npm - @ls-stack/agent-eval - Versions diffs - 0.26.2 → 0.26.3 - Mend

@ls-stack/agent-eval 0.26.2 → 0.26.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (15) hide show

package/dist/{app-BxR4SbjA.mjs → app-C-yzV9J8.mjs} +3 -3
package/dist/apps/web/dist/assets/{index-CFF1eYUm.js → index-DR2haqvV.js} +1 -1
package/dist/apps/web/dist/index.html +1 -1
package/dist/bin.mjs +1 -1
package/dist/{cli-C4yumCXE.mjs → cli-T6FYE2Bk.mjs} +3 -3
package/dist/index.d.mts +3 -4
package/dist/index.mjs +3 -3
package/dist/runChild.mjs +1 -1
package/dist/{runOrchestration-B31SV_Bq.mjs → runOrchestration-CW_Fzr2c.mjs} +10 -6
package/dist/{runner-LHN7L-xk.mjs → runner-BsRloAm3.mjs} +2 -2
package/dist/{runner-D5QMY3B_.mjs → runner-DesCrD7Z.mjs} +1 -1
package/dist/src-CbVnqjW3.mjs +3 -0
package/package.json +1 -1
package/skills/agent-eval/SKILL.md +20 -13
package/dist/src-CLMrRle2.mjs +0 -3

package/dist/apps/web/dist/index.html CHANGED Viewed

@@ -25,7 +25,7 @@
       href="https://fonts.googleapis.com/css2?family=Geist:wght@400;500;600&family=JetBrains+Mono:wght@400;500&display=swap"
       rel="stylesheet"
     />
-    <script type="module" crossorigin src="/assets/index-CFF1eYUm.js"></script>
+    <script type="module" crossorigin src="/assets/index-DR2haqvV.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-DOXT0Y9V.css">
   </head>
   <body>

package/dist/bin.mjs CHANGED Viewed

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { t as runCli } from "./cli-C4yumCXE.mjs";
+import { t as runCli } from "./cli-T6FYE2Bk.mjs";
 import { spawn } from "node:child_process";
 //#region src/bin.ts
 const moduleMocksFlag = "--experimental-test-module-mocks";

package/dist/{cli-C4yumCXE.mjs → cli-T6FYE2Bk.mjs} RENAMED Viewed

@@ -1,4 +1,4 @@
-import { B as runSummarySchema, C as validateCharts, F as getEvalDisplayStatus, I as deriveScopedSummaryFromCases, P as getEvalTitle, S as normalizeScoreDef, Un as getEvalRegistry, _ as loadEvalModule, a as getLastRunStatuses, b as loadConfig, c as loadPersistedRunSnapshots, ct as getCaseRowEvalKey, d as persistRunState, f as recomputeEvalStatusesInRuns, g as deriveEvalFreshness, h as resolveArtifactPath, i as generateRunId, l as nextShortIdFromSnapshots, m as runTouchesEval, n as getTargetEvalKeys, nt as resolveLlmCallsConfig, o as getLatestRunInfos, ot as buildEvalKey, p as recomputePersistedCaseStatus, s as loadPersistedRunSnapshot, st as getCaseRowCaseKey, tt as resolveApiCallsConfig, u as persistCaseDetail, v as parseEvalDiscovery, w as createFsCacheStore, x as buildDeclaredColumnDefs, y as resolveEvalDefaultConfig } from "./runOrchestration-B31SV_Bq.mjs";
+import { B as runSummarySchema, C as validateCharts, F as getEvalDisplayStatus, I as deriveScopedSummaryFromCases, P as getEvalTitle, S as normalizeScoreDef, Un as getEvalRegistry, _ as loadEvalModule, a as getLastRunStatuses, b as loadConfig, c as loadPersistedRunSnapshots, ct as getCaseRowEvalKey, d as persistRunState, f as recomputeEvalStatusesInRuns, g as deriveEvalFreshness, h as resolveArtifactPath, i as generateRunId, l as nextShortIdFromSnapshots, m as runTouchesEval, n as getTargetEvalKeys, nt as resolveLlmCallsConfig, o as getLatestRunInfos, ot as buildEvalKey, p as recomputePersistedCaseStatus, s as loadPersistedRunSnapshot, st as getCaseRowCaseKey, tt as resolveApiCallsConfig, u as persistCaseDetail, v as parseEvalDiscovery, w as createFsCacheStore, x as buildDeclaredColumnDefs, y as resolveEvalDefaultConfig } from "./runOrchestration-CW_Fzr2c.mjs";
 import { createHash } from "node:crypto";
 import { mkdir, readFile, rm, writeFile } from "node:fs/promises";
 import { dirname, join, relative, resolve } from "node:path";
@@ -1041,8 +1041,8 @@ async function commandApp(args) {
 	const { serve } = await import("@hono/node-server");
 	const bundledWebDist = resolve(currentDir, "apps/web/dist");
 	if (existsSync(bundledWebDist)) process.env.AGENT_EVALS_WEB_DIST = bundledWebDist;
-	const appModule = await import("./app-BxR4SbjA.mjs");
-	const runnerModule = await import("./runner-D5QMY3B_.mjs");
+	const appModule = await import("./app-C-yzV9J8.mjs");
+	const runnerModule = await import("./runner-DesCrD7Z.mjs");
 	if (!isHonoAppModule(appModule)) throw new Error("Server app module is invalid");
 	if (!isServerRunnerModule(runnerModule)) throw new Error("Server runner module is invalid");
 	await runnerModule.initRunner();

package/dist/index.d.mts CHANGED Viewed

@@ -2288,7 +2288,7 @@ type CacheMode = z$1.infer<typeof cacheModeSchema>;
 /** Options accepted by an `evalTracer.span` call to opt the span into caching. */
 declare const spanCacheOptionsSchema: z$1.ZodObject<{
   key: z$1.ZodUnknown;
-  namespace: z$1.ZodOptional<z$1.ZodString>;
+  namespace: z$1.ZodString;
   serializeFileBytes: z$1.ZodOptional<z$1.ZodBoolean>;
 }, z$1.core.$strip>;
 /** Options accepted by an `evalTracer.span` call to opt the span into caching. */
@@ -3168,7 +3168,7 @@ type CacheAdapter = {
 type CacheScopeContext = {
   adapter: CacheAdapter;
   mode: CacheMode;
-  evalId: string; /** Hash of the eval source file; used to invalidate on code changes. */
+  evalId: string; /** Hash of the eval source file, stored as cache metadata for inspection. */
   codeFingerprint: string;
 };
 /** Active recording frame captured while a cached operation body executes. */
@@ -3419,8 +3419,7 @@ type TraceCacheInfo = {
 //#region ../sdk/src/cacheKey.d.ts
 /** Components folded into a deterministic cache key hash. */
 type CacheKeyHashInput = {
-  /** Cache namespace, usually derived from the eval id and operation name. */namespace: string; /** Eval source fingerprint used to invalidate cache entries on code edits. */
-  codeFingerprint: string; /** User-authored cache key value. */
+  /** Cache namespace, usually derived from the eval id and operation name. */namespace: string; /** User-authored cache key value. */
   key: unknown;
 };
 /** Optional controls for cache key hashing. */

package/dist/index.mjs CHANGED Viewed

@@ -1,4 +1,4 @@
-import { $ as llmCallsConfigSchema, $t as traceSpanKindSchema, A as extractApiCalls, An as getEvalStartTime, At as evalChartTypeSchema, B as runSummarySchema, Bn as startEvalBackgroundJob, Bt as cacheRecordingOpSchema, Cn as advanceEvalTime, Ct as evalChartAggregateSchema, D as sseEnvelopeSchema, Dn as evalLog, Dt as evalChartConfigSchema, E as updateManualScoreRequestSchema, En as evalAssert, Et as evalChartColorSchema, F as getEvalDisplayStatus, Fn as runInEvalRuntimeScope, Ft as cacheEntryWithDebugKeySchema, G as apiCallMetricPlacementSchema, Gt as traceCacheRefSchema, H as DEFAULT_LLM_CALLS_CONFIG, Hn as defineEval, Ht as cacheStatusSchema, I as deriveScopedSummaryFromCases, In as runInEvalScope, It as cacheFileSchema, J as defaultConfigKeySchema, Jt as traceAttributeDisplayPlacementSchema, K as apiCallMetricSchema, Kt as traceAttributeDisplayFormatSchema, L as deriveStatusFromCaseRows, Ln as runInExistingEvalScope, Lt as cacheListItemSchema, M as applyDerivedCallAttributes, Mn as isInEvalScope, Mt as cacheDebugKeyEntrySchema, N as getNestedAttribute, Nn as mergeEvalOutput, Nt as cacheDebugKeyFileSchema, O as extractCacheEntries, On as getCurrentScope, Ot as evalChartMetricSchema, P as getEvalTitle, Pn as nextEvalId, Pt as cacheEntrySchema, Q as llmCallPricingSchema, Qt as traceSpanErrorSchema, R as deriveStatusFromChildStatuses, Rn as setEvalOutput, Rt as cacheModeSchema, Sn as EvalAssertionError, St as scoreTraceSchema, T as createRunRequestSchema, Tt as evalChartBuiltinMetricSchema, U as agentEvalsConfigSchema, Un as getEvalRegistry, Ut as serializedCacheSpanSchema, V as DEFAULT_API_CALLS_CONFIG, Vn as repoFile, Vt as cacheRecordingSchema, W as apiCallMetricFormatSchema, Wt as spanCacheOptionsSchema, X as llmCallMetricPlacementSchema, Xt as traceDisplayConfigSchema, Y as llmCallMetricFormatSchema, Yt as traceAttributeDisplaySchema, Z as llmCallMetricSchema, Zt as traceDisplayInputConfigSchema, _n as hashCacheKeySync, _t as evalSummarySchema, an as columnKindSchema, at as buildCaseKey, bn as serializeCacheRecording, bt as runLogLocationSchema, cn as numberDisplayOptionsSchema, ct as getCaseRowEvalKey, dn as z, dt as caseRowSchema, en as traceSpanSchema, et as removeDefaultConfigSchema, fn as buildTraceTree, ft as discoveryIssueSchema, gn as hashCacheKey, gt as evalStatsConfigSchema, hn as evalTracer, ht as evalStatItemSchema, in as columnFormatSchema, it as trialSelectionModeSchema, j as extractLlmCalls, jn as incrementEvalOutput, jt as evalChartsConfigSchema, k as extractCacheHits, kn as getEvalCaseInput, kt as evalChartTooltipExtraSchema, ln as repoFileRefSchema, lt as assertionFailureSchema, mn as evalSpan, mt as evalStatAggregateSchema, nn as cellValueSchema, nt as resolveLlmCallsConfig, on as fileRefSchema, ot as buildEvalKey, pn as captureEvalSpanError, pt as evalFreshnessStatusSchema, q as apiCallsConfigSchema, qt as traceAttributeDisplayInputSchema, rn as columnDefSchema, rt as runLogsConfigSchema, sn as jsonCellSchema, st as getCaseRowCaseKey, tn as traceSpanWarningSchema, tt as resolveApiCallsConfig, un as runArtifactRefSchema, ut as caseDetailSchema, vn as deserializeCacheRecording, vt as runLogEntrySchema, wn as appendToEvalOutput, wt as evalChartAxisSchema, xn as serializeCacheValue, xt as runLogPhaseSchema, yn as deserializeCacheValue, yt as runLogLevelSchema, z as runManifestSchema, zn as setScopeCacheContext, zt as cacheOperationTypeSchema } from "./runOrchestration-B31SV_Bq.mjs";
-import { n as createRunner, t as runCli } from "./cli-C4yumCXE.mjs";
-import "./src-CLMrRle2.mjs";
+import { $ as llmCallsConfigSchema, $t as traceSpanKindSchema, A as extractApiCalls, An as getEvalStartTime, At as evalChartTypeSchema, B as runSummarySchema, Bn as startEvalBackgroundJob, Bt as cacheRecordingOpSchema, Cn as advanceEvalTime, Ct as evalChartAggregateSchema, D as sseEnvelopeSchema, Dn as evalLog, Dt as evalChartConfigSchema, E as updateManualScoreRequestSchema, En as evalAssert, Et as evalChartColorSchema, F as getEvalDisplayStatus, Fn as runInEvalRuntimeScope, Ft as cacheEntryWithDebugKeySchema, G as apiCallMetricPlacementSchema, Gt as traceCacheRefSchema, H as DEFAULT_LLM_CALLS_CONFIG, Hn as defineEval, Ht as cacheStatusSchema, I as deriveScopedSummaryFromCases, In as runInEvalScope, It as cacheFileSchema, J as defaultConfigKeySchema, Jt as traceAttributeDisplayPlacementSchema, K as apiCallMetricSchema, Kt as traceAttributeDisplayFormatSchema, L as deriveStatusFromCaseRows, Ln as runInExistingEvalScope, Lt as cacheListItemSchema, M as applyDerivedCallAttributes, Mn as isInEvalScope, Mt as cacheDebugKeyEntrySchema, N as getNestedAttribute, Nn as mergeEvalOutput, Nt as cacheDebugKeyFileSchema, O as extractCacheEntries, On as getCurrentScope, Ot as evalChartMetricSchema, P as getEvalTitle, Pn as nextEvalId, Pt as cacheEntrySchema, Q as llmCallPricingSchema, Qt as traceSpanErrorSchema, R as deriveStatusFromChildStatuses, Rn as setEvalOutput, Rt as cacheModeSchema, Sn as EvalAssertionError, St as scoreTraceSchema, T as createRunRequestSchema, Tt as evalChartBuiltinMetricSchema, U as agentEvalsConfigSchema, Un as getEvalRegistry, Ut as serializedCacheSpanSchema, V as DEFAULT_API_CALLS_CONFIG, Vn as repoFile, Vt as cacheRecordingSchema, W as apiCallMetricFormatSchema, Wt as spanCacheOptionsSchema, X as llmCallMetricPlacementSchema, Xt as traceDisplayConfigSchema, Y as llmCallMetricFormatSchema, Yt as traceAttributeDisplaySchema, Z as llmCallMetricSchema, Zt as traceDisplayInputConfigSchema, _n as hashCacheKeySync, _t as evalSummarySchema, an as columnKindSchema, at as buildCaseKey, bn as serializeCacheRecording, bt as runLogLocationSchema, cn as numberDisplayOptionsSchema, ct as getCaseRowEvalKey, dn as z, dt as caseRowSchema, en as traceSpanSchema, et as removeDefaultConfigSchema, fn as buildTraceTree, ft as discoveryIssueSchema, gn as hashCacheKey, gt as evalStatsConfigSchema, hn as evalTracer, ht as evalStatItemSchema, in as columnFormatSchema, it as trialSelectionModeSchema, j as extractLlmCalls, jn as incrementEvalOutput, jt as evalChartsConfigSchema, k as extractCacheHits, kn as getEvalCaseInput, kt as evalChartTooltipExtraSchema, ln as repoFileRefSchema, lt as assertionFailureSchema, mn as evalSpan, mt as evalStatAggregateSchema, nn as cellValueSchema, nt as resolveLlmCallsConfig, on as fileRefSchema, ot as buildEvalKey, pn as captureEvalSpanError, pt as evalFreshnessStatusSchema, q as apiCallsConfigSchema, qt as traceAttributeDisplayInputSchema, rn as columnDefSchema, rt as runLogsConfigSchema, sn as jsonCellSchema, st as getCaseRowCaseKey, tn as traceSpanWarningSchema, tt as resolveApiCallsConfig, un as runArtifactRefSchema, ut as caseDetailSchema, vn as deserializeCacheRecording, vt as runLogEntrySchema, wn as appendToEvalOutput, wt as evalChartAxisSchema, xn as serializeCacheValue, xt as runLogPhaseSchema, yn as deserializeCacheValue, yt as runLogLevelSchema, z as runManifestSchema, zn as setScopeCacheContext, zt as cacheOperationTypeSchema } from "./runOrchestration-CW_Fzr2c.mjs";
+import { n as createRunner, t as runCli } from "./cli-T6FYE2Bk.mjs";
+import "./src-CbVnqjW3.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, columnDefSchema, columnFormatSchema, columnKindSchema, createRunRequestSchema, createRunner, defaultConfigKeySchema, defineEval, deriveScopedSummaryFromCases, deriveStatusFromCaseRows, deriveStatusFromChildStatuses, deserializeCacheRecording, deserializeCacheValue, discoveryIssueSchema, evalAssert, evalChartAggregateSchema, evalChartAxisSchema, evalChartBuiltinMetricSchema, evalChartColorSchema, evalChartConfigSchema, evalChartMetricSchema, evalChartTooltipExtraSchema, evalChartTypeSchema, evalChartsConfigSchema, 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, jsonCellSchema, llmCallMetricFormatSchema, llmCallMetricPlacementSchema, llmCallMetricSchema, llmCallPricingSchema, llmCallsConfigSchema, mergeEvalOutput, nextEvalId, numberDisplayOptionsSchema, removeDefaultConfigSchema, repoFile, repoFileRefSchema, resolveApiCallsConfig, resolveLlmCallsConfig, runArtifactRefSchema, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, runLogEntrySchema, runLogLevelSchema, runLogLocationSchema, runLogPhaseSchema, runLogsConfigSchema, runManifestSchema, runSummarySchema, scoreTraceSchema, serializeCacheRecording, serializeCacheValue, serializedCacheSpanSchema, setEvalOutput, setScopeCacheContext, spanCacheOptionsSchema, sseEnvelopeSchema, startEvalBackgroundJob, traceAttributeDisplayFormatSchema, traceAttributeDisplayInputSchema, traceAttributeDisplayPlacementSchema, traceAttributeDisplaySchema, traceCacheRefSchema, traceDisplayConfigSchema, traceDisplayInputConfigSchema, traceSpanErrorSchema, traceSpanKindSchema, traceSpanSchema, traceSpanWarningSchema, trialSelectionModeSchema, updateManualScoreRequestSchema, z };

package/dist/runChild.mjs CHANGED Viewed

@@ -1,4 +1,4 @@
-import { B as runSummarySchema, T as createRunRequestSchema, Tn as configureEvalRunLogs, b as loadConfig, gt as evalStatsConfigSchema, jt as evalChartsConfigSchema, ot as buildEvalKey, r as getTargetEvals$1, rn as columnDefSchema, t as executeRun, v as parseEvalDiscovery, w as createFsCacheStore, z as runManifestSchema } from "./runOrchestration-B31SV_Bq.mjs";
+import { B as runSummarySchema, T as createRunRequestSchema, Tn as configureEvalRunLogs, b as loadConfig, gt as evalStatsConfigSchema, jt as evalChartsConfigSchema, ot as buildEvalKey, r as getTargetEvals$1, rn as columnDefSchema, t as executeRun, v as parseEvalDiscovery, w as createFsCacheStore, z as runManifestSchema } from "./runOrchestration-CW_Fzr2c.mjs";
 import { createHash } from "node:crypto";
 import { readFile } from "node:fs/promises";
 import { relative } from "node:path";

package/dist/{runOrchestration-B31SV_Bq.mjs → runOrchestration-CW_Fzr2c.mjs} RENAMED Viewed

@@ -1436,7 +1436,6 @@ function createTraceCache(generateSpanId) {
 		const namespace = info.namespace ?? `${cacheCtx.evalId}__${info.name}`;
 		const keyHash = await hashCacheKey({
 			namespace,
-			codeFingerprint: cacheCtx.codeFingerprint,
 			key: info.key
 		}, { serializeFileBytes: info.serializeFileBytes === true });
 		const activeSpan = scope.activeSpanStack.at(-1);
@@ -1879,10 +1878,9 @@ async function traceSpanInternal(info, fn) {
 		const cacheCtx = scope.cacheContext;
 		if (cacheOpts !== void 0 && cacheCtx !== void 0 && scope.replayingDepth === 0) {
 			const ctx = cacheCtx;
-			const namespace = cacheOpts.namespace ?? `${ctx.evalId}__${info.name}`;
+			const namespace = getRequiredSpanCacheNamespace(cacheOpts);
 			const keyHash = await hashCacheKey({
 				namespace,
-				codeFingerprint: ctx.codeFingerprint,
 				key: cacheOpts.key
 			}, { serializeFileBytes: cacheOpts.serializeFileBytes === true });
 			mergeSpanAttributes(spanRecord, {
@@ -1966,6 +1964,12 @@ async function traceSpanInternal(info, fn) {
 		scope.activeSpanStack.pop();
 	}
 }
+function getRequiredSpanCacheNamespace(cacheOpts) {
+	if (!isRecordLike$1(cacheOpts)) throw new Error("Cached spans require a non-empty cache.namespace");
+	const namespace = cacheOpts.namespace;
+	if (typeof namespace !== "string" || namespace.length === 0) throw new Error("Cached spans require a non-empty cache.namespace");
+	return namespace;
+}
 /**
 * Trace builder used to create hierarchical spans and checkpoints during eval
 * execution.
@@ -2248,8 +2252,8 @@ const cacheModeSchema = z.enum([
 const spanCacheOptionsSchema = z.object({
 	/** Arbitrary JSON-safe value used to derive the cache key. */
 	key: z.unknown(),
-	/** Override the default namespace (`${evalId}__${spanName}`). */
-	namespace: z.string().optional(),
+	/** Required cache namespace shared by span cache entries in the same domain. */
+	namespace: z.string().min(1),
 	/**
 	* Include native `Blob`/`File` bytes in the cache key. By default only stable
 	* metadata (`type`, `size`, plus `name`/`lastModified` for `File`) is used.
@@ -4197,7 +4201,7 @@ async function writeDebugKeyFile(debugDir, debugFile) {
 	await mkdir(debugDir, { recursive: true });
 	const filePath = ownerPath(debugDir, debugFile.owner);
 	const tmpPath = `${filePath}.${process.pid.toString()}.tmp`;
-	await writeFile(tmpPath, JSON.stringify(debugFile));
+	await writeFile(tmpPath, JSON.stringify(debugFile, null, 2));
 	await rename(tmpPath, filePath);
 }
 function pruneEntries(entries, namespace, maxEntries, protectedKey) {

package/dist/{runner-LHN7L-xk.mjs → runner-BsRloAm3.mjs} RENAMED Viewed

@@ -1,5 +1,5 @@
-import { n as createRunner } from "./cli-C4yumCXE.mjs";
-import "./src-CLMrRle2.mjs";
+import { n as createRunner } from "./cli-T6FYE2Bk.mjs";
+import "./src-CbVnqjW3.mjs";
 //#region ../../apps/server/src/runner.ts
 let runnerInstance = null;
 function getRunnerInstance() {

package/dist/{runner-D5QMY3B_.mjs → runner-DesCrD7Z.mjs} RENAMED Viewed

@@ -1,2 +1,2 @@
-import { n as initRunner, t as getRunnerInstance } from "./runner-LHN7L-xk.mjs";
+import { n as initRunner, t as getRunnerInstance } from "./runner-BsRloAm3.mjs";
 export { getRunnerInstance, initRunner };

package/dist/src-CbVnqjW3.mjs ADDED Viewed

@@ -0,0 +1,3 @@
+import "./runOrchestration-CW_Fzr2c.mjs";
+import "./cli-T6FYE2Bk.mjs";
+export {};

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.26.2",
+  "version": "0.26.3",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"

package/skills/agent-eval/SKILL.md CHANGED Viewed

@@ -87,7 +87,10 @@ export async function runRefundWorkflow(input: RefundInput) {
         {
           kind: 'llm',
           name: 'plan-refund',
-          cache: { key: { prompt: input.message, model: 'gpt-4o-mini' } },
+          cache: {
+            namespace: 'refund-workflow__plan-refund',
+            key: { prompt: input.message, model: 'gpt-4o-mini' },
+          },
         },
         async () => {
           let text: string;
@@ -319,15 +322,18 @@ Their shapes live in the types; no need to memorize the option set.
 ## Cached operations
-Wrap a costly pure span in `cache: { key }` so later runs replay its recorded
-effects without re-executing:
+Wrap a costly pure span in `cache: { namespace, key }` so later runs replay its
+recorded effects without re-executing:
 ```ts
 await evalTracer.span(
   {
     kind: 'llm',
     name: 'plan-refund',
-    cache: { key: { prompt: input.message, model: 'gpt-4o-mini' } },
+    cache: {
+      namespace: 'refund-workflow__plan-refund',
+      key: { prompt: input.message, model: 'gpt-4o-mini' },
+    },
   },
   async () => {
     const result = await llm.complete(input.message);
@@ -368,12 +374,13 @@ Mental model:
   namespace, and hit/miss status. When called directly from the case body
   (no surrounding span), the ref is recorded on the case detail's `cacheRefs`
   array.
-- The cache key folds in a source-file fingerprint, so editing the eval busts
-  the cache automatically.
-- `cache.namespace` on spans or `namespace` on value caches can share entries
-  across operations/evals, but the source-file fingerprint still participates
-  in the final key. Shared namespaces are reusable across evals in the same
-  file; evals in different files miss even with the same namespace and key.
+- Cache identity is the namespace plus the authored key. Source-file
+  fingerprints are stored as metadata for inspection, but do not participate in
+  cache-key hashing.
+- Cached spans require an explicit `cache.namespace`; value caches default to
+  `${evalId}__${name}` and can be overridden with `namespace`. Matching
+  namespaces share entries across operations/evals that use the same authored
+  key.
 - Authored eval ids are unique within one eval file. The exact eval identity is
   the workspace-relative file path plus eval id, so the same id can be reused in
   different files. Case ids must be unique within one eval; duplicate case ids
@@ -462,9 +469,9 @@ When adding or changing evals:
    formats from the `ColumnFormat` type.
 5. Promote high-signal span attributes with `traceDisplay` so they surface in
    the trace tree and detail pane.
-6. Cache costly pure spans with `cache: { key }` and pure spanless values with
-   `evalTracer.cache(...)`; never cache operations whose external side effects
-   you depend on.
+6. Cache costly pure spans with `cache: { namespace, key }` and pure spanless
+   values with `evalTracer.cache(...)`; never cache operations whose external
+   side effects you depend on.
 7. Sanity-check after changes: `agent-evals list`, then
    `agent-evals run --eval <id>`; use `--file <path|glob>` to target one file
    when multiple files use the same eval id.

package/dist/src-CLMrRle2.mjs DELETED Viewed

@@ -1,3 +0,0 @@
-import "./runOrchestration-B31SV_Bq.mjs";
-import "./cli-C4yumCXE.mjs";
-export {};