explorbot 0.1.32-beta.1 → 0.2.1

package/src/ai/planner.ts

@@ -3,7 +3,7 @@ import { z } from 'zod';
 import { ActionResult } from '../action-result.ts';
 import { setActivity } from '../activity.ts';
 import { ConfigParser } from '../config.ts';
-import { ExperienceTracker } from '../experience-tracker.ts';
+import type { ExperienceTracker } from '../experience-tracker.ts';
 import type Explorer from '../explorer.ts';
 import { Observability } from '../observability.ts';
 import type { StateManager } from '../state-manager.js';
@@ -14,14 +14,14 @@ import { createDebug, tag } from '../utils/logger.js';
 import { jsonToTable } from '../utils/markdown-parser.ts';
 import { mdq } from '../utils/markdown-query.js';
 import { planToCompactAiContext } from '../utils/test-plan-markdown.ts';
-import type { Agent } from './agent.js';
+import type { Agent, AgentDeps } from './agent.js';
 import { Conversation } from './conversation.ts';
 import type { Fisherman } from './fisherman.ts';
 import { WithSessionDedup } from './planner/session-dedup.ts';
-import { getActiveStyle, getStyles } from './planner/styles.ts';
+import { getActiveStyle } from './planner/styles.ts';
 import { WithSubPages, getPlannedByStateHash, getRegisteredPlan, registerPlan } from './planner/subpages.ts';
 import type { Provider } from './provider.js';
-import { POSSIBLE_SECTIONS, Researcher } from './researcher.ts';
+import { POSSIBLE_SECTIONS, type Researcher } from './researcher.ts';
 import { findSimilarStateHash } from './researcher/cache.ts';
 import { hasFocusedSection } from './researcher/focus.ts';
 import { capabilityGroundingRule, dataProtectionRules, fileUploadRule } from './rules.ts';
@@ -63,13 +63,13 @@ export class Planner extends PlannerBase implements Agent {
   researcher: Researcher;
   private fisherman: Fisherman | null = null;
 
-  constructor(explorer: Explorer, provider: Provider) {
+  constructor(deps: AgentDeps, researcher: Researcher) {
     super();
-    this.explorer = explorer;
-    this.provider = provider;
-    this.researcher = new Researcher(explorer, provider);
-    this.stateManager = explorer.getStateManager();
-    this.experienceTracker = new ExperienceTracker();
+    this.explorer = deps.explorer;
+    this.provider = deps.ai;
+    this.researcher = researcher;
+    this.stateManager = deps.stateManager;
+    this.experienceTracker = deps.stateManager.getExperienceTracker();
   }
 
   setFisherman(fisherman: Fisherman): void {
@@ -98,7 +98,7 @@ export class Planner extends PlannerBase implements Agent {
     You are ISTQB certified senior manual QA planning exploratory testing session of a web application.
     Each test scenario must complete a meaningful user workflow — not just open a UI element and verify it exists.
     Bad: "Open the Help panel" — just opens something.
-    Good: "Create a new suite and verify it appears in the list" — completes a business action.
+    Good: "Create a new post and verify it appears in the list" — completes a business action.
     </role>
     <task>
       List possible testing scenarios for the web page.
@@ -171,6 +171,9 @@ export class Planner extends PlannerBase implements Agent {
 
     const tags = ['planner'];
     if (style) tags.push(style);
+
+    const existingTestScenarios = this.getExistingTestFileScenarios(state.url);
+
     const result = await Observability.run(`planner: ${state.url}`, { tags, sessionId: state.url }, async () => {
       const actionResult = ActionResult.fromState(state);
       const conversation = await this.buildConversation(actionResult, style, parentPlan, feature);
@@ -211,7 +214,6 @@ export class Planner extends PlannerBase implements Agent {
       const defaultStartUrl = this.getDefaultStartUrl(state);
       if (parentPlan) this.currentPlan.parentPlan = parentPlan;
       const allPreviousScenarios = this.getPreviousSessionScenarios();
-      const existingTestScenarios = this.getExistingTestFileScenarios(state.url);
       for (const s of existingTestScenarios) allPreviousScenarios.add(s);
       for (const t of tests) {
         if (allPreviousScenarios.has(t.scenario.toLowerCase())) continue;
@@ -266,6 +268,7 @@ export class Planner extends PlannerBase implements Agent {
   }
 
   private getExistingTestFileScenarios(currentUrl?: string): Set<string> {
+    this.lastSuite = null;
     if (!currentUrl) return new Set<string>();
     try {
       this.lastSuite = new Suite(currentUrl);
@@ -277,29 +280,36 @@ export class Planner extends PlannerBase implements Agent {
   }
 
   private cleanExperienceFlows(text: string): string | null {
-    const seenTitles = new Set<string>();
     let result = text;
+    const seenTitles = new Set<string>();
+    for (const selector of ['section2', 'section3']) {
+      result = mdq(result)
+        .query(selector)
+        .replaceEach((section) => {
+          const heading = section.query('heading').text().trim();
+          const withoutHeadings = mdq(section.text()).query('heading').replace('');
+          const body = mdq(withoutHeadings).query('hr').replace('').trim();
+          if (body && !seenTitles.has(heading)) {
+            seenTitles.add(heading);
+            return section.text();
+          }
+          return '';
+        });
+    }
 
-    for (const section of [...mdq(text).query('section2').each(), ...mdq(text).query('section3').each()]) {
-      const heading = section.query('heading').text().trim();
-      const body = mdq(section.text())
-        .query('heading')
-        .replace('')
-        .replace(/^---\s*$/gm, '')
-        .trim();
-
-      if (!body || seenTitles.has(heading)) {
-        result = result.replace(section.text(), '');
-        continue;
-      }
-      seenTitles.add(heading);
-
-      const blockquotes = section.query('blockquote').each();
-      if (blockquotes.length <= 10) continue;
-      for (const bq of blockquotes.slice(10)) {
-        result = result.replace(bq.text(), '');
-      }
-      result = result.replace(section.text().trim(), `${section.text().trim()}\n> ... and ${blockquotes.length - 10} more discoveries`);
+    const trimmedTitles = new Set<string>();
+    for (const selector of ['section2', 'section3']) {
+      result = mdq(result)
+        .query(selector)
+        .replaceEach((section) => {
+          const heading = section.query('heading').text().trim();
+          if (trimmedTitles.has(heading)) return section.text();
+          const count = section.query('blockquote').count();
+          if (count <= 10) return section.text();
+          const kept = mdq(section.text()).query('blockquote[10:]').replace('');
+          trimmedTitles.add(heading);
+          return `${kept.trimEnd()}\n> ... and ${count - 10} more discoveries\n`;
+        });
     }
 
     return result.trim() || null;
@@ -338,6 +348,7 @@ export class Planner extends PlannerBase implements Agent {
       If there are subpages (pages with same URL path) plan testing of those subpages as well
       If you plan to test CRUD operations, plan them in correct order: create, read, update.
       Do not invent specific route names, success messages, validation texts, badge counts, or welcome messages unless they are visible in research, visited pages, or prior observed flows.
+      When validation placement or wording was not observed, require feedback associated with the invalid input without inventing a specific location or message.
       If exact wording is unknown, describe the expected result generically, for example "an authentication error is shown" or "the user stays on the login page" instead of guessing the literal text.
       If exact redirect destination is unknown, describe the destination by visible page identity, for example "the dashboard page opens" or "the current workspace home page opens" instead of inventing a URL slug.
       Only propose scenarios whose prerequisites are evident from page research, visited pages, or API data preparation context.
@@ -363,31 +374,26 @@ export class Planner extends PlannerBase implements Agent {
       ${dataProtectionRules}
       ${fileUploadRule}
       </rules>
-
-      ${this.buildApproach(style)}
-
-      <context>
-      URL: ${state.url || 'Unknown'}
-      Title: ${state.title || 'Unknown'}
-      </context>
     `;
 
     conversation.addUserText(planningPrompt);
+    conversation.addUserText(this.getTasksMessage());
     const currentState = this.stateManager.getCurrentState();
     const research = await this.researcher.research(currentState || state, {
       deep: true,
     });
     let plannerResearch = mdq(research).query('code').replace('');
-    for (const table of mdq(plannerResearch).query('table').each()) {
-      const rawTable = table.text();
-      const rows = table.toJson();
-      if (rows.length === 0 || !rows[0].Element) continue;
-      const elementWithType = rows.map((r) => ({
-        Element: r.Element,
-        Type: r.Type || '',
-      }));
-      plannerResearch = plannerResearch.replace(rawTable, jsonToTable(elementWithType, ['Element', 'Type']));
-    }
+    plannerResearch = mdq(plannerResearch)
+      .query('table')
+      .replaceEach((table) => {
+        const rows = table.toJson();
+        if (rows.length === 0 || !rows[0].Element) return table.text();
+        const elementWithType = rows.map((r) => ({
+          Element: r.Element,
+          Type: r.Type || '',
+        }));
+        return jsonToTable(elementWithType, ['Element', 'Type']);
+      });
 
     const hasFocusedOverlay = hasFocusedSection(plannerResearch);
     const focusNote = hasFocusedOverlay ? "IMPORTANT: One section is marked as **Focused** — this is the user's current focus area. Concentrate testing on the Focused section FIRST — test all interactions inside it before planning tests for the rest of the page." : '';
@@ -405,6 +411,15 @@ export class Planner extends PlannerBase implements Agent {
       </page_research>
     `);
 
+    conversation.addUserText(dedent`
+      ${this.buildApproach(style)}
+
+      <context>
+      URL: ${state.url || 'Unknown'}
+      Title: ${state.title || 'Unknown'}
+      </context>
+    `);
+
     if (this.fisherman) {
       await this.fisherman.ensureReady(state.url);
       if (this.fisherman.isAvailable()) {
@@ -556,8 +571,11 @@ export class Planner extends PlannerBase implements Agent {
       `);
     }
 
-    const hasCurrentPlan = !!this.currentPlan;
-    const tasksMessage = dedent`
+    return conversation;
+  }
+
+  private getTasksMessage(): string {
+    return dedent`
     <task>
       Provide testing scenarios as structured data with the following requirements:
       1. Create a short, descriptive plan name that summarizes what will be tested (e.g., "User Authentication Testing", "Product Catalog Navigation", "Form Validation Tests")
@@ -576,17 +594,17 @@ export class Planner extends PlannerBase implements Agent {
 
          STEPS - actionable commands:
          - Each step should be a specific action to perform
-         - Good: "Click Create Suite button", "Enter 'My New Suite' as suite name", "Click Save button"
-         - Good: "Click Star icon on 'Template API Testing' suite"
+         - Good: "Click Create Post button", "Enter 'My New Post' as post title", "Click Save button"
+         - Good: "Click Star icon on 'Quarterly Report' item"
          - Bad: "Open the dropdown menu" (opening is not a goal, it's a means)
          - Bad: "Verify modal appears" (verification belongs in expected outcomes, not steps)
          - Steps should form a complete workflow, not stop at opening a UI element
          - NEVER split a workflow across two tests. One test = one complete action + its verification
 
          EXPECTED OUTCOMES - verifiable results:
-         - Good: "New suite 'My New Suite' appears in the suite list"
-         - Good: "Suite appears under Starred filter tab"
-         - Good: "Success message 'Suite created' is displayed"
+         - Good: "New post 'My New Post' appears in the posts list"
+         - Good: "Item appears under Starred filter tab"
+         - Good: "Success message 'Post created' is displayed"
          - Good when wording is unknown: "An authentication error is displayed"
          - Good when route is unknown: "The workspace home page is displayed"
          - Bad: "Modal is displayed" (just verifying existence, no business value)
@@ -603,12 +621,8 @@ export class Planner extends PlannerBase implements Agent {
          - Do not wrap text in ** or * quotes, ( or ) brackets.
          - Avoid using emojis or special characters.
       7. Only tests that can be tested from web UI should be proposed.
-      ${hasCurrentPlan ? '8. CRITICAL: Return ONLY NEW scenarios not in the existing tests list. Return empty array if no new tests needed.' : `8. At least ${this.MIN_TASKS} tests should be proposed.`}
+      8. For a fresh plan propose at least ${this.MIN_TASKS} tests; when expanding an existing plan return ONLY new scenarios not in the tested list, and an empty array if none remain.
     </task>
     `;
-
-    conversation.addUserText(tasksMessage);
-
-    return conversation;
   }
 }

package/src/ai/provider.ts

@@ -11,7 +11,7 @@ import { Stats } from '../stats.ts';
 import { createDebug, tag } from '../utils/logger.js';
 import { type RetryOptions, withRetry } from '../utils/retry.js';
 import { RulesLoader } from '../utils/rules-loader.ts';
-import { Conversation } from './conversation.js';
+import { Conversation, toToolExecution } from './conversation.js';
 
 const debugLog = createDebug('explorbot:provider');
 const promptLog = createDebug('explorbot:provider:out');
@@ -22,6 +22,8 @@ export class ContextLengthError extends Error {}
 
 let telemetryRegistered = false;
 
+const CONTEXT_LENGTH_PATTERNS = ['reduce the length', 'context length', 'maximum context', 'token limit', 'too many tokens', 'max_tokens', 'context_length_exceeded', 'output truncated at maxtokens'];
+
 function extractCachedTokens(usage: any): number {
   if (!usage) return 0;
   const direct = usage.inputTokenDetails?.cacheReadTokens ?? usage.cachedInputTokens;
@@ -31,10 +33,10 @@ function extractCachedTokens(usage: any): number {
   return typeof fromRaw === 'number' ? fromRaw : 0;
 }
 
-function rejectAfterIdle(ms: number, signal: { cancelled: boolean }, controller: AbortController): Promise<never> {
+function abortAfterIdle(ms: number, cancel: { cancelled: boolean }, controller: AbortController): Promise<never> {
   return new Promise((_, reject) => {
     const tick = () => {
-      if (signal.cancelled) return;
+      if (cancel.cancelled) return;
       if (executionController.isAwaitingInput()) {
         setTimeout(tick, ms);
         return;
@@ -46,6 +48,12 @@ function rejectAfterIdle(ms: number, signal: { cancelled: boolean }, controller:
   });
 }
 
+function combinedAbortSignal(controller: AbortController): AbortSignal {
+  const executionSignal = executionController.getAbortSignal();
+  if (!executionSignal) return controller.signal;
+  return AbortSignal.any([controller.signal, executionSignal]);
+}
+
 export class Provider {
   private config: AIConfig;
   private telemetryEnabled = false;
@@ -70,7 +78,6 @@ export class Provider {
     },
   };
 
-  static readonly CONTEXT_LENGTH_PATTERNS = ['reduce the length', 'context length', 'maximum context', 'token limit', 'too many tokens', 'max_tokens', 'context_length_exceeded', 'output truncated at maxtokens'];
   lastConversation: Conversation | null = null;
 
   constructor(config: AIConfig) {
@@ -90,7 +97,6 @@ export class Provider {
       await generateText({
         model: this.config.model,
         prompt: 'hi',
-        maxOutputTokens: 1,
       });
     } catch (error: any) {
       throw new AiError(`AI connection failed: ${error.message}`);
@@ -138,12 +144,12 @@ export class Provider {
     return parts.length > 0 ? parts.join('\n\n') : undefined;
   }
 
-  getProviderOptionsForAgent(agentName: string): Record<string, any> | undefined {
+  private getProviderOptionsForAgent(agentName: string): Record<string, any> | undefined {
     const agentConfig = this.config.agents?.[agentName as keyof typeof this.config.agents];
     return agentConfig?.providerOptions;
   }
 
-  getReasoningForAgent(agentName?: string): string | undefined {
+  private getReasoningForAgent(agentName?: string): string | undefined {
     if (!agentName) return undefined;
     const agentConfig = this.config.agents?.[agentName as keyof typeof this.config.agents];
     return agentConfig?.reasoning;
@@ -172,6 +178,50 @@ export class Provider {
     if (reasoning) config.reasoning ??= reasoning;
   }
 
+  private buildGenerateConfig(defaults: Record<string, any>, overrides: Record<string, any>, options: any): Record<string, any> {
+    const telemetry = this.getTelemetry(options);
+    const config = this.mergeProviderOptions(
+      {
+        ...defaults,
+        allowSystemInMessages: true,
+        ...(this.config.config || {}),
+        ...options,
+        ...overrides,
+      },
+      options.agentName
+    );
+    this.finalizeConfig(config, options, telemetry);
+    return config;
+  }
+
+  private recordUsage(agentName: string, modelName: string, usage: any): void {
+    if (!usage) return;
+    Stats.recordTokens(agentName, modelName, {
+      input: usage.inputTokens ?? usage.promptTokens ?? 0,
+      output: usage.outputTokens ?? usage.completionTokens ?? 0,
+      total: usage.totalTokens ?? 0,
+      cached: extractCachedTokens(usage),
+    });
+  }
+
+  private async raceWithIdleTimeout<T>(fn: (signal: AbortSignal) => Promise<T>, timeoutMs: number): Promise<T> {
+    const cancel = { cancelled: false };
+    const controller = new AbortController();
+    const combinedSignal = combinedAbortSignal(controller);
+    try {
+      return await Promise.race([fn(combinedSignal), abortAfterIdle(timeoutMs, cancel, controller)]);
+    } finally {
+      cancel.cancelled = true;
+    }
+  }
+
+  private async recoverFromContextLength(error: any, messages: ModelMessage[], options: any, retry: (messages: ModelMessage[], options: any) => Promise<any>): Promise<any> {
+    const reduced = this.tryReduceMessages(messages, options._contextRetryLevel || 0);
+    if (!reduced) throw new ContextLengthError(error.message || error.toString());
+    tag('warning').log('Context length exceeded, retrying with reduced messages...');
+    return retry(reduced.messages, { ...options, _contextRetryLevel: reduced.nextLevel });
+  }
+
   private initLangfuse() {
     const langfuseConfig = this.config.langfuse;
     const publicKey = langfuseConfig?.publicKey || process.env.LANGFUSE_PUBLIC_KEY;
@@ -207,20 +257,25 @@ export class Provider {
 
     const runTelemetry = Observability.getTelemetry();
 
-    if (!options.telemetry) {
+    let optionTelemetry = options.telemetry;
+    if (options.telemetryFunctionId && !optionTelemetry?.functionId) {
+      optionTelemetry = { ...optionTelemetry, functionId: options.telemetryFunctionId };
+    }
+
+    if (!optionTelemetry) {
       return runTelemetry;
     }
 
     if (!runTelemetry) {
-      return options.telemetry;
+      return optionTelemetry;
     }
 
     return {
       ...runTelemetry,
-      ...options.telemetry,
+      ...optionTelemetry,
       metadata: {
         ...runTelemetry.metadata,
-        ...options.telemetry.metadata,
+        ...optionTelemetry.metadata,
       },
     };
   }
@@ -254,12 +309,8 @@ export class Provider {
     const toolCalls = response.toolCalls || [];
     const toolResults = response.toolResults || [];
 
-    const toolExecutions = toolCalls.map((call: any, index: number) => ({
-      toolName: call.toolName || '',
-      input: call.input,
-      output: toolResults[index]?.output,
-      wasSuccessful: toolResults[index]?.output?.success || false,
-    }));
+    const resultsById = new Map(toolResults.map((r: any) => [r.toolCallId, r]));
+    const toolExecutions = toolCalls.map((call: any) => toToolExecution(call.toolName || '', call.input, resultsById.get(call.toolCallId)?.output));
 
     return { conversation, response, toolExecutions };
   }
@@ -269,19 +320,7 @@ export class Provider {
     setActivity(`🤖 Asking ${modelName}`, 'ai');
     promptLog(`Using model: ${modelName}`);
 
-    const telemetry = this.getTelemetry(options);
-    const config = this.mergeProviderOptions(
-      {
-        maxOutputTokens: 16384,
-        allowSystemInMessages: true,
-        ...(this.config.config || {}),
-        ...options,
-        model,
-        abortSignal: executionController.getAbortSignal(),
-      },
-      options.agentName
-    );
-    this.finalizeConfig(config, options, telemetry);
+    const config = this.buildGenerateConfig({ maxOutputTokens: 16384 }, { model, abortSignal: executionController.getAbortSignal() }, options);
 
     promptLog(messages[messages.length - 1].content);
     try {
@@ -303,14 +342,7 @@ export class Provider {
       clearActivity();
       responseLog(response.text);
 
-      if (response.usage) {
-        Stats.recordTokens(options.agentName || 'unknown', modelName, {
-          input: response.usage.inputTokens ?? response.usage.promptTokens ?? 0,
-          output: response.usage.outputTokens ?? response.usage.completionTokens ?? 0,
-          total: response.usage.totalTokens ?? 0,
-          cached: extractCachedTokens(response.usage),
-        });
-      }
+      this.recordUsage(options.agentName || 'unknown', modelName, response.usage);
 
       return response;
     } catch (error: any) {
@@ -318,12 +350,7 @@ export class Provider {
       if (error?.name === 'AbortError') throw error;
       if (error instanceof ContextLengthError) throw error;
       if (Provider.isContextLengthError(error)) {
-        const reduced = this.tryReduceMessages(messages, options._contextRetryLevel || 0);
-        if (reduced) {
-          tag('warning').log('Context length exceeded, retrying with reduced messages...');
-          return this.chat(reduced.messages, model, { ...options, _contextRetryLevel: reduced.nextLevel });
-        }
-        throw new ContextLengthError(error.message || error.toString());
+        return this.recoverFromContextLength(error, messages, options, (m, o) => this.chat(m, model, o));
       }
       const message = error.message || error.toString();
       if (message !== 'No response text from AI') {
@@ -343,50 +370,19 @@ export class Provider {
     promptLog('Available tools:', toolNames);
     promptLog(messages[messages.length - 1].content);
 
-    const telemetry = this.getTelemetry(options);
     const maxRoundtrips = options.maxToolRoundtrips ?? 5;
     const extraStop = options.stopWhen;
     const stopConditions: any[] = [isStepCount(maxRoundtrips)];
     if (extraStop) stopConditions.push(extraStop);
-    const { stopWhen: _ignoredStopWhen, ...optionsWithoutStop } = options;
-    const config = this.mergeProviderOptions(
-      {
-        tools,
-        maxOutputTokens: 16384,
-        toolChoice: 'auto',
-        allowSystemInMessages: true,
-        ...(this.config.config || {}),
-        ...optionsWithoutStop,
-        stopWhen: stopConditions,
-        model,
-        abortSignal: executionController.getAbortSignal(),
-      },
-      options.agentName
-    );
-    this.finalizeConfig(config, options, telemetry);
+    const config = this.buildGenerateConfig({ tools, maxOutputTokens: 16384, toolChoice: 'auto' }, { stopWhen: stopConditions, model }, options);
     try {
       const response = await withRetry(async () => {
-        const timeout = config.timeout || 30000;
-        const cancel = { cancelled: false };
-        const controller = new AbortController();
-        const combinedSignal = AbortSignal.any([controller.signal, executionController.getAbortSignal()].filter(Boolean) as AbortSignal[]);
-        try {
-          const result = (await Promise.race([
-            generateText({
-              messages,
-              ...config,
-              abortSignal: combinedSignal,
-            }),
-            rejectAfterIdle(timeout, cancel, controller),
-          ])) as any;
-          const hasToolCall = (result.toolCalls?.length || 0) > 0;
-          if (!result.text && !hasToolCall && result.finishReason === 'length') {
-            throw new ContextLengthError('AI response empty: output truncated at maxTokens. Increase maxOutputTokens in config or use a model with higher output capacity.');
-          }
-          return result;
-        } finally {
-          cancel.cancelled = true;
+        const result = (await this.raceWithIdleTimeout((signal) => generateText({ messages, ...config, abortSignal: signal }), config.timeout || 30000)) as any;
+        const hasToolCall = (result.toolCalls?.length || 0) > 0;
+        if (!result.text && !hasToolCall && result.finishReason === 'length') {
+          throw new ContextLengthError('AI response empty: output truncated at maxTokens. Increase maxOutputTokens in config or use a model with higher output capacity.');
         }
+        return result;
       }, this.getRetryOptions(options));
 
       clearActivity();
@@ -401,14 +397,7 @@ export class Provider {
 
       responseLog(response.text);
 
-      if (response.usage) {
-        Stats.recordTokens(options.agentName || 'unknown', modelName, {
-          input: response.usage.inputTokens ?? response.usage.promptTokens ?? 0,
-          output: response.usage.outputTokens ?? response.usage.completionTokens ?? 0,
-          total: response.usage.totalTokens ?? 0,
-          cached: extractCachedTokens(response.usage),
-        });
-      }
+      this.recordUsage(options.agentName || 'unknown', modelName, response.usage);
 
       return response;
     } catch (error: any) {
@@ -419,12 +408,7 @@ export class Provider {
       if (error?.name === 'AbortError') throw error;
       if (error instanceof ContextLengthError) throw error;
       if (Provider.isContextLengthError(error)) {
-        const reduced = this.tryReduceMessages(messages, options._contextRetryLevel || 0);
-        if (reduced) {
-          tag('warning').log('Context length exceeded, retrying with reduced messages...');
-          return this.generateWithTools(reduced.messages, model, tools, { ...options, _contextRetryLevel: reduced.nextLevel });
-        }
-        throw new ContextLengthError(error.message || error.toString());
+        return this.recoverFromContextLength(error, messages, options, (m, o) => this.generateWithTools(m, model, tools, o));
       }
       if (error.constructor?.name === 'AI_APICallError') {
         responseLog(error.message);
@@ -440,52 +424,18 @@ export class Provider {
     setActivity(`🤖 Asking ${modelName} for structured output`, 'ai');
     promptLog(`Using model: ${modelName}`);
 
-    const telemetry = this.getTelemetry(options);
-    const config = this.mergeProviderOptions(
-      {
-        schema,
-        allowSystemInMessages: true,
-        ...(this.config.config || {}),
-        ...options,
-        model: modelToUse,
-        abortSignal: executionController.getAbortSignal(),
-      },
-      options.agentName
-    );
-    this.finalizeConfig(config, options, telemetry);
+    const config = this.buildGenerateConfig({ schema }, { model: modelToUse }, options);
 
     try {
       promptLog(messages[messages.length - 1].content);
       const response = await withRetry(async () => {
-        const timeout = config.timeout || 30000;
-        const cancel = { cancelled: false };
-        const controller = new AbortController();
-        const combinedSignal = AbortSignal.any([controller.signal, executionController.getAbortSignal()].filter(Boolean) as AbortSignal[]);
-        try {
-          return (await Promise.race([
-            generateObject({
-              messages,
-              ...config,
-              abortSignal: combinedSignal,
-            }),
-            rejectAfterIdle(timeout, cancel, controller),
-          ])) as any;
-        } finally {
-          cancel.cancelled = true;
-        }
+        return (await this.raceWithIdleTimeout((signal) => generateObject({ messages, ...config, abortSignal: signal }), config.timeout || 30000)) as any;
       }, this.getRetryOptions(options));
 
       clearActivity();
       responseLog(response.object);
 
-      if (response.usage) {
-        Stats.recordTokens(options.agentName || 'unknown', modelName, {
-          input: response.usage.inputTokens ?? response.usage.promptTokens ?? 0,
-          output: response.usage.outputTokens ?? response.usage.completionTokens ?? 0,
-          total: response.usage.totalTokens ?? 0,
-          cached: extractCachedTokens(response.usage),
-        });
-      }
+      this.recordUsage(options.agentName || 'unknown', modelName, response.usage);
 
       return response;
     } catch (error: any) {
@@ -493,12 +443,7 @@ export class Provider {
       if (error?.name === 'AbortError') throw error;
       if (error instanceof ContextLengthError) throw error;
       if (Provider.isContextLengthError(error)) {
-        const reduced = this.tryReduceMessages(messages, options._contextRetryLevel || 0);
-        if (reduced) {
-          tag('warning').log('Context length exceeded, retrying with reduced messages...');
-          return this.generateObject(reduced.messages, schema, model, { ...options, _contextRetryLevel: reduced.nextLevel });
-        }
-        throw new ContextLengthError(error.message || error.toString());
+        return this.recoverFromContextLength(error, messages, options, (m, o) => this.generateObject(m, schema, model, o));
       }
       throw new AiError(error.message || error.toString());
     }
@@ -506,7 +451,7 @@ export class Provider {
 
   static isContextLengthError(error: any): boolean {
     const msg = (error?.message || error?.toString() || '').toLowerCase();
-    return Provider.CONTEXT_LENGTH_PATTERNS.some((p) => msg.includes(p));
+    return CONTEXT_LENGTH_PATTERNS.some((p) => msg.includes(p));
   }
 
   static trimMessagesForRetry(messages: ModelMessage[]): ModelMessage[] | null {
@@ -677,13 +622,7 @@ export class Provider {
       clearActivity();
       responseLog(response.text);
 
-      if (response.usage) {
-        Stats.recordTokens('vision', this.getModelName(this.config.visionModel), {
-          input: response.usage.inputTokens ?? response.usage.promptTokens ?? 0,
-          output: response.usage.outputTokens ?? response.usage.completionTokens ?? 0,
-          total: response.usage.totalTokens ?? 0,
-        });
-      }
+      this.recordUsage('vision', this.getModelName(this.config.visionModel), response.usage);
 
       return response;
     } catch (error: any) {