@ls-stack/agent-eval 0.56.0 → 0.57.0

package/dist/index.mjs

@@ -1,4 +1,4 @@
-import { $ as startEvalBackgroundJob, A as repoFile, B as getCurrentScope, C as evalTracer, D as deserializeCacheValue, E as deserializeCacheRecording, F as appendToEvalOutput, G as mergeEvalOutput, H as incrementEvalOutput, J as runInEvalScope, K as nextEvalId, L as evalAssert, M as readManualInputFile, N as evalExpect, Nt as getEvalRegistry, O as serializeCacheRecording, P as EvalAssertionError, Q as setScopeCacheContext, R as evalLog, S as evalSpan, T as hashCacheKeySync, U as isInEvalScope, V as getEvalCaseInput, Y as runInExistingEvalScope, Z as setEvalOutput, at as extractLlmCalls, b as buildTraceTree, it as extractApiCalls, j as manualInputFileValueSchema, k as serializeCacheValue, lt as getNestedAttribute, nt as extractCacheEntries, ot as simulateLlmCallCost, q as runInEvalRuntimeScope, rt as extractCacheHits, st as simulateTokenAllocation, w as hashCacheKey, x as captureEvalSpanError, y as z, z as evalTime } from "./runExecution-6lrtj48K.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-DQO2Fpt2.mjs";
-import { n as matchesEvalTags, t as defineEval } from "./src-DCGrFAmO.mjs";
+import { $ as startEvalBackgroundJob, A as repoFile, B as getCurrentScope, C as evalTracer, D as deserializeCacheValue, E as deserializeCacheRecording, F as appendToEvalOutput, G as mergeEvalOutput, H as incrementEvalOutput, J as runInEvalScope, K as nextEvalId, L as evalAssert, M as readManualInputFile, N as evalExpect, Nt as getEvalRegistry, O as serializeCacheRecording, P as EvalAssertionError, Q as setScopeCacheContext, R as evalLog, S as evalSpan, T as hashCacheKeySync, U as isInEvalScope, V as getEvalCaseInput, Y as runInExistingEvalScope, Z as setEvalOutput, at as extractLlmCalls, b as buildTraceTree, it as extractApiCalls, j as manualInputFileValueSchema, k as serializeCacheValue, lt as getNestedAttribute, nt as extractCacheEntries, ot as simulateLlmCallCost, q as runInEvalRuntimeScope, rt as extractCacheHits, st as simulateTokenAllocation, w as hashCacheKey, x as captureEvalSpanError, y as z, z as evalTime } from "./runExecution-BH7DlMXl.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-Ck0mqxd-.mjs";
+import { n as matchesEvalTags, t as defineEval } from "./src-B3iq-tuv.mjs";
 export { EvalAssertionError, appendToEvalOutput, buildTraceTree, captureEvalSpanError, cleanupStagedManualInputFiles, createRunner, defineEval, deserializeCacheRecording, deserializeCacheValue, evalAssert, evalExpect, evalLog, evalSpan, evalTime, evalTracer, extractApiCalls, extractCacheEntries, extractCacheHits, extractLlmCalls, getCurrentScope, getEvalCaseInput, getEvalRegistry, getNestedAttribute, hashCacheKey, hashCacheKeySync, incrementEvalOutput, isInEvalScope, isManualInputFileValue, manualInputFileValueSchema, matchesEvalTags, materializeManualInputFiles, mergeEvalOutput, nextEvalId, readManualInputFile, repoFile, runCli, runInEvalRuntimeScope, runInEvalScope, runInExistingEvalScope, serializeCacheRecording, serializeCacheValue, setEvalOutput, setScopeCacheContext, simulateLlmCallCost, simulateTokenAllocation, stageManualInputFile, stageManualInputFileFromPath, startEvalBackgroundJob, z };

package/dist/runChild.mjs

@@ -1,5 +1,5 @@
-import { At as evalChartsConfigSchema, Ct as buildEvalKey, Dt as evalStatAggregateSchema, I as configureEvalRunLogs, Ot as evalStatsConfigSchema, bt as runSummarySchema, et as createRunRequestSchema, jt as columnDefSchema, kt as manualInputDescriptorSchema, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, v as createFsCacheStore, yt as runManifestSchema } from "./runExecution-6lrtj48K.mjs";
-import { S as parseEvalDiscovery, m as persistRunState, r as getTargetEvals$1, t as executeRun } from "./runOrchestration-BYaN2mzS.mjs";
+import { At as evalChartsConfigSchema, Ct as buildEvalKey, Dt as evalStatAggregateSchema, I as configureEvalRunLogs, Ot as evalStatsConfigSchema, bt as runSummarySchema, et as createRunRequestSchema, jt as columnDefSchema, kt as manualInputDescriptorSchema, l as registerAgentEvalsPackageResolutionHooks, p as loadConfig, v as createFsCacheStore, yt as runManifestSchema } from "./runExecution-BH7DlMXl.mjs";
+import { S as parseEvalDiscovery, m as persistRunState, r as getTargetEvals$1, t as executeRun } from "./runOrchestration-C1Ex9QI-.mjs";
 import { z } from "zod/v4";
 import { readFile } from "node:fs/promises";
 import { relative } from "node:path";

package/dist/{runExecution-6lrtj48K.mjs → runExecution-BH7DlMXl.mjs}

@@ -6181,6 +6181,10 @@ function sumNullable(values) {
 	}
 	return hasValue ? total : void 0;
 }
+function getMaxLlmTurns(calls) {
+	if (calls.length === 0) return void 0;
+	return Math.max(...calls.map((call) => Math.max(call.stepCount ?? 1, 1)));
+}
 function assignIfMissing(params) {
 	if (!params.activeKeys.has(params.key)) return;
 	if (params.key in params.outputs) return;
@@ -6202,7 +6206,7 @@ function addDefaultOutputs(params) {
 	assignIfMissing({
 		outputs: params.outputs,
 		key: "llmTurns",
-		value: calls.length,
+		value: getMaxLlmTurns(calls),
 		activeKeys
 	});
 	assignIfMissing({

package/dist/{runOrchestration-BYaN2mzS.mjs → runOrchestration-C1Ex9QI-.mjs}

@@ -1,4 +1,4 @@
-import { Et as caseRowSchema, Pt as runWithEvalRegistry, Tt as caseDetailSchema, X as runWithEvalClock, _t as validateEvalTagName, bt as runSummarySchema, d as loadEvalModule, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, g as commitPendingCacheWrites, gt as matchesTagsFilter, ht as dedupeEvalTags, i as isCaseChildMessage, m as buildDeclaredColumnDefs, mt as deriveStatusFromChildStatuses, n as resolveRunnableEvalCases, o as stripTerminalControlCodes, pt as deriveStatusFromCaseRows, q as runInEvalRuntimeScope, t as filterEvalCases, u as runWithModuleIsolation, vt as validateTagsFilterExpression, wt as getCaseRowCaseKey, yt as runManifestSchema } from "./runExecution-6lrtj48K.mjs";
+import { Et as caseRowSchema, Pt as runWithEvalRegistry, Tt as caseDetailSchema, X as runWithEvalClock, _t as validateEvalTagName, bt as runSummarySchema, d as loadEvalModule, f as resolveEvalDefaultConfig, ft as deriveScopedSummaryFromCases, g as commitPendingCacheWrites, gt as matchesTagsFilter, ht as dedupeEvalTags, i as isCaseChildMessage, m as buildDeclaredColumnDefs, mt as deriveStatusFromChildStatuses, n as resolveRunnableEvalCases, o as stripTerminalControlCodes, pt as deriveStatusFromCaseRows, q as runInEvalRuntimeScope, t as filterEvalCases, u as runWithModuleIsolation, vt as validateTagsFilterExpression, wt as getCaseRowCaseKey, yt as runManifestSchema } from "./runExecution-BH7DlMXl.mjs";
 import { readFile, readdir, rm, writeFile } from "node:fs/promises";
 import { dirname, join } from "node:path";
 import { existsSync } from "node:fs";

package/dist/{runner-DYlwuAT3.mjs → runner-B3hEOT_I.mjs}

@@ -1,5 +1,5 @@
-import { n as createRunner } from "./cli-DQO2Fpt2.mjs";
-import "./src-DCGrFAmO.mjs";
+import { n as createRunner } from "./cli-Ck0mqxd-.mjs";
+import "./src-B3iq-tuv.mjs";
 //#region ../../apps/server/src/runner.ts
 let runnerInstance = null;
 function getRunnerInstance() {

package/dist/{runner-C3CiS2o7.mjs → runner-DbVYcapC.mjs}

@@ -1,2 +1,2 @@
-import { n as initRunner, t as getRunnerInstance } from "./runner-DYlwuAT3.mjs";
+import { n as initRunner, t as getRunnerInstance } from "./runner-B3hEOT_I.mjs";
 export { getRunnerInstance, initRunner };

package/dist/{src-DCGrFAmO.mjs → src-B3iq-tuv.mjs}

@@ -1,5 +1,5 @@
-import { Mt as defineEval$1, W as matchesEvalTags$1 } from "./runExecution-6lrtj48K.mjs";
-import "./cli-DQO2Fpt2.mjs";
+import { Mt as defineEval$1, W as matchesEvalTags$1 } from "./runExecution-BH7DlMXl.mjs";
+import "./cli-Ck0mqxd-.mjs";
 //#region src/index.ts
 /** Register an eval definition with typed tag support. */
 function defineEval(definition) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.56.0",
+  "version": "0.57.0",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"

package/skills/agent-eval/SKILL.md

@@ -393,7 +393,9 @@ See `EvalScoreDef` / `EvalManualScoreDef` in the types for the full shape
   tokens, and output tokens separately and use `dedupeConsecutiveValues: true`
   to skip repeated adjacent chart values. `totalTokens` is input + output only;
   cache read/write tokens stay separate and affect `costUsd` at their own
-  rates.
+  rates. `llmTurns` is the maximum per-call turn count in the case run, using
+  configured steps when available and otherwise one turn per matched LLM call
+  span.
   Derived base input cost uses `inputTokens - cachedInputTokens -
 cacheCreationInputTokens` so cache details are not double-counted.
   `cacheCreationInputTokens` is the total cache-write count; optional
@@ -568,6 +570,8 @@ Targeted evals in `run.json` are recorded by exact `evalKeys`
 unambiguous in saved history.
 Temporary runs use the same directory layout, but are removed before the next
 run of any kind starts.
+When a saved case needs to be handed to another agent, the app can copy the
+saved case detail path or the saved run folder path directly.
 
 Use `agent-evals show-runs` when you need stable file
 paths before reading saved output: