@ls-stack/agent-eval 0.32.0 → 0.34.0

package/dist/{runOrchestration-ZpN7xty_.mjs → runOrchestration-CYhn4DJ0.mjs}

@@ -1688,7 +1688,7 @@ function createTraceCache(generateSpanId) {
 			const hit = await cacheCtx.adapter.lookup(namespace, keyHash);
 			if (hit) {
 				const storedAt = hit.storedAt;
-				const age = Date.now() - new Date(storedAt).getTime();
+				const age = getRealDateNowMs() - new Date(storedAt).getTime();
 				recordCacheRef(scope, activeSpan, {
 					type: "value",
 					name: info.name,
@@ -2149,7 +2149,7 @@ async function traceSpanInternal(info, fn) {
 					mergeSpanAttributes(spanRecord, {
 						"cache.status": "hit",
 						"cache.storedAt": storedAt,
-						"cache.age": Date.now() - new Date(storedAt).getTime()
+						"cache.age": getRealDateNowMs() - new Date(storedAt).getTime()
 					});
 					const recording = deserializeCacheRecording(hit.recording);
 					replayRecording(scope, spanRecord, recording, { generateSpanId });
@@ -5025,30 +5025,6 @@ function mergeOverrides(base, override) {
 	};
 }
 /**
-* Populate `target` with `ColumnDef` entries for any keys in `columns`
-* that aren't already present, applying user-supplied `overrides` and
-* flagging score columns declared via `scores`.
-*/
-function mergeColumnDefs(target, columns, overrides, scores, manualScores) {
-	const scoreKeys = new Set(Object.keys(scores ?? {}));
-	const manualScoreKeys = new Set(Object.keys(manualScores ?? {}));
-	const overrideMap = overrides ?? {};
-	for (const [key, value] of Object.entries(columns)) {
-		if (target.has(key)) continue;
-		const override = mergeOverrides(getScoreOverride(scores?.[key]) ?? manualScores?.[key], overrideMap[key]);
-		const isScore = scoreKeys.has(key) || manualScoreKeys.has(key);
-		target.set(key, createColumnDef({
-			key,
-			override,
-			scoreDef: scores?.[key],
-			manualScoreDef: manualScores?.[key],
-			inferredKind: isScore ? "number" : inferKind(value),
-			isScore,
-			isManualScore: manualScoreKeys.has(key)
-		}));
-	}
-}
-/**
 * Build the column definitions declared directly on an eval before any runtime
 * output values exist. This lets discovery metadata describe authored rich
 * output columns even for runs created by another process.
@@ -5092,30 +5068,30 @@ function buildDeclaredColumnDefs(overrides, scores, manualScores) {
 	}
 	return [...declaredDefs.values()];
 }
-/** Infer a `ColumnKind` from a runtime value when no override is set. */
-function inferKind(value) {
-	if (typeof value === "number") return "number";
-	if (typeof value === "boolean") return "boolean";
-	return "string";
-}
 /**
 * Coerce an arbitrary runtime value into a serializable `CellValue`.
-* Non-primitive values fall back to `JSON.stringify`.
+* JSON-safe objects and arrays stay structured so saved run artifacts preserve
+* the authored output shape. Rich runtime values fall back to `JSON.stringify`.
 */
-function toCellValue(value, override = void 0) {
+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;
-	if (override?.format === "image" || override?.format === "audio" || override?.format === "video" || override?.format === "file") {
-		const parsed = fileRefSchema.safeParse(value);
-		if (parsed.success) return parsed.data;
-	}
-	if (override?.format === "json") {
+	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;
+}
 function inferKindFromFormat(format) {
 	if (format === "boolean") return "boolean";
 	if (format === "duration" || format === "percent" || format === "number" || format === "passFail" || format === "stars") return "number";
@@ -6586,7 +6562,7 @@ async function runDeriveFromTracingConfig(params) {
 	}
 }
 async function runCase(params) {
-	const { evalDef, evalId, evalKey = evalId, evalCase, globalTraceDisplay, globalColumns, globalDeriveFromTracing, llmCallsConfig = resolveLlmCallsConfig(void 0), apiCallsConfig = resolveApiCallsConfig(void 0), globalRemoveDefaultConfig, trial, startTime, cacheAdapter, cacheMode, moduleIsolation, evalFilePath, evalFileRelativePath = evalFilePath, workspaceRoot, artifactDir, runId } = params;
+	const { evalDef, evalId, evalKey = evalId, evalCase, globalTraceDisplay, globalDeriveFromTracing, llmCallsConfig = resolveLlmCallsConfig(void 0), apiCallsConfig = resolveApiCallsConfig(void 0), globalRemoveDefaultConfig, trial, startTime, cacheAdapter, cacheMode, moduleIsolation, evalFilePath, evalFileRelativePath = evalFilePath, workspaceRoot, artifactDir, runId } = params;
 	const scopedIdPrefix = buildScopedEvalIdPrefix({
 		evalId,
 		evalFilePath,
@@ -6739,12 +6715,6 @@ async function runCase(params) {
 	const status = nonAssertError ? "error" : passed ? "pass" : "fail";
 	const { trace: displayTrace, traceDisplay } = resolveTracePresentation(spansWithDerivedAttributes, globalTraceDisplay, evalDef.traceDisplay);
 	const columns = {};
-	const columnOverrides = mergeDefaultColumns({
-		globalColumns,
-		columns: evalDef.columns,
-		globalRemove: globalRemoveDefaultConfig,
-		evalRemove: evalDef.removeDefaultConfig
-	});
 	for (const [key, value] of Object.entries(scope.outputs)) {
 		const cell = isBlob(value) ? await persistInlineArtifact({
 			artifactDir,
@@ -6753,7 +6723,7 @@ async function runCase(params) {
 			outputKey: key,
 			trial,
 			value
-		}) : toCellValue(value, columnOverrides?.[key]);
+		}) : toCellValue(value);
 		if (cell !== void 0) columns[key] = cell;
 	}
 	for (const key of Object.keys(evalDef.manualScores ?? {})) columns[key] = null;
@@ -6989,7 +6959,6 @@ async function finalizePreparedCase(params) {
 	const artifactFileId = getCaseArtifactFileId(runState, winningTrial.caseRow);
 	runState.cases.push(winningTrial.caseRow);
 	runState.caseDetails.set(getCaseRowCaseKey(winningTrial.caseRow), winningTrial.caseDetail);
-	preparedEval.mergeColumns(winningTrial.caseDetail.columns);
 	if (winningTrial.caseRow.status === "pass") runState.summary.passedCases++;
 	else if (winningTrial.caseRow.status === "error") runState.summary.errorCases++;
 	else runState.summary.failedCases++;
@@ -7106,13 +7075,13 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 								globalRemove: config.removeDefaultConfig
 							});
 							const declaredColumnDefs = buildDeclaredColumnDefs(defaultConfig.columns, evalDef.scores, evalDef.manualScores);
-							const accumulatedColumns = new Map(declaredColumnDefs.map((def) => [def.key, def]));
 							const validatedCharts = validateCharts({
 								charts: defaultConfig.charts,
 								columnDefs: declaredColumnDefs,
 								evalId: evalMeta.id
 							});
 							for (const warning of validatedCharts.warnings) console.warn(warning);
+							evalMeta.columnDefs = declaredColumnDefs;
 							evalMeta.stats = defaultConfig.stats;
 							evalMeta.charts = validatedCharts.charts;
 							const evalCaseRows = [];
@@ -7121,13 +7090,9 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 							const manualScoreKeys = Object.freeze(Object.keys(evalDef.manualScores ?? {}));
 							const preparedEval = {
 								evalMeta,
-								accumulatedColumns,
 								evalCaseRows,
 								preparedCases,
-								scoreKeys: Object.freeze([...scoreKeys, ...manualScoreKeys]),
-								mergeColumns: (columns) => {
-									mergeColumnDefs(accumulatedColumns, columns, defaultConfig.columns, evalDef.scores, evalDef.manualScores);
-								}
+								scoreKeys: Object.freeze([...scoreKeys, ...manualScoreKeys])
 							};
 							preparedEvals.push(preparedEval);
 							for (const evalCase of cases) {
@@ -7236,7 +7201,6 @@ async function executeRun({ runState, request, runDir, config, cacheStore, lastR
 				onCaseFinished,
 				emitEvent
 			});
-			preparedEval.evalMeta.columnDefs = [...preparedEval.accumulatedColumns.values()];
 			lastRunStatusMap.set(preparedEval.evalMeta.key, toLastRunStatus(deriveStatusFromCaseRows({ caseRows: preparedEval.evalCaseRows })));
 			const latestStatus = lastRunStatusMap.get(preparedEval.evalMeta.key) ?? null;
 			latestRunInfoMap.set(preparedEval.evalMeta.key, {

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

@@ -1,2 +1,2 @@
-import { n as initRunner, t as getRunnerInstance } from "./runner-Dkol2ukD.mjs";
+import { n as initRunner, t as getRunnerInstance } from "./runner-j9b6g0h0.mjs";
 export { getRunnerInstance, initRunner };

package/dist/{runner-Dkol2ukD.mjs → runner-j9b6g0h0.mjs}

@@ -1,5 +1,5 @@
-import { n as createRunner } from "./cli-huuJbDNb.mjs";
-import "./src-1Qvuh0NH.mjs";
+import { n as createRunner } from "./cli-CG66f0GN.mjs";
+import "./src-B0b1vgk0.mjs";
 //#region ../../apps/server/src/runner.ts
 let runnerInstance = null;
 function getRunnerInstance() {

package/dist/src-B0b1vgk0.mjs

@@ -0,0 +1,3 @@
+import "./runOrchestration-CYhn4DJ0.mjs";
+import "./cli-CG66f0GN.mjs";
+export {};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.32.0",
+  "version": "0.34.0",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"
@@ -31,9 +31,9 @@
   "devDependencies": {
     "@types/node": "^24.7.2",
     "typescript": "^5.9.2",
-    "@agent-evals/sdk": "0.0.1",
     "@agent-evals/runner": "0.0.1",
-    "@agent-evals/shared": "0.0.1"
+    "@agent-evals/shared": "0.0.1",
+    "@agent-evals/sdk": "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

@@ -299,10 +299,13 @@ See `EvalScoreDef` / `EvalManualScoreDef` in the types for the full shape
 
 - `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. 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.
+  `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
+  `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.
 - Use `incrementEvalOutput(key, delta)` for numeric totals,
   `appendToEvalOutput(key, value)` for arrays that preserve existing scalar
   values, and `mergeEvalOutput(key, patch)` for shallow object updates.