@wix/pathgrade 1.0.23 → 1.0.24

package/dist/sdk/evaluate.js

@@ -1,4 +1,4 @@
-import { extractSkillsFromLog } from '../tool-events.js';
+import { extractSkillsFromLog, extractToolEventsFromLog } from '../tool-events.js';
 import { getRuntime } from './eval-runtime.js';
 import { emitEvalResult } from './result-capture.js';
 import { runJudgePipeline } from './judge-pipeline.js';
@@ -34,9 +34,7 @@ export class EvalScorerError extends Error {
 function makeEvaluateAgent() {
     const conversationAttributed = new WeakSet();
     return async function evaluateAgent(agent, scorers, opts) {
-        const toolEvents = agent.log
-            .filter((e) => e.type === 'tool_event' && e.tool_event)
-            .map((e) => e.tool_event);
+        const toolEvents = extractToolEventsFromLog(agent.log);
         const ctx = {
             workspace: agent.workspace,
             log: agent.log,

package/dist/sdk/index.d.ts

@@ -14,6 +14,7 @@ export { runScorer } from './run-scorer.js';
 export { previewReactions } from './reaction-preview.js';
 export { setRuntime, resetRuntime } from './eval-runtime.js';
 export { DEFAULT_COPY_IGNORE } from '../providers/copy-filter.js';
+export { extractToolEventsFromLog } from '../tool-events.js';
 export { createAskBus, requireAskBusForLiveBatches, AskBusTimeoutError } from './ask-bus/bus.js';
 export { toAskUserToolEvent } from './ask-bus/projection.js';
 export type { AskUserToolEvent, AskUserToolEventArguments, AskUserToolEventQuestionArgument, } from './ask-bus/projection.js';

package/dist/sdk/index.js

@@ -14,6 +14,7 @@ export { runScorer } from './run-scorer.js';
 export { previewReactions } from './reaction-preview.js';
 export { setRuntime, resetRuntime } from './eval-runtime.js';
 export { DEFAULT_COPY_IGNORE } from '../providers/copy-filter.js';
+export { extractToolEventsFromLog } from '../tool-events.js';
 export { createAskBus, requireAskBusForLiveBatches, AskBusTimeoutError } from './ask-bus/bus.js';
 export { toAskUserToolEvent } from './ask-bus/projection.js';
 export { buildAskBatchLogEntries } from './agent-result-log.js';

package/dist/sdk/mcp-evidence.d.ts

@@ -44,6 +44,8 @@ export interface ExpectedMcpToolCall {
     serverName?: string;
     toolName?: string;
     status?: string;
+    invocation?: McpInvocationResult['invocation'];
+    outcome?: McpInvocationResult['outcome'];
     argumentsContaining?: Record<string, unknown>;
 }
 export interface McpStartupStatusEvidence {

package/dist/sdk/mcp-evidence.js

@@ -127,6 +127,17 @@ function matchesMcpToolCall(call, expected) {
         return false;
     if (expected.status !== undefined && call.status !== expected.status)
         return false;
+    if (expected.invocation !== undefined || expected.outcome !== undefined) {
+        const invocation = getMcpInvocation(call.event);
+        if (!invocation)
+            return false;
+        if (invocation.serverName !== call.serverName || invocation.toolName !== call.toolName)
+            return false;
+        if (expected.invocation !== undefined && invocation.invocation !== expected.invocation)
+            return false;
+        if (expected.outcome !== undefined && invocation.outcome !== expected.outcome)
+            return false;
+    }
     if (expected.argumentsContaining && !containsArguments(call.arguments, expected.argumentsContaining)) {
         return false;
     }

package/dist/sdk/snapshots.js

@@ -1,11 +1,10 @@
+import { extractToolEventsFromLog } from '../tool-events.js';
 import { sanitizePersistenceValue } from '../tool-event-results.js';
 import fs from 'fs-extra';
 export const RUN_SNAPSHOT_VERSION = 2;
 export function buildRunSnapshot(params) {
     const { agent, messages, log, conversationResult, workspace, timestamp } = params;
-    const toolEvents = log
-        .filter((entry) => entry.type === 'tool_event' && entry.tool_event)
-        .map((entry) => entry.tool_event);
+    const toolEvents = extractToolEventsFromLog(log);
     return sanitizePersistenceValue({
         version: agent === 'opencode' ? 2 : 1,
         timestamp: timestamp ?? new Date().toISOString(),

package/dist/tool-events.d.ts

@@ -59,6 +59,11 @@ export declare function extractSkillsFromLog(log: ReadonlyArray<{
     type: string;
     tool_event?: ToolEvent;
 }>): string[];
+/** Extract normalized tool events from a session log. */
+export declare function extractToolEventsFromLog(log: ReadonlyArray<{
+    type: string;
+    tool_event?: ToolEvent;
+}>): ToolEvent[];
 /**
  * Extract deduplicated skill names from tool events that have action 'use_skill'.
  */

package/dist/tool-events.js

@@ -93,10 +93,13 @@ export function buildSummary(action, toolName, args) {
  * Filters for tool_event entries, then delegates to extractSkillsFromToolEvents.
  */
 export function extractSkillsFromLog(log) {
-    const toolEvents = log
+    return extractSkillsFromToolEvents(extractToolEventsFromLog(log));
+}
+/** Extract normalized tool events from a session log. */
+export function extractToolEventsFromLog(log) {
+    return log
         .filter((e) => e.type === 'tool_event' && e.tool_event != null)
         .map((e) => e.tool_event);
-    return extractSkillsFromToolEvents(toolEvents);
 }
 /**
  * Extract deduplicated skill names from tool events that have action 'use_skill'.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wix/pathgrade",
-  "version": "1.0.23",
+  "version": "1.0.24",
   "packageManager": "yarn@4.12.0",
   "description": "Evaluate whether AI agents discover and use your skills correctly",
   "exports": {
@@ -140,5 +140,5 @@
     "typescript": "^5.9.3",
     "zod": "4.3.6"
   },
-  "falconPackageHash": "0debc4519da71de54bc8ca58c57f1ddbf0899cc2b06cbf908371e1e8"
+  "falconPackageHash": "45feda0f936d0f7a5e2c85b4b55cd80ebe1f411b76a3c5c38b3fc149"
 }