@ls-stack/agent-eval 0.17.0 → 0.19.0

package/dist/{runner-DyM0Gp8G.mjs → runner-DxlahWDo.mjs}

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

package/dist/{runner-BSXZiQIi.mjs → runner-RmZPRz-h.mjs}

@@ -1,5 +1,5 @@
-import { n as createRunner } from "./cli-CULTt3Xp.mjs";
-import "./src-CNf3xwVw.mjs";
+import { n as createRunner } from "./cli-3zANEAhG.mjs";
+import "./src-BC4OrajN.mjs";
 //#region ../../apps/server/src/runner.ts
 let runnerInstance = null;
 function getRunnerInstance() {

package/dist/src-BC4OrajN.mjs

@@ -0,0 +1,3 @@
+import "./runOrchestration-BBg_VUH5.mjs";
+import "./cli-3zANEAhG.mjs";
+export {};

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ls-stack/agent-eval",
-  "version": "0.17.0",
+  "version": "0.19.0",
   "type": "module",
   "bin": {
     "agent-evals": "./dist/bin.mjs"

package/skills/agent-eval/SKILL.md

@@ -55,8 +55,9 @@ a per-case sequence number, and throws outside an eval case scope.
 Use `evalLog(level, ...args)` for intentional per-case logs. The runner also
 captures `console.log`, `console.info`, `console.warn`, and `console.error`
 during case-owned phases by default; log arguments are stored as JSON-safe
-values and rendered with the JSON viewer, collapsed previews are capped, and
-logs inside cached operations are not replayed from cache hits.
+values and rendered with the JSON viewer, collapsed previews include best-effort
+code locations when stack data is available, previews are capped, and logs
+inside cached operations are not replayed from cache hits.
 
 ### Product code (instrumented once, reused everywhere)
 
@@ -158,7 +159,9 @@ events, use `evalTracer.startSpan(...)`, `evalTracer.updateSpan(...)`,
 `evalTracer.endSpan(...)`, or `evalTracer.recordSpan(...)` to translate those
 events into the eval trace tree without wrapping the upstream work in a
 callback. Pass the upstream span id and parent id when available so the UI keeps
-the original hierarchy.
+the original hierarchy. The Trace tab can switch between that recorded hierarchy
+and UI-only timeline nesting for flat exported traces; saved trace JSON and
+`deriveFromTracing` continue to use the recorded parent ids.
 
 ### Eval file (thin)
 
@@ -341,7 +344,8 @@ Mental model:
 - Authored raw cache keys are stored for debugging under
   `.agent-evals/cache-debug/<owner>.json`. This folder may include prompts,
   user inputs, or other sensitive data, should be gitignored, and is not needed
-  for cache reuse. The UI Cache hits tab shows the raw key when it is available.
+  for cache reuse. The UI Cache tab shows the raw key when it is available and
+  can be filtered to hits or new entries added by cache misses/refreshes.
 - Cached payloads use advance serialization/deserialization with the Web API plugin set, so return values and
   recorded SDK effects preserve richer built-ins such as `Date`, `Map`, `Set`,
   typed arrays, `URL`, `Headers`, `Blob`, and `File` on hits. Cache keys still