@wix/evalforge-evaluator 0.31.0 → 0.32.0

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.
package/build/index.js CHANGED
@@ -6607,8 +6607,9 @@ function extractToolActionDescription(toolName, toolArgs) {
6607
6607
  try {
6608
6608
  const args = JSON.parse(toolArgs);
6609
6609
  if ((toolName === "Task" || toolName === "dispatch_agent") && args.description) {
6610
- const desc2 = String(args.description).slice(0, 60);
6611
- return desc2.length < String(args.description).length ? `${desc2}...` : desc2;
6610
+ const desc2 = String(args.description).slice(0, 55);
6611
+ const truncatedDesc = desc2.length < String(args.description).length ? `${desc2}...` : desc2;
6612
+ return `Task: ${truncatedDesc}`;
6612
6613
  }
6613
6614
  if ((toolName === "Bash" || toolName === "bash" || toolName === "execute") && args.command) {
6614
6615
  const cmd = String(args.command).slice(0, 50);
@@ -6974,15 +6975,27 @@ async function executeWithClaudeCode(skill, scenario, options) {
6974
6975
  }, SDK_TIMEOUT_MS);
6975
6976
  });
6976
6977
  if (traceContext) {
6978
+ let lastReportedAction = "";
6979
+ let sameActionCount = 0;
6977
6980
  heartbeatHandle = setInterval(() => {
6978
6981
  const elapsedMs = Date.now() - executionStartTime;
6979
6982
  let progressMessage = lastAction;
6980
- if (lastToolName && lastFilePath) {
6983
+ if (lastAction === lastReportedAction) {
6984
+ sameActionCount++;
6985
+ } else {
6986
+ sameActionCount = 1;
6987
+ lastReportedAction = lastAction;
6988
+ }
6989
+ const isTaskTool = lastToolName === "Task" || lastToolName === "dispatch_agent";
6990
+ if (isTaskTool && sameActionCount > 1) {
6991
+ progressMessage = `Waiting for ${lastAction}`;
6992
+ } else if (lastToolName && lastFilePath) {
6981
6993
  progressMessage = `${lastToolName}: ${lastFilePath}`;
6982
- } else if (lastToolName) {
6994
+ } else if (lastToolName && !isTaskTool) {
6983
6995
  progressMessage = `Using ${lastToolName}...`;
6984
6996
  }
6985
- progressMessage += ` (${Math.round(elapsedMs / 1e3)}s)`;
6997
+ const elapsedSec = Math.round(elapsedMs / 1e3);
6998
+ progressMessage += ` (${elapsedSec}s, step ${traceStepNumber})`;
6986
6999
  const progressEvent = {
6987
7000
  evalRunId: traceContext.evalRunId,
6988
7001
  scenarioId: traceContext.scenarioId,