@wix/evalforge-evaluator 0.120.0 → 0.122.0

package/build/index.mjs

@@ -163,9 +163,6 @@ function createApiClient(serverUrl, options = "") {
     getPreset(projectId2, id) {
       return fetchJson(`/projects/${projectId2}/presets/${id}`);
     },
-    getAssertion(projectId2, id) {
-      return fetchJson(`/projects/${projectId2}/assertions/${id}`);
-    },
     addResult(projectId2, evalRunId2, result) {
       return postJson(
         `/projects/${projectId2}/eval-runs/${evalRunId2}/results`,
@@ -230,9 +227,6 @@ function resolveValue(value, placeholders) {
   }
   return value;
 }
-function resolvePlaceholdersInString(text, placeholders) {
-  return resolveValue(text, placeholders);
-}
 
 // src/fetch-evaluation-data.ts
 function parseSkillNamesFromParams(value) {
@@ -245,59 +239,6 @@ function parseSkillNamesFromParams(value) {
   }
   return [];
 }
-function applyParamsToAssertion(assertion, params) {
-  if (!params || Object.keys(params).length === 0) {
-    return assertion;
-  }
-  if (assertion.type === "llm_judge") {
-    const stringParams = {};
-    for (const [key, value] of Object.entries(params)) {
-      stringParams[key] = String(value ?? "");
-    }
-    const prompt = resolvePlaceholdersInString(assertion.prompt, stringParams);
-    return {
-      ...assertion,
-      prompt,
-      ...params.model !== void 0 && { model: params.model },
-      ...params.maxTokens !== void 0 && {
-        maxTokens: params.maxTokens
-      },
-      ...params.temperature !== void 0 && {
-        temperature: params.temperature
-      },
-      ...params.minScore !== void 0 && {
-        minScore: params.minScore
-      }
-    };
-  }
-  if (assertion.type === "time_limit" && params.maxDurationMs !== void 0) {
-    return {
-      ...assertion,
-      maxDurationMs: params.maxDurationMs
-    };
-  }
-  if (assertion.type === "skill_was_called" && params.skillNames !== void 0) {
-    return {
-      ...assertion,
-      skillNames: parseSkillNamesFromParams(params.skillNames)
-    };
-  }
-  if (assertion.type === "tool_called_with_param") {
-    return {
-      ...assertion,
-      ...params.toolName !== void 0 && {
-        toolName: params.toolName
-      },
-      ...params.expectedParams !== void 0 && {
-        expectedParams: params.expectedParams
-      },
-      ...params.requireSuccess !== void 0 && {
-        requireSuccess: params.requireSuccess
-      }
-    };
-  }
-  return { ...assertion, ...params };
-}
 function resolveSystemAssertion(assertionId, params) {
   const systemAssertion = SYSTEM_ASSERTIONS[assertionId];
   let baseAssertion;
@@ -352,18 +293,6 @@ function resolveSystemAssertion(assertionId, params) {
   }
   return baseAssertion;
 }
-function customAssertionToAssertion(ca, params) {
-  const config = ca.config;
-  const baseAssertion = {
-    type: "llm_judge",
-    prompt: config?.prompt ?? "",
-    minScore: config?.minScore,
-    model: config?.model,
-    maxTokens: config?.maxTokens,
-    temperature: config?.temperature
-  };
-  return applyParamsToAssertion(baseAssertion, params);
-}
 async function fetchEvaluationData(api, projectId2, evalRunId2) {
   const evalRun = await api.getEvalRun(projectId2, evalRunId2);
   const scenarios = await Promise.all(
@@ -433,30 +362,14 @@ async function fetchEvaluationData(api, projectId2, evalRunId2) {
     templateIds.map((id) => api.getTemplate(projectId2, id))
   ) : [];
   const templateMap = new Map(templates.map((t) => [t.id, t]));
-  const assertionIds = [
-    ...new Set(
-      scenarios.flatMap((s) => s.assertionLinks ?? []).map((link) => link.assertionId).filter((id) => !isSystemAssertionId(id))
-    )
-  ];
-  const assertions = assertionIds.length > 0 ? await Promise.all(
-    assertionIds.map((id) => api.getAssertion(projectId2, id))
-  ) : [];
-  const assertionMap = new Map(assertions.map((a) => [a.id, a]));
   const scenarioItems = scenarios.map((scenario) => {
     const resolvedAssertions = (scenario.assertionLinks ?? []).map((link) => {
       const { assertionId, params } = link;
-      if (isSystemAssertionId(assertionId)) {
-        return resolveSystemAssertion(
-          assertionId,
-          params
-        );
-      }
-      const customAssertion = assertionMap.get(assertionId);
-      if (!customAssertion) {
+      if (!isSystemAssertionId(assertionId)) {
         return null;
       }
-      return customAssertionToAssertion(
-        customAssertion,
+      return resolveSystemAssertion(
+        assertionId,
         params
       );
     }).filter((a) => a !== null);
@@ -2693,6 +2606,7 @@ import { writeFile as writeFile6, mkdir as mkdir7 } from "fs/promises";
 import { join as join8 } from "path";
 var KILL_GRACE_PERIOD_MS = 5e3;
 var IDLE_TIMEOUT_MS = 12e4;
+var TOOL_RUNNING_IDLE_TIMEOUT_MS = 36e4;
 var IDLE_CHECK_INTERVAL_MS = 15e3;
 function extractToolAction(toolName, args) {
   if (!toolName) return "Using tool...";
@@ -2848,6 +2762,7 @@ async function executeWithOpenCode(skills, scenario, options) {
   let lastAction = "Starting...";
   let lastToolName;
   let lastFilePath;
+  let isToolRunning = false;
   if (traceContext) {
     emitTraceEvent(
       {
@@ -3048,15 +2963,16 @@ Stderr: ${stderr.slice(0, 1e3)}`
     timers.idleCheck = setInterval(() => {
       if (resolved) return;
       const idleTime = Date.now() - lastOutputTime;
-      if (idleTime >= IDLE_TIMEOUT_MS) {
+      const effectiveTimeout = isToolRunning ? TOOL_RUNNING_IDLE_TIMEOUT_MS : IDLE_TIMEOUT_MS;
+      if (idleTime >= effectiveTimeout) {
         console.warn(
-          `[OpenCode] Process appears stuck - no output for ${Math.round(idleTime / 1e3)}s. Killing process.`
+          `[OpenCode] Process appears stuck - no output for ${Math.round(idleTime / 1e3)}s (tool running: ${isToolRunning}). Killing process.`
         );
         killProcess(child, resolved);
         finalize(
           false,
           new Error(
-            `OpenCode process stuck - no output for ${Math.round(idleTime / 1e3)} seconds (idle timeout). Skills: ${skillNames}, Scenario: ${scenario.name}`
+            `OpenCode process stuck - no output for ${Math.round(idleTime / 1e3)} seconds (idle timeout, tool running: ${isToolRunning}). Skills: ${skillNames}, Scenario: ${scenario.name}`
           )
         );
       }
@@ -3117,6 +3033,13 @@ Stderr: ${stderr.slice(0, 1e3)}`
         const evt = tryParseJson(line);
         if (!evt || !evt.type) continue;
         allEvents.push({ event: evt, receivedAt: Date.now() });
+        if (evt.type === "tool_use") {
+          const tu = evt;
+          const status = tu.part.state.status;
+          isToolRunning = status !== "completed" && status !== "error";
+        } else {
+          isToolRunning = false;
+        }
         if (traceContext) {
           traceStepNumber++;
           const traceEvt = createTraceEventFromNdjson(