explorbot 0.3.4 → 0.3.5

package/dist/package.json

@@ -1,6 +1,6 @@
 {
   "name": "explorbot",
-  "version": "0.3.4",
+  "version": "0.3.5",
   "description": "CLI app built with React Ink, CodeceptJS, and Playwright",
   "license": "Elastic-2.0",
   "type": "module",

package/dist/src/ai/pilot.js

@@ -1005,6 +1005,8 @@ export class Pilot {
       state), instruct Tester to verify() and finish(). If goal was already true at the start, propose
       different input data so the test is meaningful. If Tester repeats the same successful action, STOP.
 
+      If needed you should pick the exact item the scenario should act on (from the page, or precondition() one) and pass it to tester
+
       Action classification: GOAL-ADVANCING actions mutate the scenario's subject data (create/edit/delete/submit/verify).
       VIEW-ONLY actions toggle filters/tabs/sort/collapse without changing data. One VIEW-ONLY to reveal a
       target is fine; ≥2 consecutive VIEW-ONLY actions with no GOAL-ADVANCING action in between is thrashing

package/dist/src/ai/planner.js

@@ -24,7 +24,7 @@ const TasksSchema = z.object({
     planName: z.string().describe('Short descriptive name for the test plan (e.g., "User Authentication Testing", "Product Catalog Navigation", "Form Validation Tests")'),
     scenarios: z
         .array(z.object({
-        scenario: z.string().describe('A single sentence describing what to test'),
+        scenario: z.string().describe('A single sentence describing the behavior to test.'),
         priority: z.enum(['critical', 'important', 'high', 'normal', 'low']).describe('Priority of the task based on business importance'),
         startUrl: z.string().nullable().describe('Start URL for the test if different from plan URL. Use only stable feature/list/detail pages, not transient create/edit/modal URLs unless the scenario specifically starts inside that form.'),
         steps: z.array(z.string()).describe('List of steps to perform for this scenario. Each step should be a specific action (e.g., "Open the form", "Enter required data", "Submit the form"). Keep steps atomic and actionable.'),

package/dist/src/ai/provider.js

@@ -2,7 +2,8 @@ import { OpenTelemetry } from '@ai-sdk/otel';
 import { LangfuseSpanProcessor } from '@langfuse/otel';
 import { NodeSDK } from '@opentelemetry/sdk-node';
 import { AsyncLocalStorage } from 'node:async_hooks';
-import { generateObject, generateText, isStepCount, registerTelemetry, tool } from 'ai';
+import dedent from 'dedent';
+import { APICallError, generateObject, generateText, isStepCount, registerTelemetry, tool } from 'ai';
 import { z } from 'zod';
 import { clearActivity, setActivity } from "../activity.js";
 import { configuredModels, modelName as getModelName } from '../config.js';
@@ -382,9 +383,18 @@ export class Provider {
         if (extraStop)
             stopConditions.push(extraStop);
         const config = this.buildGenerateConfig({ tools: toolsWithCommentary, maxOutputTokens: 16384, toolChoice: 'auto', experimental_repairToolCall: repairToolCall }, { stopWhen: stopConditions, model }, options);
+        let attemptMessages = messages;
+        let invalidRequestFeedbackAdded = false;
         try {
             const response = await this.withModelRequestSlot(() => withRetry(async () => {
-                const result = (await this.raceWithIdleTimeout((signal) => generateText({ messages, ...config, abortSignal: signal }), config.timeout || 30000));
+                const result = (await this.raceWithIdleTimeout((signal) => generateText({ messages: attemptMessages, ...config, abortSignal: signal }), config.timeout || 30000).catch((error) => {
+                    if (!invalidRequestFeedbackAdded) {
+                        const amended = withInvalidRequestFeedback(attemptMessages, error);
+                        invalidRequestFeedbackAdded = amended !== attemptMessages;
+                        attemptMessages = amended;
+                    }
+                    throw error;
+                }));
                 this.recordUsage(options.agentName || 'unknown', modelName, result.usage);
                 const hasToolCall = (result.toolCalls?.length || 0) > 0;
                 if (!result.text && !hasToolCall && result.finishReason === 'length') {
@@ -620,6 +630,22 @@ function repairToolCall(options) {
         return repairChannelMarker(options);
     return repairHarmonyChannel(options);
 }
+function withInvalidRequestFeedback(messages, error) {
+    if (!(error instanceof APICallError) || error.statusCode !== 400)
+        return messages;
+    tag('warning').log('Provider rejected the request as invalid — relaying its reason before the retry');
+    return [
+        ...messages,
+        {
+            role: 'user',
+            content: dedent `
+        The previous request was rejected by the provider as invalid:
+        "${error.message}"
+        Fix what it describes and re-issue the request.
+      `,
+        },
+    ];
+}
 function repairChannelMarker({ toolCall, tools }) {
     const markerIndex = toolCall.toolName.indexOf('<|channel|>');
     if (markerIndex <= 0)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "explorbot",
-  "version": "0.3.4",
+  "version": "0.3.5",
   "description": "CLI app built with React Ink, CodeceptJS, and Playwright",
   "license": "Elastic-2.0",
   "type": "module",

package/src/ai/pilot.ts

@@ -1114,6 +1114,8 @@ export class Pilot implements Agent {
       state), instruct Tester to verify() and finish(). If goal was already true at the start, propose
       different input data so the test is meaningful. If Tester repeats the same successful action, STOP.
 
+      If needed you should pick the exact item the scenario should act on (from the page, or precondition() one) and pass it to tester
+
       Action classification: GOAL-ADVANCING actions mutate the scenario's subject data (create/edit/delete/submit/verify).
       VIEW-ONLY actions toggle filters/tabs/sort/collapse without changing data. One VIEW-ONLY to reveal a
       target is fine; ≥2 consecutive VIEW-ONLY actions with no GOAL-ADVANCING action in between is thrashing

package/src/ai/planner.ts

@@ -33,7 +33,7 @@ const TasksSchema = z.object({
   scenarios: z
     .array(
       z.object({
-        scenario: z.string().describe('A single sentence describing what to test'),
+        scenario: z.string().describe('A single sentence describing the behavior to test.'),
         priority: z.enum(['critical', 'important', 'high', 'normal', 'low']).describe('Priority of the task based on business importance'),
         startUrl: z.string().nullable().describe('Start URL for the test if different from plan URL. Use only stable feature/list/detail pages, not transient create/edit/modal URLs unless the scenario specifically starts inside that form.'),
         steps: z.array(z.string()).describe('List of steps to perform for this scenario. Each step should be a specific action (e.g., "Open the form", "Enter required data", "Submit the form"). Keep steps atomic and actionable.'),

package/src/ai/provider.ts

@@ -2,7 +2,8 @@ import { OpenTelemetry } from '@ai-sdk/otel';
 import { LangfuseSpanProcessor } from '@langfuse/otel';
 import { NodeSDK } from '@opentelemetry/sdk-node';
 import { AsyncLocalStorage } from 'node:async_hooks';
-import { generateObject, generateText, isStepCount, registerTelemetry, tool } from 'ai';
+import dedent from 'dedent';
+import { APICallError, generateObject, generateText, isStepCount, registerTelemetry, tool } from 'ai';
 import type { ModelMessage } from 'ai';
 import { z } from 'zod';
 import { clearActivity, setActivity } from '../activity.ts';
@@ -423,10 +424,19 @@ export class Provider {
     const stopConditions: any[] = [isStepCount(maxRoundtrips)];
     if (extraStop) stopConditions.push(extraStop);
     const config = this.buildGenerateConfig({ tools: toolsWithCommentary, maxOutputTokens: 16384, toolChoice: 'auto', experimental_repairToolCall: repairToolCall }, { stopWhen: stopConditions, model }, options);
+    let attemptMessages = messages;
+    let invalidRequestFeedbackAdded = false;
     try {
       const response = await this.withModelRequestSlot(() =>
         withRetry(async () => {
-          const result = (await this.raceWithIdleTimeout((signal) => generateText({ messages, ...config, abortSignal: signal }), config.timeout || 30000)) as any;
+          const result = (await this.raceWithIdleTimeout((signal) => generateText({ messages: attemptMessages, ...config, abortSignal: signal }), config.timeout || 30000).catch((error) => {
+            if (!invalidRequestFeedbackAdded) {
+              const amended = withInvalidRequestFeedback(attemptMessages, error);
+              invalidRequestFeedbackAdded = amended !== attemptMessages;
+              attemptMessages = amended;
+            }
+            throw error;
+          })) as any;
           this.recordUsage(options.agentName || 'unknown', modelName, result.usage);
           const hasToolCall = (result.toolCalls?.length || 0) > 0;
           if (!result.text && !hasToolCall && result.finishReason === 'length') {
@@ -693,6 +703,22 @@ function repairToolCall(options: ToolCallRepairOptions): any | null {
   return repairHarmonyChannel(options);
 }
 
+function withInvalidRequestFeedback(messages: ModelMessage[], error: unknown): ModelMessage[] {
+  if (!(error instanceof APICallError) || error.statusCode !== 400) return messages;
+  tag('warning').log('Provider rejected the request as invalid — relaying its reason before the retry');
+  return [
+    ...messages,
+    {
+      role: 'user',
+      content: dedent`
+        The previous request was rejected by the provider as invalid:
+        "${error.message}"
+        Fix what it describes and re-issue the request.
+      `,
+    },
+  ];
+}
+
 function repairChannelMarker({ toolCall, tools }: ToolCallRepairOptions): any | null {
   const markerIndex = toolCall.toolName.indexOf('<|channel|>');
   if (markerIndex <= 0) return null;