),
+            agentVersion,
+            entriesTested: results.length,
+            passCount,
+            failCount,
+            avgQualityScore,
+            avgLatencyMs,
+            regressionDetected,
+            regressionDetails,
+        };
+    }
+}
+//# sourceMappingURL=dataset-runner.js.map

package/dist/src/eval/dataset-runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"dataset-runner.js","sourceRoot":"","sources":["../../../src/eval/dataset-runner.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AAmBpC,MAAM,OAAO,aAAa;IACxB;;;OAGG;IACH,KAAK,CAAC,GAAG,CAAC,IAAoB;QAC5B,MAAM,EACJ,SAAS,EACT,YAAY,EACZ,MAAM,EACN,WAAW,EACX,cAAc,EACd,mBAAmB,GAAG,GAAG,GAC1B,GAAG,IAAI,CAAC;QAET,MAAM,OAAO,GAA2D,EAAE,CAAC;QAE3E,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,KAAK,CAAC,CAAC;YACxC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAClC,CAAC;QAED,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;QAC/E,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC;QAC7C,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC;QAChF,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QAC7E,MAAM,eAAe,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAC/E,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAE5E,uBAAuB;QACvB,MAAM,iBAAiB,GAAuB,EAAE,CAAC;QACjD,IAAI,kBAAkB,GAAG,KAAK,CAAC;QAE/B,IAAI,cAAc,EAAE,CAAC;YACnB,MAAM,KAAK,GAAG,cAAc,CAAC,eAAe,GAAG,eAAe,CAAC;YAC/D,IAAI,KAAK,GAAG,mBAAmB,EAAE,CAAC;gBAChC,kBAAkB,GAAG,IAAI,CAAC;gBAC1B,8EAA8E;gBAC9E,KAAK,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,OAAO,EAAE,CAAC;oBACxC,IAAI,MAAM,CAAC,YAAY,GAAG,cAAc,CAAC,eAAe,EAAE,CAAC;wBACzD,iBAAiB,CAAC,IAAI,CAAC;4BACrB,OAAO,EAAE,KAAK,CAAC,OAAO;4BACtB,SAAS,EAAE,KAAK,CAAC,SAAS;4BAC1B,aAAa,EAAE,cAAc,CAAC,eAAe;4BAC7C,YAAY,EAAE,MAAM,CAAC,YAAY;4BACjC,KAAK,EAAE,cAAc,CAAC,eAAe,GAAG,MAAM,CAAC,YAAY;yBAC5D,CAAC,CAAC;oBACL,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO;YACL,KAAK,EAAE,UAAU,EAAE;YACnB,SAAS;YACT,KAAK,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YAC/B,YAAY;YACZ,aAAa,EAAE,OAAO,CAAC,MAAM;YAC7B,SAAS;YACT,SAAS;YACT,eAAe;YACf,YAAY;YACZ,kBAAkB;YAClB,iBAAiB;SAClB,CAAC;IACJ,CAAC;CACF"}

package/dist/src/eval/index.d.ts

@@ -0,0 +1,10 @@
+/**
+ * Eval Module - Automated eval dataset from production traces (Task 33)
+ */
+export { TraceCollector } from './trace-collector.js';
+export type { RecordTraceInput } from './trace-collector.js';
+export { DatasetManager } from './dataset-manager.js';
+export type { CreateFromTracesOpts } from './dataset-manager.js';
+export { DatasetRunner } from './dataset-runner.js';
+export type { AgentRunnerResult, DatasetRunOpts } from './dataset-runner.js';
+//# sourceMappingURL=index.d.ts.map

package/dist/src/eval/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/eval/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,YAAY,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAE7D,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,YAAY,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAEjE,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,YAAY,EAAE,iBAAiB,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC"}

package/dist/src/eval/index.js

@@ -0,0 +1,7 @@
+/**
+ * Eval Module - Automated eval dataset from production traces (Task 33)
+ */
+export { TraceCollector } from './trace-collector.js';
+export { DatasetManager } from './dataset-manager.js';
+export { DatasetRunner } from './dataset-runner.js';
+//# sourceMappingURL=index.js.map

package/dist/src/eval/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/eval/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAGtD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAGtD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC"}

package/dist/src/eval/trace-collector.d.ts

@@ -0,0 +1,40 @@
+import type { EvalTrace } from '../../../shared/src/types/eval.js';
+export interface RecordTraceInput {
+    agentSlug: string;
+    agentVersion: string;
+    taskDescription: string;
+    taskInput: string;
+    agentOutput: string;
+    retryCount: number;
+    qualityScore?: number;
+    outcome: 'success' | 'failure' | 'timeout';
+    latencyMs: number;
+    tokenCount?: number;
+    costUsd?: number;
+    correctedOutput?: string;
+}
+export declare class TraceCollector {
+    private filePath;
+    constructor(filePath: string);
+    /**
+     * Determine auto review status based on trace quality signals.
+     */
+    autoReviewStatus(input: RecordTraceInput): 'pending' | 'approved';
+    /**
+     * Auto-generate tags based on trace characteristics.
+     */
+    autoTag(input: RecordTraceInput): string[];
+    /**
+     * Record a trace, auto-generating traceId, capturedAt, reviewStatus, and tags.
+     */
+    record(input: RecordTraceInput): EvalTrace;
+    /**
+     * Read all traces from the JSONL file.
+     */
+    readAll(): EvalTrace[];
+    /**
+     * Get traces pending review, with optional limit.
+     */
+    getTracesPendingReview(limit?: number): EvalTrace[];
+}
+//# sourceMappingURL=trace-collector.d.ts.map

package/dist/src/eval/trace-collector.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"trace-collector.d.ts","sourceRoot":"","sources":["../../../src/eval/trace-collector.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,mCAAmC,CAAC;AAEnE,MAAM,WAAW,gBAAgB;IAC/B,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,OAAO,EAAE,SAAS,GAAG,SAAS,GAAG,SAAS,CAAC;IAC3C,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,qBAAa,cAAc;IACzB,OAAO,CAAC,QAAQ,CAAS;gBAEb,QAAQ,EAAE,MAAM;IAI5B;;OAEG;IACH,gBAAgB,CAAC,KAAK,EAAE,gBAAgB,GAAG,SAAS,GAAG,UAAU;IAOjE;;OAEG;IACH,OAAO,CAAC,KAAK,EAAE,gBAAgB,GAAG,MAAM,EAAE;IAQ1C;;OAEG;IACH,MAAM,CAAC,KAAK,EAAE,gBAAgB,GAAG,SAAS;IAwB1C;;OAEG;IACH,OAAO,IAAI,SAAS,EAAE;IAOtB;;OAEG;IACH,sBAAsB,CAAC,KAAK,CAAC,EAAE,MAAM,GAAG,SAAS,EAAE;CAKpD"}

package/dist/src/eval/trace-collector.js

@@ -0,0 +1,82 @@
+/**
+ * TraceCollector - JSONL-based production trace collection (Task 33)
+ */
+import { randomUUID } from 'crypto';
+import { appendFileSync, readFileSync, existsSync } from 'fs';
+export class TraceCollector {
+    filePath;
+    constructor(filePath) {
+        this.filePath = filePath;
+    }
+    /**
+     * Determine auto review status based on trace quality signals.
+     */
+    autoReviewStatus(input) {
+        if (input.retryCount > 1)
+            return 'pending';
+        if (input.qualityScore !== undefined && input.qualityScore < 0.6)
+            return 'pending';
+        if (input.outcome === 'failure')
+            return 'pending';
+        return 'approved';
+    }
+    /**
+     * Auto-generate tags based on trace characteristics.
+     */
+    autoTag(input) {
+        const tags = [];
+        if (input.retryCount > 1)
+            tags.push('high-retry');
+        if (input.outcome === 'failure')
+            tags.push('failure');
+        if (input.outcome === 'timeout')
+            tags.push('timeout');
+        return tags;
+    }
+    /**
+     * Record a trace, auto-generating traceId, capturedAt, reviewStatus, and tags.
+     */
+    record(input) {
+        const trace = {
+            traceId: randomUUID(),
+            agentSlug: input.agentSlug,
+            agentVersion: input.agentVersion,
+            taskDescription: input.taskDescription,
+            taskInput: input.taskInput,
+            agentOutput: input.agentOutput,
+            retryCount: input.retryCount,
+            qualityScore: input.qualityScore,
+            outcome: input.outcome,
+            latencyMs: input.latencyMs,
+            tokenCount: input.tokenCount,
+            costUsd: input.costUsd,
+            capturedAt: new Date().toISOString(),
+            reviewStatus: this.autoReviewStatus(input),
+            correctedOutput: input.correctedOutput,
+            tags: this.autoTag(input),
+        };
+        appendFileSync(this.filePath, JSON.stringify(trace) + '\n', 'utf-8');
+        return trace;
+    }
+    /**
+     * Read all traces from the JSONL file.
+     */
+    readAll() {
+        if (!existsSync(this.filePath))
+            return [];
+        const content = readFileSync(this.filePath, 'utf-8').trim();
+        if (!content)
+            return [];
+        return content.split('\n').map((line) => JSON.parse(line));
+    }
+    /**
+     * Get traces pending review, with optional limit.
+     */
+    getTracesPendingReview(limit) {
+        const all = this.readAll().filter((t) => t.reviewStatus === 'pending');
+        if (limit !== undefined)
+            return all.slice(0, limit);
+        return all;
+    }
+}
+//# sourceMappingURL=trace-collector.js.map

package/dist/src/eval/trace-collector.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"trace-collector.js","sourceRoot":"","sources":["../../../src/eval/trace-collector.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AACpC,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,IAAI,CAAC;AAkB9D,MAAM,OAAO,cAAc;IACjB,QAAQ,CAAS;IAEzB,YAAY,QAAgB;QAC1B,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAC3B,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,KAAuB;QACtC,IAAI,KAAK,CAAC,UAAU,GAAG,CAAC;YAAE,OAAO,SAAS,CAAC;QAC3C,IAAI,KAAK,CAAC,YAAY,KAAK,SAAS,IAAI,KAAK,CAAC,YAAY,GAAG,GAAG;YAAE,OAAO,SAAS,CAAC;QACnF,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS;YAAE,OAAO,SAAS,CAAC;QAClD,OAAO,UAAU,CAAC;IACpB,CAAC;IAED;;OAEG;IACH,OAAO,CAAC,KAAuB;QAC7B,MAAM,IAAI,GAAa,EAAE,CAAC;QAC1B,IAAI,KAAK,CAAC,UAAU,GAAG,CAAC;YAAE,IAAI,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAClD,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS;YAAE,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtD,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS;YAAE,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtD,OAAO,IAAI,CAAC;IACd,CAAC;IAED;;OAEG;IACH,MAAM,CAAC,KAAuB;QAC5B,MAAM,KAAK,GAAc;YACvB,OAAO,EAAE,UAAU,EAAE;YACrB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACpC,YAAY,EAAE,IAAI,CAAC,gBAAgB,CAAC,KAAK,CAAC;YAC1C,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,IAAI,EAAE,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC;SAC1B,CAAC;QAEF,cAAc,CAAC,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QACrE,OAAO,KAAK,CAAC;IACf,CAAC;IAED;;OAEG;IACH,OAAO;QACL,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC;YAAE,OAAO,EAAE,CAAC;QAC1C,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;QAC5D,IAAI,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;QACxB,OAAO,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAc,CAAC,CAAC;IAC1E,CAAC;IAED;;OAEG;IACH,sBAAsB,CAAC,KAAc;QACnC,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,SAAS,CAAC,CAAC;QACvE,IAAI,KAAK,KAAK,SAAS;YAAE,OAAO,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;QACpD,OAAO,GAAG,CAAC;IACb,CAAC;CACF"}

package/dist/src/index.d.ts

@@ -0,0 +1,82 @@
+/**
+ * CLI Main Entry Point
+ * Modernized CLI for Monobrain
+ *
+ * github.com/nokhodian/monobrain
+ */
+export declare const VERSION: string;
+export interface CLIOptions {
+    name?: string;
+    description?: string;
+    version?: string;
+    interactive?: boolean;
+}
+/**
+ * CLI Application
+ */
+export declare class CLI {
+    private name;
+    private description;
+    private version;
+    private parser;
+    private output;
+    private interactive;
+    constructor(options?: CLIOptions);
+    /**
+     * Run the CLI with given arguments
+     */
+    run(args?: string[]): Promise<void>;
+    /**
+     * Show main help
+     */
+    private showHelp;
+    /**
+     * Show command-specific help
+     */
+    private showCommandHelp;
+    /**
+     * Show version
+     */
+    private showVersion;
+    /**
+     * Check for updates on startup (non-blocking)
+     * Shows notification if updates are available
+     */
+    private checkForUpdatesOnStartup;
+    /**
+     * Load configuration file
+     */
+    private loadConfig;
+    /**
+     * Initialize optional subsystems at startup (non-blocking, all failures are silent).
+     * Wires TierManager, ObservabilityBus + TraceCollector, and SwarmCheckpointer
+     * so that packages/@monobrain/* actually contribute to the live runtime.
+     */
+    private initSubsystems;
+    /**
+     * Handle errors
+     */
+    private handleError;
+}
+export * from './types.js';
+export { CommandParser, commandParser } from './parser.js';
+export { OutputFormatter, output, Progress, Spinner, type VerbosityLevel } from './output.js';
+export * from './prompt.js';
+export * from './commands/index.js';
+export { MCPServerManager, createMCPServerManager, getServerManager, startMCPServer, stopMCPServer, getMCPServerStatus, type MCPServerOptions, type MCPServerStatus, } from './mcp-server.js';
+export { initializeMemoryDatabase, generateEmbedding, generateBatchEmbeddings, storeEntry, searchEntries, getHNSWIndex, addToHNSWIndex, searchHNSWIndex, getHNSWStatus, clearHNSWIndex, quantizeInt8, dequantizeInt8, quantizedCosineSim, getQuantizationStats, batchCosineSim, softmaxAttention, topKIndices, flashAttentionSearch, type MemoryInitResult, } from './memory/memory-initializer.js';
+export { initializeIntelligence, recordStep, recordTrajectory, findSimilarPatterns, getIntelligenceStats, getSonaCoordinator, getReasoningBank, clearIntelligence, benchmarkAdaptation, endTrajectoryWithVerdict, distillLearning, getAllPatterns, getPatternsByType, flushPatterns, deletePattern, clearAllPatterns, getNeuralDataDir, getPersistenceStatus, type SonaConfig, type TrajectoryStep, type Pattern, type IntelligenceStats, } from './memory/intelligence.js';
+export { EWCConsolidator, getEWCConsolidator, resetEWCConsolidator, consolidatePatterns, recordPatternOutcome, getEWCStats, type PatternWeights, type EWCConfig, type ConsolidationResult, type EWCStats, } from './memory/ewc-consolidation.js';
+export { SONAOptimizer, getSONAOptimizer, resetSONAOptimizer, processTrajectory, getSuggestion, getSONAStats, type TrajectoryOutcome, type LearnedPattern, type RoutingSuggestion, type SONAStats, } from './memory/sona-optimizer.js';
+export { ErrorHandler, withErrorHandling, } from './production/error-handler.js';
+export type { ErrorContext, ErrorHandlerConfig, } from './production/error-handler.js';
+export { RateLimiter, createRateLimiter, } from './production/rate-limiter.js';
+export type { RateLimiterConfig, RateLimitResult, } from './production/rate-limiter.js';
+export { withRetry, makeRetryable, } from './production/retry.js';
+export type { RetryConfig, RetryResult, RetryStrategy, } from './production/retry.js';
+export { CircuitBreaker, getCircuitBreaker, getAllCircuitStats, resetAllCircuits, } from './production/circuit-breaker.js';
+export type { CircuitBreakerConfig, CircuitState, } from './production/circuit-breaker.js';
+export { MonitoringHooks, createMonitor, getMonitor, } from './production/monitoring.js';
+export type { MonitorConfig, MetricEvent, HealthStatus, PerformanceMetrics, } from './production/monitoring.js';
+export default CLI;
+//# sourceMappingURL=index.d.ts.map

package/dist/src/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AA0BH,eAAO,MAAM,OAAO,QAAsB,CAAC;AAE3C,MAAM,WAAW,UAAU;IACzB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,WAAW,CAAC,EAAE,OAAO,CAAC;CACvB;AAED;;GAEG;AACH,qBAAa,GAAG;IACd,OAAO,CAAC,IAAI,CAAS;IACrB,OAAO,CAAC,WAAW,CAAS;IAC5B,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,MAAM,CAAgB;IAC9B,OAAO,CAAC,MAAM,CAAkB;IAChC,OAAO,CAAC,WAAW,CAAU;gBAEjB,OAAO,GAAE,UAAe;IAcpC;;OAEG;IACG,GAAG,CAAC,IAAI,GAAE,MAAM,EAA0B,GAAG,OAAO,CAAC,IAAI,CAAC;IAgMhE;;OAEG;IACH,OAAO,CAAC,QAAQ;IAsFhB;;OAEG;YACW,eAAe;IAoD7B;;OAEG;IACH,OAAO,CAAC,WAAW;IAInB;;;OAGG;YACW,wBAAwB;IA4BtC;;OAEG;YACW,UAAU;IAkCxB;;;;OAIG;YACW,cAAc;IAqG5B;;OAEG;IACH,OAAO,CAAC,WAAW;CAuBpB;AAOD,cAAc,YAAY,CAAC;AAG3B,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAG3D,OAAO,EAAE,eAAe,EAAE,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,cAAc,EAAE,MAAM,aAAa,CAAC;AAG9F,cAAc,aAAa,CAAC;AAG5B,cAAc,qBAAqB,CAAC;AAGpC,OAAO,EACL,gBAAgB,EAChB,sBAAsB,EACtB,gBAAgB,EAChB,cAAc,EACd,aAAa,EACb,kBAAkB,EAClB,KAAK,gBAAgB,EACrB,KAAK,eAAe,GACrB,MAAM,iBAAiB,CAAC;AAGzB,OAAO,EACL,wBAAwB,EACxB,iBAAiB,EACjB,uBAAuB,EACvB,UAAU,EACV,aAAa,EACb,YAAY,EACZ,cAAc,EACd,eAAe,EACf,aAAa,EACb,cAAc,EACd,YAAY,EACZ,cAAc,EACd,kBAAkB,EAClB,oBAAoB,EAEpB,cAAc,EACd,gBAAgB,EAChB,WAAW,EACX,oBAAoB,EACpB,KAAK,gBAAgB,GACtB,MAAM,gCAAgC,CAAC;AAExC,OAAO,EACL,sBAAsB,EACtB,UAAU,EACV,gBAAgB,EAChB,mBAAmB,EACnB,oBAAoB,EACpB,kBAAkB,EAClB,gBAAgB,EAChB,iBAAiB,EACjB,mBAAmB,EAEnB,wBAAwB,EACxB,eAAe,EAEf,cAAc,EACd,iBAAiB,EACjB,aAAa,EACb,aAAa,EACb,gBAAgB,EAChB,gBAAgB,EAChB,oBAAoB,EACpB,KAAK,UAAU,EACf,KAAK,cAAc,EACnB,KAAK,OAAO,EACZ,KAAK,iBAAiB,GACvB,MAAM,0BAA0B,CAAC;AAGlC,OAAO,EACL,eAAe,EACf,kBAAkB,EAClB,oBAAoB,EACpB,mBAAmB,EACnB,oBAAoB,EACpB,WAAW,EACX,KAAK,cAAc,EACnB,KAAK,SAAS,EACd,KAAK,mBAAmB,EACxB,KAAK,QAAQ,GACd,MAAM,+BAA+B,CAAC;AAGvC,OAAO,EACL,aAAa,EACb,gBAAgB,EAChB,kBAAkB,EAClB,iBAAiB,EACjB,aAAa,EACb,YAAY,EACZ,KAAK,iBAAiB,EACtB,KAAK,cAAc,EACnB,KAAK,iBAAiB,EACtB,KAAK,SAAS,GACf,MAAM,4BAA4B,CAAC;AAGpC,OAAO,EACL,YAAY,EACZ,iBAAiB,GAClB,MAAM,+BAA+B,CAAC;AACvC,YAAY,EACV,YAAY,EACZ,kBAAkB,GACnB,MAAM,+BAA+B,CAAC;AAEvC,OAAO,EACL,WAAW,EACX,iBAAiB,GAClB,MAAM,8BAA8B,CAAC;AACtC,YAAY,EACV,iBAAiB,EACjB,eAAe,GAChB,MAAM,8BAA8B,CAAC;AAEtC,OAAO,EACL,SAAS,EACT,aAAa,GACd,MAAM,uBAAuB,CAAC;AAC/B,YAAY,EACV,WAAW,EACX,WAAW,EACX,aAAa,GACd,MAAM,uBAAuB,CAAC;AAE/B,OAAO,EACL,cAAc,EACd,iBAAiB,EACjB,kBAAkB,EAClB,gBAAgB,GACjB,MAAM,iCAAiC,CAAC;AACzC,YAAY,EACV,oBAAoB,EACpB,YAAY,GACb,MAAM,iCAAiC,CAAC;AAEzC,OAAO,EACL,eAAe,EACf,aAAa,EACb,UAAU,GACX,MAAM,4BAA4B,CAAC;AACpC,YAAY,EACV,aAAa,EACb,WAAW,EACX,YAAY,EACZ,kBAAkB,GACnB,MAAM,4BAA4B,CAAC;AAGpC,eAAe,GAAG,CAAC"}