@dotsetlabs/bellwether 1.0.2 → 1.0.3

package/dist/interview/interviewer.js

@@ -4,7 +4,7 @@ import { DEFAULT_PERSONA } from '../persona/builtins.js';
 import { getLogger, startTiming } from '../logging/logger.js';
 import { evaluateAssertions } from '../scenarios/evaluator.js';
 import { withTimeout, DEFAULT_TIMEOUTS, parallelLimit, createMutex } from '../utils/index.js';
-import { INTERVIEW, WORKFLOW, DISPLAY_LIMITS, SCHEMA_TESTING, OUTCOME_ASSESSMENT } from '../constants.js';
+import { INTERVIEW, WORKFLOW, DISPLAY_LIMITS, SCHEMA_TESTING, OUTCOME_ASSESSMENT, } from '../constants.js';
 import { generateSchemaTests } from './schema-test-generator.js';
 import { WorkflowDiscoverer } from '../workflow/discovery.js';
 import { WorkflowExecutor } from '../workflow/executor.js';
@@ -62,7 +62,8 @@ export class Interviewer {
         // Use multiple personas by default for better coverage
         // Fall back to DEFAULT_PERSONAS if no personas provided or empty array
         const providedPersonas = config?.personas;
-        this.personas = (providedPersonas && providedPersonas.length > 0) ? providedPersonas : DEFAULT_PERSONAS;
+        this.personas =
+            providedPersonas && providedPersonas.length > 0 ? providedPersonas : DEFAULT_PERSONAS;
         // Store cache reference for tool response and analysis caching
         this.cache = config?.cache;
         if (this.config.rateLimit?.enabled) {
@@ -270,12 +271,12 @@ export class Interviewer {
         };
         // Look for tools that reveal server constraints
         for (const toolName of INTERVIEW.CONSTRAINT_DISCOVERY_TOOLS) {
-            const tool = discovery.tools.find(t => t.name === toolName);
+            const tool = discovery.tools.find((t) => t.name === toolName);
             if (tool) {
                 try {
                     const result = await client.callTool(toolName, {});
                     if (result?.content) {
-                        const textContent = result.content.find(c => c.type === 'text');
+                        const textContent = result.content.find((c) => c.type === 'text');
                         if (textContent && 'text' in textContent) {
                             const text = String(textContent.text);
                             // Parse allowed directories from response
@@ -344,7 +345,7 @@ export class Interviewer {
         try {
             const parsed = JSON.parse(text);
             if (Array.isArray(parsed)) {
-                return parsed.filter(d => typeof d === 'string' && d.startsWith('/'));
+                return parsed.filter((d) => typeof d === 'string' && d.startsWith('/'));
             }
         }
         catch (error) {
@@ -443,7 +444,7 @@ export class Interviewer {
                 concurrency,
             }, 'Running persona interviews in parallel');
             // Create tasks for each persona
-            const personaTasks = this.personas.map(persona => async () => {
+            const personaTasks = this.personas.map((persona) => async () => {
                 progress.currentPersona = persona.name;
                 onProgress?.(progress);
                 const result = await this.interviewPersona(client, discovery, persona, toolCallMutex);
@@ -501,7 +502,10 @@ export class Interviewer {
             if (statefulEnabled) {
                 this.logger.info({ toolCount: orderedTools.length }, 'Stateful testing enabled');
             }
-            this.logger.info({ parallel: this.config.parallelTools && !statefulEnabled, concurrency: effectiveConcurrency }, 'Using check mode tool testing');
+            this.logger.info({
+                parallel: this.config.parallelTools && !statefulEnabled,
+                concurrency: effectiveConcurrency,
+            }, 'Using check mode tool testing');
             const statefulRunner = statefulEnabled
                 ? new StatefulTestRunner({ shareOutputs: statefulConfig?.shareOutputsBetweenTools ?? true })
                 : undefined;
@@ -516,13 +520,15 @@ export class Interviewer {
                 const toolData = toolInteractionsMap.get(profile.name);
                 if (toolData) {
                     toolData.interactions = profile.interactions;
-                    toolData.findingsByPersona = [{
+                    toolData.findingsByPersona = [
+                        {
                             personaId: 'check_mode',
                             personaName: 'Check Mode',
                             behavioralNotes: [],
                             limitations: [],
                             securityNotes: [],
-                        }];
+                        },
+                    ];
                 }
             }
             // Update persona stats with aggregated counts
@@ -557,7 +563,7 @@ export class Interviewer {
                         const scenarioResults = await this.executeToolScenarios(client, tool.name, customScenarios);
                         allScenarioResults.push(...scenarioResults);
                         // Convert scenarios to interview questions for integration with profiling
-                        questions = customScenarios.map(s => this.scenarioToQuestion(s));
+                        questions = customScenarios.map((s) => this.scenarioToQuestion(s));
                         // If not custom-only mode, also generate LLM questions (skip in fast CI mode)
                         if (!this.config.customScenariosOnly && !this.config.checkMode) {
                             const llmQuestions = await orchestrator.generateQuestions(tool, this.config.maxQuestionsPerTool, this.config.skipErrorTests);
@@ -568,7 +574,8 @@ export class Interviewer {
                         // No custom scenarios - generate questions
                         if (this.config.checkMode) {
                             // Fast CI mode: use fallback questions (no LLM call)
-                            questions = orchestrator.getFallbackQuestions(tool, this.config.skipErrorTests)
+                            questions = orchestrator
+                                .getFallbackQuestions(tool, this.config.skipErrorTests)
                                 .slice(0, this.config.maxQuestionsPerTool);
                         }
                         else {
@@ -589,8 +596,10 @@ export class Interviewer {
                             });
                             // If we have multiple failures, regenerate remaining questions with error context
                             // Skip in scenarios-only mode and fast CI mode
-                            if (!this.config.customScenariosOnly && !this.config.checkMode &&
-                                previousErrors.length >= 2 && personaInteractions.length < questions.length) {
+                            if (!this.config.customScenariosOnly &&
+                                !this.config.checkMode &&
+                                previousErrors.length >= 2 &&
+                                personaInteractions.length < questions.length) {
                                 const remaining = this.config.maxQuestionsPerTool - personaInteractions.length;
                                 if (remaining > 0) {
                                     this.logger.debug({ tool: tool.name, errors: previousErrors.length }, 'Regenerating questions after errors');
@@ -616,7 +625,7 @@ export class Interviewer {
                         };
                     }
                     else {
-                        personaProfile = await orchestrator.synthesizeToolProfile(tool, personaInteractions.map(i => ({
+                        personaProfile = await orchestrator.synthesizeToolProfile(tool, personaInteractions.map((i) => ({
                             question: i.question,
                             response: i.response,
                             error: i.error,
@@ -664,7 +673,9 @@ export class Interviewer {
             progress.promptsCompleted = 0;
             onProgress?.(progress);
             // Only create orchestrator if NOT in check mode (requires LLM)
-            const primaryOrchestrator = this.isCheckMode() ? null : this.createOrchestrator(this.personas[0]);
+            const primaryOrchestrator = this.isCheckMode()
+                ? null
+                : this.createOrchestrator(this.personas[0]);
             for (const prompt of discovery.prompts) {
                 progress.currentTool = `prompt:${prompt.name}`;
                 onProgress?.(progress);
@@ -678,7 +689,7 @@ export class Interviewer {
                     const scenarioResults = await this.executePromptScenarios(client, prompt.name, customScenarios);
                     allScenarioResults.push(...scenarioResults);
                     // Convert scenarios to prompt questions for profiling
-                    questions = customScenarios.map(s => ({
+                    questions = customScenarios.map((s) => ({
                         description: s.description,
                         args: s.args,
                     }));
@@ -688,7 +699,9 @@ export class Interviewer {
                         questions = [...questions, ...llmQuestions];
                     }
                 }
-                else if (!this.config.customScenariosOnly && !this.config.checkMode && primaryOrchestrator) {
+                else if (!this.config.customScenariosOnly &&
+                    !this.config.checkMode &&
+                    primaryOrchestrator) {
                     // No custom scenarios - generate LLM questions as usual
                     questions = await primaryOrchestrator.generatePromptQuestions(prompt, 2);
                 }
@@ -740,7 +753,7 @@ export class Interviewer {
                     };
                 }
                 else {
-                    profile = await primaryOrchestrator.synthesizePromptProfile(prompt, promptInteractions.map(i => ({
+                    profile = await primaryOrchestrator.synthesizePromptProfile(prompt, promptInteractions.map((i) => ({
                         question: i.question,
                         response: i.response,
                         error: i.error,
@@ -768,7 +781,9 @@ export class Interviewer {
             progress.resourcesCompleted = 0;
             onProgress?.(progress);
             // Only create orchestrator if NOT in check mode (requires LLM)
-            const primaryOrchestrator = this.isCheckMode() ? null : this.createOrchestrator(this.personas[0]);
+            const primaryOrchestrator = this.isCheckMode()
+                ? null
+                : this.createOrchestrator(this.personas[0]);
             for (const resource of discoveredResources) {
                 progress.currentTool = `resource:${resource.name}`;
                 onProgress?.(progress);
@@ -777,7 +792,9 @@ export class Interviewer {
                 let questions;
                 if (this.config.checkMode || !primaryOrchestrator) {
                     // Fast CI mode: use simple fallback question
-                    questions = [{ description: 'Basic resource read test', category: 'happy_path' }];
+                    questions = [
+                        { description: 'Basic resource read test', category: 'happy_path' },
+                    ];
                 }
                 else {
                     questions = await primaryOrchestrator.generateResourceQuestions(resource, 2);
@@ -787,8 +804,9 @@ export class Interviewer {
                     let response = null;
                     let error = null;
                     try {
+                        const abortController = new AbortController();
                         // Apply timeout to resource read to prevent indefinite hangs
-                        response = await withTimeout(client.readResource(resource.uri), this.config.resourceTimeout ?? DEFAULT_TIMEOUTS.resourceRead, `Resource read: ${resource.uri}`);
+                        response = await withTimeout(client.readResource(resource.uri, { signal: abortController.signal }), this.config.resourceTimeout ?? DEFAULT_TIMEOUTS.resourceRead, `Resource read: ${resource.uri}`, { abortController });
                         resourceReadCount++;
                     }
                     catch (e) {
@@ -829,7 +847,7 @@ export class Interviewer {
                     };
                 }
                 else {
-                    profile = await primaryOrchestrator.synthesizeResourceProfile(resource, resourceInteractions.map(i => ({
+                    profile = await primaryOrchestrator.synthesizeResourceProfile(resource, resourceInteractions.map((i) => ({
                         question: i.question,
                         response: i.response,
                         error: i.error,
@@ -838,13 +856,14 @@ export class Interviewer {
                 }
                 // Extract content preview from first successful read
                 let contentPreview;
-                const successfulRead = resourceInteractions.find(i => i.response && !i.error);
+                const successfulRead = resourceInteractions.find((i) => i.response && !i.error);
                 if (successfulRead?.response?.contents?.[0]) {
                     const content = successfulRead.response.contents[0];
                     if (content.text) {
-                        contentPreview = content.text.length > DISPLAY_LIMITS.CONTENT_TEXT_PREVIEW
-                            ? `${content.text.substring(0, DISPLAY_LIMITS.CONTENT_TEXT_PREVIEW)}...`
-                            : content.text;
+                        contentPreview =
+                            content.text.length > DISPLAY_LIMITS.CONTENT_TEXT_PREVIEW
+                                ? `${content.text.substring(0, DISPLAY_LIMITS.CONTENT_TEXT_PREVIEW)}...`
+                                : content.text;
                     }
                     else if (content.blob) {
                         contentPreview = `[Binary data: ${content.blob.length} bytes base64]`;
@@ -1058,7 +1077,7 @@ export class Interviewer {
                 if (response.isError) {
                     stats.errorCount++;
                     hadError = true;
-                    const errorContent = response.content?.find(c => c.type === 'text');
+                    const errorContent = response.content?.find((c) => c.type === 'text');
                     if (errorContent && 'text' in errorContent) {
                         error = String(errorContent.text);
                     }
@@ -1143,7 +1162,7 @@ export class Interviewer {
         // Extract allowed directories explicitly mentioned
         const allowedMatch = error.match(/allowed director(?:y|ies)[:\s]+([^\n]+)/i);
         if (allowedMatch) {
-            const dirs = allowedMatch[1].split(/[,\s]+/).filter(d => d.startsWith('/'));
+            const dirs = allowedMatch[1].split(/[,\s]+/).filter((d) => d.startsWith('/'));
             if (dirs.length > 0) {
                 const currentContext = orchestrator.getServerContext() ?? { allowedDirectories: [] };
                 const existingDirs = currentContext.allowedDirectories ?? [];
@@ -1197,7 +1216,7 @@ export class Interviewer {
                     toolCallMutex.release();
                 }
                 // Convert scenarios to interview questions
-                questions = customScenarios.map(s => this.scenarioToQuestion(s));
+                questions = customScenarios.map((s) => this.scenarioToQuestion(s));
                 // If not custom-only mode, also generate LLM questions
                 if (!this.config.customScenariosOnly) {
                     const llmQuestions = await orchestrator.generateQuestions(tool, this.config.maxQuestionsPerTool, this.config.skipErrorTests);
@@ -1231,7 +1250,8 @@ export class Interviewer {
                     });
                     // If we have multiple failures, regenerate remaining questions
                     if (!this.config.customScenariosOnly &&
-                        previousErrors.length >= 2 && personaInteractions.length < questions.length) {
+                        previousErrors.length >= 2 &&
+                        personaInteractions.length < questions.length) {
                         const remaining = this.config.maxQuestionsPerTool - personaInteractions.length;
                         if (remaining > 0) {
                             this.logger.debug({ tool: tool.name, errors: previousErrors.length }, 'Regenerating questions after errors');
@@ -1253,7 +1273,7 @@ export class Interviewer {
                 };
             }
             else {
-                personaProfile = await orchestrator.synthesizeToolProfile(tool, personaInteractions.map(i => ({
+                personaProfile = await orchestrator.synthesizeToolProfile(tool, personaInteractions.map((i) => ({
                     question: i.question,
                     response: i.response,
                     error: i.error,
@@ -1361,20 +1381,19 @@ export class Interviewer {
                 const results = await this.executeToolScenarios(client, tool.name, customScenarios);
                 scenarioResults.push(...results);
                 toolCallCount += results.length;
-                errorCount += results.filter(r => !r.passed).length;
+                errorCount += results.filter((r) => !r.passed).length;
             }
             finally {
                 toolCallMutex.release();
             }
             // Convert scenarios to interview questions
-            questions = customScenarios.map(s => this.scenarioToQuestion(s));
+            questions = customScenarios.map((s) => this.scenarioToQuestion(s));
         }
         else {
             // No custom scenarios - use fallback questions (check mode, no LLM)
             // We need an orchestrator for fallback questions, but we won't use LLM
             // Get fallback questions directly
-            questions = this.getFallbackQuestionsForTool(tool, this.config.skipErrorTests)
-                .slice(0, this.config.maxQuestionsPerTool);
+            questions = this.getFallbackQuestionsForTool(tool, this.config.skipErrorTests).slice(0, this.config.maxQuestionsPerTool);
         }
         // Execute warmup runs if configured (helps reduce cold-start timing variance)
         // Warmup runs are not recorded in interactions
@@ -1444,7 +1463,10 @@ export class Interviewer {
             // Generate simple analysis (no LLM in check mode)
             const analysis = this.generateSimpleAnalysis(error, !!response, 'Tool call succeeded.');
             const outcomeAssessment = this.assessOutcome(resolvedQuestion, response, error);
-            if (this.config.assertions?.enabled && outcomeAssessment.expected === 'success' && response && !response.isError) {
+            if (this.config.assertions?.enabled &&
+                outcomeAssessment.expected === 'success' &&
+                response &&
+                !response.isError) {
                 let schema = this.responseSchemas.get(tool.name);
                 if (!schema && this.config.assertions?.infer) {
                     const inferred = inferResponseSchema(response);
@@ -1546,7 +1568,7 @@ export class Interviewer {
             parallel: this.config.parallelTools,
         }, 'Running check mode tool testing');
         // Create tasks for each tool
-        const toolTasks = tools.map(tool => async () => {
+        const toolTasks = tools.map((tool) => async () => {
             progress.currentTool = tool.name;
             onProgress?.(progress);
             const result = await this.interviewToolInCheckMode(client, tool, toolCallMutex, options?.statefulRunner, options?.dependencyMap?.get(tool.name), options?.statefulConfig);
@@ -1575,7 +1597,7 @@ export class Interviewer {
         let totalErrorCount = 0;
         let totalQuestionsAsked = 0;
         for (const result of successfulResults) {
-            const tool = tools.find(t => t.name === result.toolName);
+            const tool = tools.find((t) => t.name === result.toolName);
             if (!tool)
                 continue;
             // Classify errors to separate tool correctness from environment issues
@@ -1618,7 +1640,7 @@ export class Interviewer {
         };
     }
     buildToolProgressSummary(result) {
-        const interactions = result.interactions.filter(i => !i.mocked);
+        const interactions = result.interactions.filter((i) => !i.mocked);
         const totalTests = interactions.length;
         let passedTests = 0;
         let validationTotal = 0;
@@ -1674,14 +1696,14 @@ export class Interviewer {
      */
     getScenariosForTool(toolName) {
         const scenarios = this.config.customScenarios?.toolScenarios ?? [];
-        return scenarios.filter(s => s.tool === toolName && !s.skip);
+        return scenarios.filter((s) => s.tool === toolName && !s.skip);
     }
     /**
      * Get custom scenarios for a specific prompt.
      */
     getScenariosForPrompt(promptName) {
         const scenarios = this.config.customScenarios?.promptScenarios ?? [];
-        return scenarios.filter(s => s.prompt === promptName && !s.skip);
+        return scenarios.filter((s) => s.prompt === promptName && !s.skip);
     }
     /**
      * Execute custom test scenarios for a tool.
@@ -1708,7 +1730,7 @@ export class Interviewer {
                     response = result.response;
                     isError = response?.isError ?? false;
                     if (isError) {
-                        const errorContent = response?.content?.find(c => c.type === 'text');
+                        const errorContent = response?.content?.find((c) => c.type === 'text');
                         if (errorContent && 'text' in errorContent) {
                             error = String(errorContent.text);
                         }
@@ -1728,7 +1750,7 @@ export class Interviewer {
                 ? evaluateAssertions(scenario.assertions, response, isError)
                 : [];
             // Scenario passes if no error (or expected error) and all assertions pass
-            const allAssertionsPassed = assertionResults.every(r => r.passed);
+            const allAssertionsPassed = assertionResults.every((r) => r.passed);
             const passed = allAssertionsPassed && (!isError || scenario.category === 'error_handling');
             const result = {
                 scenario,
@@ -1771,9 +1793,9 @@ export class Interviewer {
             const assertionResults = scenario.assertions
                 ? evaluateAssertions(scenario.assertions, response, !!error)
                 : [];
-            const allAssertionsPassed = assertionResults.every(r => r.passed);
+            const allAssertionsPassed = assertionResults.every((r) => r.passed);
             // Check if this scenario expects an error (has an assertion checking for 'error' to exist)
-            const expectsError = scenario.assertions?.some(a => a.path === 'error' && a.condition === 'exists') ?? false;
+            const expectsError = scenario.assertions?.some((a) => a.path === 'error' && a.condition === 'exists') ?? false;
             // Scenario passes if assertions pass AND (no error OR scenario expects error)
             const passed = allAssertionsPassed && (!error || expectsError);
             const result = {
@@ -1824,7 +1846,7 @@ export class Interviewer {
                     discoveredCount = discovered.length;
                     this.logger.info({
                         count: discoveredCount,
-                        workflows: discovered.map(w => w.name),
+                        workflows: discovered.map((w) => w.name),
                     }, 'Discovered workflows');
                 }
                 else {
@@ -1904,7 +1926,7 @@ export class Interviewer {
             }
         }
         // Build summary
-        const successfulCount = results.filter(r => r.success).length;
+        const successfulCount = results.filter((r) => r.success).length;
         const summary = {
             workflowCount: results.length,
             successfulCount,
@@ -1923,9 +1945,7 @@ export class Interviewer {
     }
 }
 function summarizeAssertions(interactions) {
-    const allResults = interactions
-        .filter((i) => !i.mocked)
-        .flatMap((i) => i.assertionResults ?? []);
+    const allResults = interactions.filter((i) => !i.mocked).flatMap((i) => i.assertionResults ?? []);
     if (allResults.length === 0)
         return undefined;
     const passed = allResults.filter((r) => r.passed).length;

package/dist/interview/orchestrator.js

@@ -44,7 +44,11 @@ function categorizeLLMError(error) {
     if (message.includes('empty or whitespace') ||
         message.includes('token exhaustion') ||
         message.includes('unexpected end of json')) {
-        return { category: 'format_error', isRetryable: true, message: 'LLM returned empty response (possible token exhaustion)' };
+        return {
+            category: 'format_error',
+            isRetryable: true,
+            message: 'LLM returned empty response (possible token exhaustion)',
+        };
     }
     // Check for format errors (LLM returned wrong format) - retryable once
     if (message.includes('invalid question format') ||
@@ -53,7 +57,11 @@ function categorizeLLMError(error) {
         message.includes('not valid json')) {
         return { category: 'format_error', isRetryable: true, message: 'LLM returned invalid format' };
     }
-    return { category: 'unknown', isRetryable: false, message: error instanceof Error ? error.message : String(error) };
+    return {
+        category: 'unknown',
+        isRetryable: false,
+        message: error instanceof Error ? error.message : String(error),
+    };
 }
 /**
  * Orchestrator uses an LLM to generate interview questions and synthesize findings.
@@ -248,11 +256,13 @@ export class Orchestrator {
         for (let attempt = 0; attempt <= maxRetries; attempt++) {
             let rawResponse;
             try {
+                const abortController = new AbortController();
                 // Apply timeout to LLM call - use streaming if enabled
                 const response = await withTimeout(this.completeWithStreaming(prompt, {
                     ...COMPLETION_OPTIONS.questionGeneration,
                     systemPrompt: this.getSystemPrompt(),
-                }, `generate-questions:${tool.name}`), DEFAULT_TIMEOUTS.questionGeneration, `Question generation for ${tool.name}`);
+                    signal: abortController.signal,
+                }, `generate-questions:${tool.name}`), DEFAULT_TIMEOUTS.questionGeneration, `Question generation for ${tool.name}`, { abortController });
                 rawResponse = response;
                 // Check for empty/whitespace-only responses (common with token exhaustion)
                 const trimmed = response.trim();
@@ -304,7 +314,7 @@ export class Orchestrator {
                 // Wait before retry with exponential backoff
                 if (attempt < maxRetries) {
                     const delay = Math.min(RETRY.INITIAL_DELAY * Math.pow(2, attempt), RETRY.MAX_DELAY);
-                    await new Promise(resolve => setTimeout(resolve, delay));
+                    await new Promise((resolve) => setTimeout(resolve, delay));
                 }
             }
         }
@@ -361,7 +371,7 @@ export class Orchestrator {
                 return `Tool returned an error: ${error}`;
             }
             if (response?.content) {
-                const textContent = response.content.find(c => c.type === 'text');
+                const textContent = response.content.find((c) => c.type === 'text');
                 if (textContent && 'text' in textContent) {
                     return `Tool returned: ${String(textContent.text).substring(0, DISPLAY_LIMITS.TOOL_RESPONSE_PREVIEW)}`;
                 }
@@ -397,7 +407,7 @@ export class Orchestrator {
             return {
                 name: tool.name,
                 description: tool.description ?? 'No description provided',
-                behavioralNotes: interactions.map(i => i.analysis).filter(a => a),
+                behavioralNotes: interactions.map((i) => i.analysis).filter((a) => a),
                 limitations: [],
                 securityNotes: [],
             };
@@ -694,7 +704,9 @@ export class Orchestrator {
         // Check name-based hints
         if (lowerName.includes('path') || lowerName.includes('file')) {
             const baseDir = this.serverContext?.allowedDirectories?.[0] ?? '/tmp';
-            if (lowerName.includes('dir') || lowerName.includes('directory') || lowerName.includes('folder')) {
+            if (lowerName.includes('dir') ||
+                lowerName.includes('directory') ||
+                lowerName.includes('folder')) {
                 return baseDir;
             }
             return `${baseDir}/test.txt`;
@@ -717,12 +729,16 @@ export class Orchestrator {
             }
             return 'test-name';
         }
-        if (lowerName.includes('query') || lowerName.includes('search') || lowerName.includes('filter')) {
+        if (lowerName.includes('query') ||
+            lowerName.includes('search') ||
+            lowerName.includes('filter')) {
             // Use a more realistic search term based on description
             if (description.includes('movie') || description.includes('film')) {
                 return 'The Matrix';
             }
-            if (description.includes('music') || description.includes('song') || description.includes('artist')) {
+            if (description.includes('music') ||
+                description.includes('song') ||
+                description.includes('artist')) {
                 return 'Beatles';
             }
             if (description.includes('book') || description.includes('author')) {
@@ -733,10 +749,14 @@ export class Orchestrator {
         if (lowerName.includes('title')) {
             return 'Test Title';
         }
-        if (lowerName.includes('description') || lowerName.includes('summary') || lowerName.includes('text')) {
+        if (lowerName.includes('description') ||
+            lowerName.includes('summary') ||
+            lowerName.includes('text')) {
             return 'This is a test description for validation purposes.';
         }
-        if (lowerName.includes('content') || lowerName.includes('body') || lowerName.includes('message')) {
+        if (lowerName.includes('content') ||
+            lowerName.includes('body') ||
+            lowerName.includes('message')) {
             return 'Test content for the operation.';
         }
         if (lowerName.includes('comment')) {
@@ -745,7 +765,9 @@ export class Orchestrator {
         if (lowerName.includes('code') || lowerName.includes('snippet')) {
             return 'function example() { return "Hello"; }';
         }
-        if (lowerName.includes('pattern') || lowerName.includes('glob') || lowerName.includes('regex')) {
+        if (lowerName.includes('pattern') ||
+            lowerName.includes('glob') ||
+            lowerName.includes('regex')) {
             return '*.txt';
         }
         if (lowerName.includes('format') || lowerName.includes('type')) {
@@ -830,13 +852,17 @@ export class Orchestrator {
         if (lowerName.includes('count') || lowerName.includes('limit') || lowerName.includes('num')) {
             return 10;
         }
-        if (lowerName.includes('enabled') || lowerName.includes('active') || lowerName.includes('flag')) {
+        if (lowerName.includes('enabled') ||
+            lowerName.includes('active') ||
+            lowerName.includes('flag')) {
             return true;
         }
         if (lowerName.includes('list') || lowerName.includes('items') || lowerName.includes('array')) {
             return [];
         }
-        if (lowerName.includes('config') || lowerName.includes('options') || lowerName.includes('settings')) {
+        if (lowerName.includes('config') ||
+            lowerName.includes('options') ||
+            lowerName.includes('settings')) {
             return {};
         }
         return 'test';
@@ -919,8 +945,7 @@ export class Orchestrator {
         if (!schema?.properties)
             return tests;
         const required = new Set(schema.required ?? []);
-        const optionalParams = Object.entries(schema.properties)
-            .filter(([name]) => !required.has(name));
+        const optionalParams = Object.entries(schema.properties).filter(([name]) => !required.has(name));
         if (optionalParams.length === 0)
             return tests;
         const allArgs = {};
@@ -1053,7 +1078,7 @@ export class Orchestrator {
             const result = this.llm.parseJSON(response);
             // Extract example output from first successful interaction
             let exampleOutput;
-            const successful = interactions.find(i => !i.error && i.response?.messages?.length);
+            const successful = interactions.find((i) => !i.error && i.response?.messages?.length);
             if (successful?.response) {
                 const firstMsg = successful.response.messages[0];
                 if (firstMsg?.content?.type === 'text' && firstMsg.content.text) {
@@ -1078,7 +1103,7 @@ export class Orchestrator {
                 name: prompt.name,
                 description: prompt.description ?? 'No description provided',
                 arguments: prompt.arguments ?? [],
-                behavioralNotes: interactions.map(i => i.analysis).filter(a => a),
+                behavioralNotes: interactions.map((i) => i.analysis).filter((a) => a),
                 limitations: [],
             };
         }
@@ -1102,7 +1127,7 @@ export class Orchestrator {
             args,
         });
         // If there are optional args, add a test with all args
-        const optionalArgs = prompt.arguments?.filter(a => !a.required) ?? [];
+        const optionalArgs = prompt.arguments?.filter((a) => !a.required) ?? [];
         if (optionalArgs.length > 0) {
             const allArgs = { ...args };
             for (const arg of optionalArgs) {
@@ -1210,10 +1235,12 @@ Description: ${resource.description ?? 'No description'}
 MIME Type: ${resource.mimeType ?? 'Not specified'}
 
 Test interactions:
-${interactions.map((i, idx) => `
+${interactions
+            .map((i, idx) => `
 ${idx + 1}. ${i.question.description}
    ${i.error ? `Error: ${i.error}` : `Analysis: ${i.analysis}`}
-`).join('')}
+`)
+            .join('')}
 
 Generate a JSON object with:
 {
@@ -1247,7 +1274,7 @@ Return ONLY valid JSON, no explanation.`;
                 name: resource.name,
                 description: resource.description ?? 'No description provided',
                 mimeType: resource.mimeType,
-                behavioralNotes: interactions.map(i => i.analysis).filter(a => a),
+                behavioralNotes: interactions.map((i) => i.analysis).filter((a) => a),
                 limitations: [],
             };
         }