@ls-stack/agent-eval 0.34.0 → 0.35.1

package/dist/index.mjs

@@ -1,4 +1,4 @@
-import { $ as apiCallMetricSchema, $n as getCurrentScope, $t as cacheDebugKeyEntrySchema, A as createRunRequestSchema, An as repoFileRefSchema, At as runLogEntrySchema, B as getNestedAttribute, Bn as deserializeCacheValue, Bt as manualInputNumberFieldSchema, Cn as cellValueSchema, Ct as caseRowSchema, Dn as fileRefSchema, Dt as evalStatItemSchema, En as columnKindSchema, Et as evalStatAggregateSchema, F as extractApiCalls, Fn as evalSpan, Ft as manualInputBooleanFieldSchema, G as deriveStatusFromChildStatuses, Gn as readManualInputFile, Gt as evalChartAxisSchema, H as getEvalDisplayStatus, Hn as serializeCacheValue, Ht as manualInputSelectOptionSchema, I as extractLlmCalls, In as evalTracer, It as manualInputDescriptorSchema, J as DEFAULT_API_CALLS_CONFIG, Jn as advanceEvalTime, Jt as evalChartConfigSchema, K as runManifestSchema, Kn as evalExpect, Kt as evalChartBuiltinMetricSchema, L as simulateLlmCallCost, Ln as hashCacheKey, Lt as manualInputFieldDescriptorSchema, M as sseEnvelopeSchema, Mn as z, Mt as runLogLocationSchema, N as extractCacheEntries, Nn as buildTraceTree, Nt as runLogPhaseSchema, O as configReloadStateSchema, On as jsonCellSchema, Ot as evalStatsConfigSchema, P as extractCacheHits, Pn as captureEvalSpanError, Pt as scoreTraceSchema, Q as apiCallMetricPlacementSchema, Qn as evalLog, Qt as evalChartsConfigSchema, R as simulateTokenAllocation, Rn as hashCacheKeySync, Rt as manualInputJsonFieldSchema, Sn as traceSpanWarningSchema, St as caseDetailSchema, Tn as columnFormatSchema, Tt as evalFreshnessStatusSchema, U as deriveScopedSummaryFromCases, Un as repoFile, Ut as manualInputTextFieldSchema, V as getEvalTitle, Vn as serializeCacheRecording, Vt as manualInputSelectFieldSchema, W as deriveStatusFromCaseRows, Wn as manualInputFileValueSchema, Wt as evalChartAggregateSchema, X as agentEvalsConfigSchema, Xt as evalChartTooltipExtraSchema, Y as DEFAULT_LLM_CALLS_CONFIG, Yn as appendToEvalOutput, Yt as evalChartMetricSchema, Z as apiCallMetricFormatSchema, Zn as evalAssert, Zt as evalChartTypeSchema, _n as traceDisplayConfigSchema, _t as buildCaseKey, an as cacheModeSchema, ar as nextEvalId, at as llmCallCostCurrencySchema, bn as traceSpanKindSchema, bt as getCaseRowEvalKey, cn as cacheRecordingSchema, cr as runInExistingEvalScope, ct as llmCallMetricSchema, dn as spanCacheOptionsSchema, dr as startEvalBackgroundJob, dt as llmCallsConfigSchema, en as cacheDebugKeyFileSchema, er as getEvalCaseInput, et as apiCallsConfigSchema, fn as traceCacheRefSchema, fr as defineEval, ft as removeDefaultConfigSchema, gn as traceAttributeDisplaySchema, gt as trialSelectionModeSchema, hn as traceAttributeDisplayPlacementSchema, ht as runLogsConfigSchema, in as cacheListItemSchema, ir as mergeEvalOutput, it as evalDeriveConfigSchema, j as updateManualScoreRequestSchema, jn as runArtifactRefSchema, jt as runLogLevelSchema, k as configReloadStatusSchema, kn as numberDisplayOptionsSchema, kt as evalSummarySchema, ln as cacheStatusSchema, lr as setEvalOutput, lt as llmCallPricingRateSchema, mn as traceAttributeDisplayInputSchema, mt as resolveLlmCallsConfig, nn as cacheEntryWithDebugKeySchema, nr as incrementEvalOutput, nt as evalColumnOverrideSchema, on as cacheOperationTypeSchema, or as runInEvalRuntimeScope, ot as llmCallMetricFormatSchema, pn as traceAttributeDisplayFormatSchema, pr as getEvalRegistry, pt as resolveApiCallsConfig, q as runSummarySchema, qn as EvalAssertionError, qt as evalChartColorSchema, rn as cacheFileSchema, rr as isInEvalScope, rt as evalColumnsSchema, sn as cacheRecordingOpSchema, sr as runInEvalScope, st as llmCallMetricPlacementSchema, tn as cacheEntrySchema, tr as getEvalStartTime, tt as defaultConfigKeySchema, un as serializedCacheSpanSchema, ur as setScopeCacheContext, ut as llmCallPricingSchema, vn as traceDisplayInputConfigSchema, vt as buildEvalKey, wn as columnDefSchema, wt as discoveryIssueSchema, xn as traceSpanSchema, xt as assertionFailureSchema, yn as traceSpanErrorSchema, yt as getCaseRowCaseKey, z as applyDerivedCallAttributes, zn as deserializeCacheRecording, zt as manualInputMultilineFieldSchema } from "./runOrchestration-CYhn4DJ0.mjs";
-import { a as materializeManualInputFiles, i as isManualInputFileValue, n as createRunner, o as stageManualInputFile, r as cleanupStagedManualInputFiles, s as stageManualInputFileFromPath, t as runCli } from "./cli-CG66f0GN.mjs";
-import "./src-B0b1vgk0.mjs";
+import { $ as apiCallMetricSchema, $n as getCurrentScope, $t as cacheDebugKeyEntrySchema, A as createRunRequestSchema, An as repoFileRefSchema, At as runLogEntrySchema, B as getNestedAttribute, Bn as deserializeCacheValue, Bt as manualInputNumberFieldSchema, Cn as cellValueSchema, Ct as caseRowSchema, Dn as fileRefSchema, Dt as evalStatItemSchema, En as columnKindSchema, Et as evalStatAggregateSchema, F as extractApiCalls, Fn as evalSpan, Ft as manualInputBooleanFieldSchema, G as deriveStatusFromChildStatuses, Gn as readManualInputFile, Gt as evalChartAxisSchema, H as getEvalDisplayStatus, Hn as serializeCacheValue, Ht as manualInputSelectOptionSchema, I as extractLlmCalls, In as evalTracer, It as manualInputDescriptorSchema, J as DEFAULT_API_CALLS_CONFIG, Jn as advanceEvalTime, Jt as evalChartConfigSchema, K as runManifestSchema, Kn as evalExpect, Kt as evalChartBuiltinMetricSchema, L as simulateLlmCallCost, Ln as hashCacheKey, Lt as manualInputFieldDescriptorSchema, M as sseEnvelopeSchema, Mn as z, Mt as runLogLocationSchema, N as extractCacheEntries, Nn as buildTraceTree, Nt as runLogPhaseSchema, O as configReloadStateSchema, On as jsonCellSchema, Ot as evalStatsConfigSchema, P as extractCacheHits, Pn as captureEvalSpanError, Pt as scoreTraceSchema, Q as apiCallMetricPlacementSchema, Qn as evalLog, Qt as evalChartsConfigSchema, R as simulateTokenAllocation, Rn as hashCacheKeySync, Rt as manualInputJsonFieldSchema, Sn as traceSpanWarningSchema, St as caseDetailSchema, Tn as columnFormatSchema, Tt as evalFreshnessStatusSchema, U as deriveScopedSummaryFromCases, Un as repoFile, Ut as manualInputTextFieldSchema, V as getEvalTitle, Vn as serializeCacheRecording, Vt as manualInputSelectFieldSchema, W as deriveStatusFromCaseRows, Wn as manualInputFileValueSchema, Wt as evalChartAggregateSchema, X as agentEvalsConfigSchema, Xt as evalChartTooltipExtraSchema, Y as DEFAULT_LLM_CALLS_CONFIG, Yn as appendToEvalOutput, Yt as evalChartMetricSchema, Z as apiCallMetricFormatSchema, Zn as evalAssert, Zt as evalChartTypeSchema, _n as traceDisplayConfigSchema, _t as buildCaseKey, an as cacheModeSchema, ar as nextEvalId, at as llmCallCostCurrencySchema, bn as traceSpanKindSchema, bt as getCaseRowEvalKey, cn as cacheRecordingSchema, cr as runInExistingEvalScope, ct as llmCallMetricSchema, dn as spanCacheOptionsSchema, dr as startEvalBackgroundJob, dt as llmCallsConfigSchema, en as cacheDebugKeyFileSchema, er as getEvalCaseInput, et as apiCallsConfigSchema, fn as traceCacheRefSchema, fr as defineEval, ft as removeDefaultConfigSchema, gn as traceAttributeDisplaySchema, gt as trialSelectionModeSchema, hn as traceAttributeDisplayPlacementSchema, ht as runLogsConfigSchema, in as cacheListItemSchema, ir as mergeEvalOutput, it as evalDeriveConfigSchema, j as updateManualScoreRequestSchema, jn as runArtifactRefSchema, jt as runLogLevelSchema, k as configReloadStatusSchema, kn as numberDisplayOptionsSchema, kt as evalSummarySchema, ln as cacheStatusSchema, lr as setEvalOutput, lt as llmCallPricingRateSchema, mn as traceAttributeDisplayInputSchema, mt as resolveLlmCallsConfig, nn as cacheEntryWithDebugKeySchema, nr as incrementEvalOutput, nt as evalColumnOverrideSchema, on as cacheOperationTypeSchema, or as runInEvalRuntimeScope, ot as llmCallMetricFormatSchema, pn as traceAttributeDisplayFormatSchema, pr as getEvalRegistry, pt as resolveApiCallsConfig, q as runSummarySchema, qn as EvalAssertionError, qt as evalChartColorSchema, rn as cacheFileSchema, rr as isInEvalScope, rt as evalColumnsSchema, sn as cacheRecordingOpSchema, sr as runInEvalScope, st as llmCallMetricPlacementSchema, tn as cacheEntrySchema, tr as getEvalStartTime, tt as defaultConfigKeySchema, un as serializedCacheSpanSchema, ur as setScopeCacheContext, ut as llmCallPricingSchema, vn as traceDisplayInputConfigSchema, vt as buildEvalKey, wn as columnDefSchema, wt as discoveryIssueSchema, xn as traceSpanSchema, xt as assertionFailureSchema, yn as traceSpanErrorSchema, yt as getCaseRowCaseKey, z as applyDerivedCallAttributes, zn as deserializeCacheRecording, zt as manualInputMultilineFieldSchema } from "./runOrchestration-V1TxX8es.mjs";
+import { a as materializeManualInputFiles, i as isManualInputFileValue, n as createRunner, o as stageManualInputFile, r as cleanupStagedManualInputFiles, s as stageManualInputFileFromPath, t as runCli } from "./cli-Dg3abrOv.mjs";
+import "./src-DBypR4TV.mjs";
 export { DEFAULT_API_CALLS_CONFIG, DEFAULT_LLM_CALLS_CONFIG, EvalAssertionError, advanceEvalTime, agentEvalsConfigSchema, apiCallMetricFormatSchema, apiCallMetricPlacementSchema, apiCallMetricSchema, apiCallsConfigSchema, appendToEvalOutput, applyDerivedCallAttributes, assertionFailureSchema, buildCaseKey, buildEvalKey, buildTraceTree, cacheDebugKeyEntrySchema, cacheDebugKeyFileSchema, cacheEntrySchema, cacheEntryWithDebugKeySchema, cacheFileSchema, cacheListItemSchema, cacheModeSchema, cacheOperationTypeSchema, cacheRecordingOpSchema, cacheRecordingSchema, cacheStatusSchema, captureEvalSpanError, caseDetailSchema, caseRowSchema, cellValueSchema, cleanupStagedManualInputFiles, columnDefSchema, columnFormatSchema, columnKindSchema, configReloadStateSchema, configReloadStatusSchema, createRunRequestSchema, createRunner, defaultConfigKeySchema, defineEval, deriveScopedSummaryFromCases, deriveStatusFromCaseRows, deriveStatusFromChildStatuses, deserializeCacheRecording, deserializeCacheValue, discoveryIssueSchema, evalAssert, evalChartAggregateSchema, evalChartAxisSchema, evalChartBuiltinMetricSchema, evalChartColorSchema, evalChartConfigSchema, evalChartMetricSchema, evalChartTooltipExtraSchema, evalChartTypeSchema, evalChartsConfigSchema, evalColumnOverrideSchema, evalColumnsSchema, evalDeriveConfigSchema, evalExpect, evalFreshnessStatusSchema, evalLog, evalSpan, evalStatAggregateSchema, evalStatItemSchema, evalStatsConfigSchema, evalSummarySchema, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, fileRefSchema, getCaseRowCaseKey, getCaseRowEvalKey, getCurrentScope, getEvalCaseInput, getEvalDisplayStatus, getEvalRegistry, getEvalStartTime, getEvalTitle, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, isManualInputFileValue, jsonCellSchema, llmCallCostCurrencySchema, llmCallMetricFormatSchema, llmCallMetricPlacementSchema, llmCallMetricSchema, llmCallPricingRateSchema, llmCallPricingSchema, llmCallsConfigSchema, manualInputBooleanFieldSchema, manualInputDescriptorSchema, manualInputFieldDescriptorSchema, manualInputFileValueSchema, manualInputJsonFieldSchema, manualInputMultilineFieldSchema, manualInputNumberFieldSchema, manualInputSelectFieldSchema, manualInputSelectOptionSchema, manualInputTextFieldSchema, materializeManualInputFiles, mergeEvalOutput, nextEvalId, numberDisplayOptionsSchema, readManualInputFile, removeDefaultConfigSchema, repoFile, repoFileRefSchema, resolveApiCallsConfig, resolveLlmCallsConfig, runArtifactRefSchema, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, runLogEntrySchema, runLogLevelSchema, runLogLocationSchema, runLogPhaseSchema, runLogsConfigSchema, runManifestSchema, runSummarySchema, scoreTraceSchema, serializeCacheRecording, serializeCacheValue, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, simulateLlmCallCost, simulateTokenAllocation, spanCacheOptionsSchema, sseEnvelopeSchema, stageManualInputFile, stageManualInputFileFromPath, startEvalBackgroundJob, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceCacheRefSchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanErrorSchema, traceSpanKindSchema, traceSpanSchema, traceSpanWarningSchema, trialSelectionModeSchema, updateManualScoreRequestSchema, z };

package/dist/runChild.mjs

@@ -1,4 +1,4 @@
-import { A as createRunRequestSchema, C as loadConfig, D as createFsCacheStore, It as manualInputDescriptorSchema, K as runManifestSchema, Ot as evalStatsConfigSchema, Qt as evalChartsConfigSchema, Xn as configureEvalRunLogs, q as runSummarySchema, r as getTargetEvals$1, t as executeRun, vt as buildEvalKey, wn as columnDefSchema, x as parseEvalDiscovery } from "./runOrchestration-CYhn4DJ0.mjs";
+import { A as createRunRequestSchema, C as loadConfig, D as createFsCacheStore, It as manualInputDescriptorSchema, K as runManifestSchema, Ot as evalStatsConfigSchema, Qt as evalChartsConfigSchema, Xn as configureEvalRunLogs, q as runSummarySchema, r as getTargetEvals$1, t as executeRun, vt as buildEvalKey, wn as columnDefSchema, x as parseEvalDiscovery } from "./runOrchestration-V1TxX8es.mjs";
 import { createHash } from "node:crypto";
 import { readFile } from "node:fs/promises";
 import { relative } from "node:path";

package/dist/{runOrchestration-CYhn4DJ0.mjs → runOrchestration-V1TxX8es.mjs}

@@ -913,13 +913,16 @@ async function cloneCacheValue(value, options = void 0) {
 	return deserializeCacheValue(await serializeCacheValue(value, options));
 }
 function normalizeCacheSerializationOptions(options) {
-	return { preserveUndefined: options?.preserveUndefined === true };
+	return {
+		compress: options?.compress !== false,
+		preserveUndefined: options?.preserveUndefined === true
+	};
 }
 async function serializeJsonSafeValue(value, refs, depth, config) {
 	if (value === void 0) return config.preserveUndefined ? jsonSafeValue("Undefined") : void 0;
 	if (typeof value === "bigint") return jsonSafeValue("BigInt", value.toString());
 	if (typeof value === "number") return serializeNumber(value);
-	if (typeof value === "string") return serializeString(value, depth);
+	if (typeof value === "string") return serializeString(value, depth, config);
 	if (value instanceof Date) return jsonSafeValue("Date", value.toISOString());
 	if (value instanceof Map) return serializeMap(value, refs, depth, config);
 	if (value instanceof Set) return serializeSet(value, refs, depth, config);
@@ -959,7 +962,7 @@ async function serializeJsonSafeValue(value, refs, depth, config) {
 			if (serializedItem !== void 0) items.push(serializedItem);
 		}
 		refs.delete(value);
-		return compressNestedJsonValue(items, depth) ?? items;
+		return compressNestedJsonValue(items, depth, config) ?? items;
 	}
 	const entries = [];
 	for (const [key, entryValue] of Object.entries(value)) {
@@ -968,7 +971,7 @@ async function serializeJsonSafeValue(value, refs, depth, config) {
 	}
 	refs.delete(value);
 	const serialized = hasSerializationMarkerKey(value) ? jsonSafeValue("Object", entries) : Object.fromEntries(entries);
-	return compressNestedJsonValue(serialized, depth) ?? serialized;
+	return compressNestedJsonValue(serialized, depth, config) ?? serialized;
 }
 function serializeNumber(value) {
 	if (Number.isNaN(value)) return jsonSafeValue("Number", "NaN");
@@ -977,8 +980,9 @@ function serializeNumber(value) {
 	if (Object.is(value, -0)) return jsonSafeValue("Number", "-0");
 	return value;
 }
-function serializeString(value, depth) {
+function serializeString(value, depth, config) {
 	if (depth === 0) return value;
+	if (!config.compress) return value;
 	return compressNestedStringValue(value) ?? value;
 }
 function isDenseNumberArray(value) {
@@ -1018,8 +1022,9 @@ function compressNestedStringValue(value) {
 	};
 	return compressionIsWorthIt(serialized, rawSize) ? serialized : void 0;
 }
-function compressNestedJsonValue(value, depth) {
+function compressNestedJsonValue(value, depth, config) {
 	if (depth === 0) return void 0;
+	if (!config.compress) return void 0;
 	const raw = JSON.stringify(value);
 	const rawSize = Buffer$1.byteLength(raw);
 	if (rawSize < compressedJsonMinBytes) return void 0;
@@ -5070,27 +5075,16 @@ function buildDeclaredColumnDefs(overrides, scores, manualScores) {
 }
 /**
 * Coerce an arbitrary runtime value into a serializable `CellValue`.
-* JSON-safe objects and arrays stay structured so saved run artifacts preserve
-* the authored output shape. Rich runtime values fall back to `JSON.stringify`.
+* Runtime values use the SDK's tagged serializer so saved run artifacts keep
+* structured data instead of storing JSON strings. Native binary/file root
+* values are handled before this helper.
 */
-function toCellValue(value) {
-	if (value === null) return null;
-	if (typeof value === "string" || typeof value === "number" || typeof value === "boolean") return value;
-	if (value === void 0) return void 0;
+async function toCellValue(value) {
 	const fileRef = fileRefSchema.safeParse(value);
 	if (fileRef.success) return fileRef.data;
-	if (isPlainJsonContainer(value)) {
-		const parsed = jsonCellSchema.safeParse(value);
-		if (parsed.success) return parsed.data;
-	}
-	if (value instanceof Date) return value.toISOString();
-	return JSON.stringify(value);
-}
-function isPlainJsonContainer(value) {
-	if (Array.isArray(value)) return true;
-	if (typeof value !== "object" || value === null) return false;
-	const prototype = Object.getPrototypeOf(value);
-	return prototype === Object.prototype || prototype === null;
+	const serialized = await serializeCacheValue(value, { compress: false });
+	const parsed = jsonCellSchema.safeParse(serialized);
+	if (parsed.success) return parsed.data;
 }
 function inferKindFromFormat(format) {
 	if (format === "boolean") return "boolean";
@@ -6723,7 +6717,7 @@ async function runCase(params) {
 			outputKey: key,
 			trial,
 			value
-		}) : toCellValue(value);
+		}) : await toCellValue(value);
 		if (cell !== void 0) columns[key] = cell;
 	}
 	for (const key of Object.keys(evalDef.manualScores ?? {})) columns[key] = null;

package/dist/{runner-j9b6g0h0.mjs → runner-BCs5rzej.mjs}

@@ -1,5 +1,5 @@
-import { n as createRunner } from "./cli-CG66f0GN.mjs";
-import "./src-B0b1vgk0.mjs";
+import { n as createRunner } from "./cli-Dg3abrOv.mjs";
+import "./src-DBypR4TV.mjs";
 //#region ../../apps/server/src/runner.ts
 let runnerInstance = null;
 function getRunnerInstance() {

package/dist/{runner-D5i7HLcE.mjs → runner-znY6PY1M.mjs}

@@ -1,2 +1,2 @@
-import { n as initRunner, t as getRunnerInstance } from "./runner-j9b6g0h0.mjs";
+import { n as initRunner, t as getRunnerInstance } from "./runner-BCs5rzej.mjs";
 export { getRunnerInstance, initRunner };

package/dist/src-DBypR4TV.mjs

@@ -0,0 +1,3 @@
+import "./runOrchestration-V1TxX8es.mjs";
+import "./cli-Dg3abrOv.mjs";
+export {};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.34.0",
+  "version": "0.35.1",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"
@@ -32,8 +32,8 @@
     "@types/node": "^24.7.2",
     "typescript": "^5.9.2",
     "@agent-evals/runner": "0.0.1",
-    "@agent-evals/shared": "0.0.1",
-    "@agent-evals/sdk": "0.0.1"
+    "@agent-evals/sdk": "0.0.1",
+    "@agent-evals/shared": "0.0.1"
   },
   "scripts": {
     "build": "pnpm --filter @agent-evals/web build && pnpm --filter @agent-evals/shared build && pnpm --filter @agent-evals/sdk build && pnpm --filter @agent-evals/runner build && tsdown --filter cli-js && tsdown --filter cli-types",

package/skills/agent-eval/SKILL.md

@@ -298,11 +298,12 @@ See `EvalScoreDef` / `EvalManualScoreDef` in the types for the full shape
 ## Outputs, columns, trace display
 
 - `setEvalOutput(key, value)` writes reviewable data for the case. Values are
-  plain data (strings, numbers, booleans, JSON-safe objects) plus native
-  `Blob`/`File` or `FileRef` variants for media columns. Stored output values
-  are not coerced by `columns.format`; column formats only control
-  visualization. Native `Blob`/`File` values are copied to run artifacts
-  because saved run files are JSON. Inside `execute`, prefer the context
+  stored as received: primitives, objects/arrays, explicit file refs, and
+  native `Blob`/`File` values. `columns.format` only controls visualization.
+  Non-JSON runtime values such as `Date`, `Map`, `Set`, `BigInt`, typed arrays,
+  and class instances use the tagged value serializer instead of a string
+  fallback. Native `Blob`/`File` values are copied to run artifacts because
+  saved run files are JSON. Inside `execute`, prefer the context
   `setOutput(key, value)` helper when writing schema-backed outputs; it is
   typed from the eval's outputs generic. Keep `setEvalOutput` for shared
   workflow code that does not receive the execute context.