agentic-qe 3.14.3 → 3.14.5

package/dist/coordination/consensus/providers/ollama-provider.js

@@ -2,7 +2,7 @@
  * Agentic QE v3 - Ollama Model Provider
  * Local/on-prem model support for multi-model consensus verification
  *
- * Ollama provides access to local LLMs (Llama, Mistral, Codellama, etc.)
+ * Ollama provides access to local LLMs (Qwen3, gpt-oss, Llama 4, Gemma, etc.)
  * for privacy-sensitive verification or air-gapped environments.
  *
  * @see https://ollama.ai
@@ -20,10 +20,10 @@ import { resolveOllamaBaseUrl } from '../../../shared/llm/ollama-url.js';
  */
 export function getRecommendedOllamaModels() {
     return [
-        'llama3.1:70b', // Best reasoning
-        'codellama:34b', // Best for code analysis
-        'qwen2.5:72b', // Strong multilingual
-        'mistral', // Good balance
+        'qwen3-coder:30b', // Best for code analysis (ADR-111 default)
+        'gpt-oss:120b', // Best reasoning
+        'qwen3:30b-a3b', // Strong general-purpose MoE
+        'llama4:scout', // Different model family for consensus diversity
     ];
 }
 /**
@@ -31,10 +31,9 @@ export function getRecommendedOllamaModels() {
  */
 export function getLightweightOllamaModels() {
     return [
-        'llama3.2:3b',
-        'phi3',
-        'gemma2:2b',
-        'qwen2.5:7b',
+        'gpt-oss:20b',
+        'qwen3:30b-a3b', // MoE: ~3B active params, fast despite its size
+        'gemma4',
     ];
 }
 /**
@@ -42,10 +41,9 @@ export function getLightweightOllamaModels() {
  */
 export function getCodeOllamaModels() {
     return [
-        'codellama:34b',
-        'qwen2.5-coder',
-        'deepseek-coder-v2',
-        'starcoder2:15b',
+        'qwen3-coder:30b',
+        'gpt-oss:120b',
+        'gpt-oss:20b',
     ];
 }
 // ============================================================================
@@ -53,7 +51,7 @@ export function getCodeOllamaModels() {
 // ============================================================================
 const DEFAULT_CONFIG = {
     baseUrl: resolveOllamaBaseUrl(OLLAMA_CONSTANTS.DEFAULT_BASE_URL),
-    defaultModel: 'llama3.1',
+    defaultModel: 'qwen3-coder:30b',
     defaultTimeout: CONSENSUS_CONSTANTS.OLLAMA_TIMEOUT_MS, // 5 minutes (local models can be slow)
     maxRetries: 2,
     retryDelayMs: CONSENSUS_CONSTANTS.OLLAMA_RETRY_DELAY_MS,
@@ -69,7 +67,7 @@ const DEFAULT_CONFIG = {
  * ```typescript
  * const provider = new OllamaModelProvider({
  *   baseUrl: 'http://localhost:11434',
- *   defaultModel: 'codellama:34b',
+ *   defaultModel: 'qwen3-coder:30b',
  * });
  *
  * const result = await provider.complete('Analyze this code for vulnerabilities...');
@@ -294,18 +292,18 @@ Format your response with: verdict (confirmed/rejected/uncertain), confidence (0
  *
  * @example
  * ```typescript
- * // Default (localhost:11434, llama3.1)
+ * // Default (localhost:11434, qwen3-coder:30b)
  * const provider = createOllamaProvider();
  *
  * // Custom model
  * const provider = createOllamaProvider({
- *   defaultModel: 'codellama:34b',
+ *   defaultModel: 'gpt-oss:20b',
  * });
  *
  * // Remote Ollama server
  * const provider = createOllamaProvider({
  *   baseUrl: 'http://gpu-server:11434',
- *   defaultModel: 'llama3.1:70b',
+ *   defaultModel: 'gpt-oss:120b',
  * });
  * ```
  */

package/dist/coordination/consensus/providers/openai-provider.d.ts

@@ -3,7 +3,7 @@
  * MM-004: OpenAI implementation for multi-model consensus verification
  *
  * Provides security finding verification using OpenAI GPT models.
- * Supports GPT-4, GPT-4-turbo with configurable parameters.
+ * Supports the GPT-6 family (sol, luna, astra) with configurable parameters.
  *
  * @see docs/plans/AQE_V3_IMPROVEMENTS_PLAN.md - Phase 2: Multi-Model Verification
  */
@@ -12,7 +12,17 @@ import { BaseModelProvider } from '../model-provider';
 /**
  * OpenAI-specific model versions
  */
-export type OpenAIModel = 'gpt-4' | 'gpt-4-turbo' | 'gpt-4-turbo-preview' | 'gpt-4-0125-preview' | 'gpt-4-1106-preview';
+export type OpenAIModel = 'gpt-6-sol' | 'gpt-6-luna' | 'gpt-6-astra' | 'gpt-5.6-sol' | 'gpt-5.4-mini'
+/** @deprecated Legacy GPT-4 generation — kept for config compatibility only. */
+ | 'gpt-4'
+/** @deprecated Legacy GPT-4 generation — kept for config compatibility only. */
+ | 'gpt-4-turbo'
+/** @deprecated Legacy GPT-4 generation — kept for config compatibility only. */
+ | 'gpt-4-turbo-preview'
+/** @deprecated Legacy GPT-4 generation — kept for config compatibility only. */
+ | 'gpt-4-0125-preview'
+/** @deprecated Legacy GPT-4 generation — kept for config compatibility only. */
+ | 'gpt-4-1106-preview';
 /**
  * Configuration for OpenAI provider
  */
@@ -45,7 +55,7 @@ export interface OpenAIProviderConfig {
  * ```typescript
  * const provider = new OpenAIModelProvider({
  *   apiKey: process.env.OPENAI_API_KEY,
- *   defaultModel: 'gpt-4-turbo',
+ *   defaultModel: 'gpt-6-sol',
  * });
  *
  * const response = await provider.complete(prompt);
@@ -120,7 +130,7 @@ export declare class OpenAIModelProvider extends BaseModelProvider {
  * ```typescript
  * const provider = createOpenAIProvider({
  *   apiKey: process.env.OPENAI_API_KEY,
- *   defaultModel: 'gpt-4-turbo',
+ *   defaultModel: 'gpt-6-sol',
  * });
  * ```
  */

package/dist/coordination/consensus/providers/openai-provider.js

@@ -3,13 +3,30 @@
  * MM-004: OpenAI implementation for multi-model consensus verification
  *
  * Provides security finding verification using OpenAI GPT models.
- * Supports GPT-4, GPT-4-turbo with configurable parameters.
+ * Supports the GPT-6 family (sol, luna, astra) with configurable parameters.
  *
  * @see docs/plans/AQE_V3_IMPROVEMENTS_PLAN.md - Phase 2: Multi-Model Verification
  */
 import { BaseModelProvider, } from '../model-provider.js';
 import { CONSENSUS_CONSTANTS } from '../../constants.js';
 import { toErrorMessage, toError } from '../../../shared/error-utils.js';
+import { applyOpenAIParams } from '../../../shared/llm/openai-params.js';
+/**
+ * Cost per million tokens (USD), from the OpenAI API catalog (2026-09-27).
+ * Legacy GPT-4 rows are kept so configs pinned to them still price correctly.
+ */
+const OPENAI_MODEL_COSTS = {
+    'gpt-6-astra': { input: 10, output: 50 },
+    'gpt-6-sol': { input: 2, output: 10 },
+    'gpt-6-luna': { input: 0.1, output: 0.5 },
+    'gpt-5.4-mini': { input: 0.75, output: 4.5 },
+    // Deprecated (legacy GPT-4 generation)
+    'gpt-4': { input: 30, output: 60 },
+    'gpt-4-turbo': { input: 10, output: 30 },
+    'gpt-4-turbo-preview': { input: 10, output: 30 },
+    'gpt-4-0125-preview': { input: 10, output: 30 },
+    'gpt-4-1106-preview': { input: 10, output: 30 },
+};
 // ============================================================================
 // OpenAI Model Provider Implementation
 // ============================================================================
@@ -24,7 +41,7 @@ import { toErrorMessage, toError } from '../../../shared/error-utils.js';
  * ```typescript
  * const provider = new OpenAIModelProvider({
  *   apiKey: process.env.OPENAI_API_KEY,
- *   defaultModel: 'gpt-4-turbo',
+ *   defaultModel: 'gpt-6-sol',
  * });
  *
  * const response = await provider.complete(prompt);
@@ -34,19 +51,17 @@ export class OpenAIModelProvider extends BaseModelProvider {
     id = 'openai';
     name = 'OpenAI GPT';
     type = 'openai';
-    // Cost per million tokens (as of 2024)
-    // GPT-4: $30 input, $60 output per 1M tokens
-    // GPT-4-turbo: $10 input, $30 output per 1M tokens
+    // Overridden per model from OPENAI_MODEL_COSTS in the constructor
     costPerToken = {
-        input: 10 / 1_000_000, // Will be overridden per model
-        output: 30 / 1_000_000,
+        input: 2 / 1_000_000,
+        output: 10 / 1_000_000,
     };
     supportedModels = [
-        'gpt-4',
-        'gpt-4-turbo',
-        'gpt-4-turbo-preview',
-        'gpt-4-0125-preview',
-        'gpt-4-1106-preview',
+        'gpt-6-sol',
+        'gpt-6-luna',
+        'gpt-6-astra',
+        'gpt-5.6-sol',
+        'gpt-5.4-mini',
     ];
     config;
     apiKey;
@@ -69,7 +84,7 @@ export class OpenAIModelProvider extends BaseModelProvider {
         this.config = {
             apiKey,
             organization: config.organization || '',
-            defaultModel: config.defaultModel || 'gpt-4-turbo',
+            defaultModel: config.defaultModel || 'gpt-6-sol',
             baseUrl: this.baseUrl,
             defaultTimeout: config.defaultTimeout || CONSENSUS_CONSTANTS.MODEL_TIMEOUT_MS,
             maxRetries: config.maxRetries ?? CONSENSUS_CONSTANTS.DEFAULT_RETRY_ATTEMPTS,
@@ -105,12 +120,7 @@ export class OpenAIModelProvider extends BaseModelProvider {
                 content: prompt,
             },
         ];
-        const request = {
-            model,
-            messages,
-            temperature,
-            max_tokens: maxTokens,
-        };
+        const request = applyOpenAIParams({ model, messages }, model, { maxTokens, temperature });
         if (this.config.enableLogging) {
             console.log(`[OpenAI] Sending request to ${model}`);
         }
@@ -243,20 +253,12 @@ Focus on accuracy over speed. It's better to mark something as "INCONCLUSIVE" if
      * Update cost based on model selection
      */
     updateCostForModel(model) {
-        if (model === 'gpt-4' && !model.includes('turbo')) {
-            // GPT-4: $30 input, $60 output per 1M tokens
-            this.costPerToken = {
-                input: 30 / 1_000_000,
-                output: 60 / 1_000_000,
-            };
-        }
-        else {
-            // GPT-4-turbo: $10 input, $30 output per 1M tokens
-            this.costPerToken = {
-                input: 10 / 1_000_000,
-                output: 30 / 1_000_000,
-            };
-        }
+        // Unknown models are priced as the default tier (gpt-6-sol)
+        const pricing = OPENAI_MODEL_COSTS[model] ?? OPENAI_MODEL_COSTS['gpt-6-sol'];
+        this.costPerToken = {
+            input: pricing.input / 1_000_000,
+            output: pricing.output / 1_000_000,
+        };
     }
     /**
      * Sleep for specified milliseconds
@@ -284,7 +286,7 @@ Focus on accuracy over speed. It's better to mark something as "INCONCLUSIVE" if
  * ```typescript
  * const provider = createOpenAIProvider({
  *   apiKey: process.env.OPENAI_API_KEY,
- *   defaultModel: 'gpt-4-turbo',
+ *   defaultModel: 'gpt-6-sol',
  * });
  * ```
  */

package/dist/coordination/consensus/providers/openrouter-provider.d.ts

@@ -14,7 +14,7 @@ import { BaseModelProvider } from '../model-provider';
  * Popular models available through OpenRouter
  * OpenRouter supports 100+ models - these are the most useful for security verification
  */
-export type OpenRouterModel = 'anthropic/claude-3.5-sonnet' | 'anthropic/claude-3-opus' | 'anthropic/claude-3-sonnet' | 'anthropic/claude-3-haiku' | 'openai/gpt-4-turbo' | 'openai/gpt-4' | 'openai/gpt-4o' | 'openai/gpt-4o-mini' | 'google/gemini-pro-1.5' | 'google/gemini-flash-1.5' | 'meta-llama/llama-3.1-405b-instruct' | 'meta-llama/llama-3.1-70b-instruct' | 'meta-llama/llama-3.1-8b-instruct' | 'mistralai/mistral-large' | 'mistralai/mistral-medium' | 'mistralai/mixtral-8x22b-instruct' | 'deepseek/deepseek-chat' | 'deepseek/deepseek-coder' | 'qwen/qwen-2.5-72b-instruct' | 'cohere/command-r-plus' | string;
+export type OpenRouterModel = 'anthropic/claude-sonnet-5' | 'anthropic/claude-opus-5' | 'anthropic/claude-opus-5.5' | 'anthropic/claude-haiku-4.5' | 'openai/gpt-6-sol' | 'openai/gpt-6-luna' | 'openai/gpt-6-astra' | 'openai/gpt-oss-120b' | 'google/gemini-3.8-flash' | 'google/gemini-3.1-pro-preview' | 'google/gemini-2.5-pro' | 'google/gemini-2.5-flash' | 'meta-llama/llama-4-maverick' | 'meta-llama/llama-4-scout' | 'mistralai/mistral-large-2512' | 'deepseek/deepseek-v4-pro' | 'deepseek/deepseek-v4.1-flash' | 'qwen/qwen3-coder' | string;
 /**
  * Model tier for cost/capability classification
  */
@@ -67,7 +67,7 @@ export declare function getCostOptimizedModels(): OpenRouterModel[];
  * ```typescript
  * const provider = new OpenRouterModelProvider({
  *   apiKey: process.env.OPENROUTER_API_KEY,
- *   defaultModel: 'anthropic/claude-3.5-sonnet',
+ *   defaultModel: 'anthropic/claude-sonnet-5',
  * });
  *
  * const result = await provider.complete('Analyze this code for vulnerabilities...');
@@ -135,13 +135,13 @@ export declare class OpenRouterModelProvider extends BaseModelProvider {
  *
  * // With specific model
  * const provider = createOpenRouterProvider({
- *   defaultModel: 'meta-llama/llama-3.1-70b-instruct',
+ *   defaultModel: 'meta-llama/llama-4-maverick',
  * });
  *
  * // With fallback models
  * const provider = createOpenRouterProvider({
- *   defaultModel: 'anthropic/claude-3.5-sonnet',
- *   fallbackModels: ['openai/gpt-4o', 'google/gemini-pro-1.5'],
+ *   defaultModel: 'anthropic/claude-sonnet-5',
+ *   fallbackModels: ['openai/gpt-6-sol', 'google/gemini-2.5-pro'],
  * });
  * ```
  */

package/dist/coordination/consensus/providers/openrouter-provider.js

@@ -10,42 +10,69 @@
  */
 import { BaseModelProvider } from '../model-provider.js';
 import { toErrorMessage, toError } from '../../../shared/error-utils.js';
+import { resolveModelPricing } from '../../../shared/llm/cost-tracker.js';
 // ============================================================================
 // Model Cost and Tier Information
 // ============================================================================
 /**
- * Cost per 1M tokens for popular models (input/output)
- * Updated as of 2024 - check OpenRouter for current prices
+ * Cost per 1M tokens for current models (input/output)
+ * From the OpenRouter public catalog (2026-09-27) - check OpenRouter for current prices
  */
 const MODEL_COSTS = {
     // Claude models
-    'anthropic/claude-3.5-sonnet': { input: 3, output: 15, tier: 'standard' },
-    'anthropic/claude-3-opus': { input: 15, output: 75, tier: 'premium' },
-    'anthropic/claude-3-sonnet': { input: 3, output: 15, tier: 'standard' },
-    'anthropic/claude-3-haiku': { input: 0.25, output: 1.25, tier: 'cheap' },
+    'anthropic/claude-sonnet-5': { input: 2, output: 10, tier: 'standard' },
+    'anthropic/claude-opus-5': { input: 5, output: 25, tier: 'premium' },
+    'anthropic/claude-opus-5.5': { input: 4, output: 20, tier: 'premium' },
+    'anthropic/claude-haiku-4.5': { input: 1, output: 5, tier: 'cheap' },
     // OpenAI models
-    'openai/gpt-4-turbo': { input: 10, output: 30, tier: 'standard' },
-    'openai/gpt-4': { input: 30, output: 60, tier: 'premium' },
-    'openai/gpt-4o': { input: 5, output: 15, tier: 'standard' },
-    'openai/gpt-4o-mini': { input: 0.15, output: 0.6, tier: 'cheap' },
+    'openai/gpt-6-astra': { input: 10, output: 50, tier: 'premium' },
+    'openai/gpt-6-sol': { input: 2, output: 10, tier: 'standard' },
+    'openai/gpt-6-luna': { input: 0.1, output: 0.5, tier: 'cheap' },
+    'openai/gpt-oss-120b': { input: 0.15, output: 0.6, tier: 'cheap' },
     // Google models
-    'google/gemini-pro-1.5': { input: 3.5, output: 10.5, tier: 'standard' },
-    'google/gemini-flash-1.5': { input: 0.075, output: 0.3, tier: 'cheap' },
+    'google/gemini-3.1-pro-preview': { input: 2, output: 12, tier: 'standard' },
+    'google/gemini-2.5-pro': { input: 1.25, output: 10, tier: 'standard' },
+    'google/gemini-3.8-flash': { input: 0.75, output: 3.75, tier: 'cheap' },
+    'google/gemini-2.5-flash': { input: 0.3, output: 2.5, tier: 'cheap' },
     // Llama models
-    'meta-llama/llama-3.1-405b-instruct': { input: 2.7, output: 2.7, tier: 'standard' },
-    'meta-llama/llama-3.1-70b-instruct': { input: 0.52, output: 0.75, tier: 'cheap' },
-    'meta-llama/llama-3.1-8b-instruct': { input: 0.055, output: 0.055, tier: 'free' },
+    'meta-llama/llama-4-maverick': { input: 0.1875, output: 0.6525, tier: 'cheap' },
+    'meta-llama/llama-4-scout': { input: 0.1, output: 0.3, tier: 'cheap' },
     // Mistral models
-    'mistralai/mistral-large': { input: 3, output: 9, tier: 'standard' },
-    'mistralai/mixtral-8x22b-instruct': { input: 0.65, output: 0.65, tier: 'cheap' },
+    'mistralai/mistral-large-2512': { input: 0.5, output: 1.5, tier: 'cheap' },
     // DeepSeek models
-    'deepseek/deepseek-chat': { input: 0.14, output: 0.28, tier: 'cheap' },
-    'deepseek/deepseek-coder': { input: 0.14, output: 0.28, tier: 'cheap' },
+    'deepseek/deepseek-v4-pro': { input: 0.348, output: 0.696, tier: 'cheap' },
+    'deepseek/deepseek-v4.1-flash': { input: 0.035, output: 0.29, tier: 'cheap' },
     // Qwen models
-    'qwen/qwen-2.5-72b-instruct': { input: 0.35, output: 0.4, tier: 'cheap' },
-    // Cohere models
-    'cohere/command-r-plus': { input: 2.5, output: 10, tier: 'standard' },
+    'qwen/qwen3-coder': { input: 0.3, output: 1, tier: 'cheap' },
 };
+/**
+ * Deprecated models no longer served by OpenRouter. Pricing only, so historical
+ * usage still costs correctly — excluded from supported models and tier lists.
+ */
+const DEPRECATED_MODEL_COSTS = {
+    'anthropic/claude-3.5-sonnet': { input: 3, output: 15 },
+    'anthropic/claude-3-opus': { input: 15, output: 75 },
+    'anthropic/claude-3-sonnet': { input: 3, output: 15 },
+    'anthropic/claude-3-haiku': { input: 0.25, output: 1.25 },
+    'openai/gpt-4-turbo': { input: 10, output: 30 },
+    'openai/gpt-4': { input: 30, output: 60 },
+    'openai/gpt-4o': { input: 5, output: 15 },
+    'openai/gpt-4o-mini': { input: 0.15, output: 0.6 },
+    'google/gemini-pro-1.5': { input: 3.5, output: 10.5 },
+    'google/gemini-flash-1.5': { input: 0.075, output: 0.3 },
+    'meta-llama/llama-3.1-405b-instruct': { input: 2.7, output: 2.7 },
+    'meta-llama/llama-3.1-70b-instruct': { input: 0.52, output: 0.75 },
+    'meta-llama/llama-3.1-8b-instruct': { input: 0.055, output: 0.055 },
+    'mistralai/mistral-large': { input: 3, output: 9 },
+    'mistralai/mixtral-8x22b-instruct': { input: 0.65, output: 0.65 },
+    'deepseek/deepseek-chat': { input: 0.14, output: 0.28 },
+    'deepseek/deepseek-coder': { input: 0.14, output: 0.28 },
+    'qwen/qwen-2.5-72b-instruct': { input: 0.35, output: 0.4 },
+    'cohere/command-r-plus': { input: 2.5, output: 10 },
+};
+function lookupModelCost(model) {
+    return MODEL_COSTS[model] ?? DEPRECATED_MODEL_COSTS[model] ?? resolveModelPricing(model);
+}
 /**
  * Get models by tier for cost optimization
  */
@@ -60,10 +87,10 @@ export function getModelsByTier(tier) {
  */
 export function getRecommendedSecurityModels() {
     return [
-        'anthropic/claude-3.5-sonnet', // Best for security analysis
-        'openai/gpt-4o', // Strong reasoning
-        'google/gemini-pro-1.5', // Good code understanding
-        'mistralai/mistral-large', // Cost-effective alternative
+        'anthropic/claude-sonnet-5', // Best for security analysis
+        'openai/gpt-6-sol', // Strong reasoning
+        'google/gemini-2.5-pro', // Good code understanding
+        'deepseek/deepseek-v4-pro', // Cost-effective alternative
     ];
 }
 /**
@@ -71,17 +98,17 @@ export function getRecommendedSecurityModels() {
  */
 export function getCostOptimizedModels() {
     return [
-        'anthropic/claude-3-haiku', // Fast and cheap Claude
-        'openai/gpt-4o-mini', // Cheap GPT-4
-        'google/gemini-flash-1.5', // Very cheap Gemini
-        'deepseek/deepseek-chat', // Very cheap alternative
+        'anthropic/claude-haiku-4.5', // Fast and cheap Claude
+        'openai/gpt-6-luna', // Cheap GPT-6
+        'google/gemini-3.8-flash', // Cheap Gemini
+        'qwen/qwen3-coder', // Very cheap alternative
     ];
 }
 // ============================================================================
 // OpenRouter Provider Implementation
 // ============================================================================
 const DEFAULT_CONFIG = {
-    defaultModel: 'anthropic/claude-3.5-sonnet',
+    defaultModel: 'anthropic/claude-sonnet-5',
     defaultTimeout: 120000, // 2 minutes
     maxRetries: 3,
     retryDelayMs: 1000,
@@ -97,7 +124,7 @@ const DEFAULT_CONFIG = {
  * ```typescript
  * const provider = new OpenRouterModelProvider({
  *   apiKey: process.env.OPENROUTER_API_KEY,
- *   defaultModel: 'anthropic/claude-3.5-sonnet',
+ *   defaultModel: 'anthropic/claude-sonnet-5',
  * });
  *
  * const result = await provider.complete('Analyze this code for vulnerabilities...');
@@ -127,7 +154,7 @@ export class OpenRouterModelProvider extends BaseModelProvider {
         this.id = `openrouter-${(this.config.defaultModel ?? 'default').replace(/[^a-z0-9]/gi, '-')}`;
         this.name = `OpenRouter (${this.config.defaultModel ?? 'default'})`;
         // Set cost per token based on default model
-        const modelCosts = MODEL_COSTS[this.config.defaultModel];
+        const modelCosts = lookupModelCost(this.config.defaultModel);
         if (modelCosts) {
             this.costPerToken = {
                 input: modelCosts.input / 1_000_000,
@@ -183,7 +210,7 @@ export class OpenRouterModelProvider extends BaseModelProvider {
                     this.requestCount++;
                     // Track cost
                     if (response.usage) {
-                        const costs = MODEL_COSTS[tryModel] || { input: 1, output: 2 };
+                        const costs = lookupModelCost(tryModel) || { input: 1, output: 2 };
                         const cost = (response.usage.prompt_tokens * costs.input +
                             response.usage.completion_tokens * costs.output) /
                             1_000_000;
@@ -330,13 +357,13 @@ Format your response with: verdict (confirmed/rejected/uncertain), confidence (0
  *
  * // With specific model
  * const provider = createOpenRouterProvider({
- *   defaultModel: 'meta-llama/llama-3.1-70b-instruct',
+ *   defaultModel: 'meta-llama/llama-4-maverick',
  * });
  *
  * // With fallback models
  * const provider = createOpenRouterProvider({
- *   defaultModel: 'anthropic/claude-3.5-sonnet',
- *   fallbackModels: ['openai/gpt-4o', 'google/gemini-pro-1.5'],
+ *   defaultModel: 'anthropic/claude-sonnet-5',
+ *   fallbackModels: ['openai/gpt-6-sol', 'google/gemini-2.5-pro'],
  * });
  * ```
  */

package/dist/coordination/handlers/test-execution-handlers.js

@@ -186,18 +186,20 @@ export function registerTestExecutionHandlers(ctx) {
             let output;
             try {
                 execution = spawnSync('npx', ['vitest', 'run', ...testFiles, '--reporter=json', ...report.args], options);
-                output = report.read(execution.stdout || '');
+                output = execution.error || execution.signal
+                    ? ''
+                    : report.read(execution.stdout || '') ?? '';
             }
             finally {
                 report.cleanup();
             }
             // Preserve the existing Jest fallback when Vitest cannot produce a report.
-            if (!output.includes('{') && execution.status !== 0) {
+            if (!output.includes('{') && execution.status !== 0 && !execution.error && !execution.signal) {
                 runner = 'jest';
                 execution = spawnSync('npx', ['jest', ...testFiles, '--json'], options);
                 output = execution.stdout || '';
             }
-            const diagnostics = [execution.error?.message, execution.stderr, output].filter(Boolean).join('\n');
+            const diagnostics = [execution.error?.message, execution.signal && `Terminated by ${execution.signal}`, execution.stderr, output].filter(Boolean).join('\n');
             if (execution.error) {
                 return err(new TestRunnerExecutionError(`${runner} could not complete: ${diagnostics.slice(0, 4000)}`));
             }

package/dist/coordination/interfaces.d.ts

@@ -69,7 +69,7 @@ export interface Protocol {
 /**
  * Status of a protocol action execution
  */
-export type ActionExecutionStatus = 'pending' | 'running' | 'completed' | 'failed' | 'skipped' | 'cancelled';
+export type ActionExecutionStatus = 'pending' | 'running' | 'completed' | 'failed' | 'unknown' | 'skipped' | 'cancelled';
 /**
  * Result of a single action execution
  */
@@ -94,7 +94,7 @@ export interface ActionExecutionResult {
 /**
  * Status of protocol execution
  */
-export type ProtocolExecutionStatus = 'pending' | 'running' | 'completed' | 'failed' | 'cancelled' | 'paused';
+export type ProtocolExecutionStatus = 'pending' | 'running' | 'completed' | 'failed' | 'unknown' | 'cancelled' | 'paused';
 /**
  * Protocol execution state
  */

package/dist/coordination/protocol-executor.js

@@ -6,6 +6,13 @@ import { v4 as uuidv4 } from 'uuid';
 import { ok, err, } from '../shared/types/index.js';
 import { createEvent } from '../shared/events/index.js';
 import { toErrorMessage, toError } from '../shared/error-utils.js';
+/** The caller stopped waiting, but the domain operation may still take effect. */
+class ActionTimeoutError extends Error {
+    constructor(timeoutMs) {
+        super(`Action timed out after ${timeoutMs}ms; outcome is unknown`);
+        this.name = 'ActionTimeoutError';
+    }
+}
 // ============================================================================
 // Protocol Events
 // ============================================================================
@@ -13,10 +20,12 @@ const ProtocolEvents = {
     ProtocolStarted: 'coordination.ProtocolStarted',
     ProtocolCompleted: 'coordination.ProtocolCompleted',
     ProtocolFailed: 'coordination.ProtocolFailed',
+    ProtocolOutcomeUnknown: 'coordination.ProtocolOutcomeUnknown',
     ProtocolCancelled: 'coordination.ProtocolCancelled',
     ActionStarted: 'coordination.ActionStarted',
     ActionCompleted: 'coordination.ActionCompleted',
     ActionFailed: 'coordination.ActionFailed',
+    ActionOutcomeUnknown: 'coordination.ActionOutcomeUnknown',
 };
 // ============================================================================
 // Protocol Executor Implementation
@@ -256,18 +265,22 @@ export class DefaultProtocolExecutor {
             // Execute actions
             await this.executeActions(context);
             // Check for failures
-            const hasFailures = Array.from(context.actionResults.values()).some((r) => r.status === 'failed');
+            const results = Array.from(context.actionResults.values());
+            const unknownActions = results.filter((r) => r.status === 'unknown').map((r) => r.actionId);
+            const failedActions = results.filter((r) => r.status === 'failed').map((r) => r.actionId);
             if (context.cancelled) {
                 execution.status = 'cancelled';
             }
-            else if (hasFailures) {
+            else if (unknownActions.length > 0) {
+                execution.status = 'unknown';
+                await this.publishEvent(ProtocolEvents.ProtocolOutcomeUnknown, { executionId, protocolId: protocol.id, unknownActions, failedActions }, correlationId);
+            }
+            else if (failedActions.length > 0) {
                 execution.status = 'failed';
                 await this.publishEvent(ProtocolEvents.ProtocolFailed, {
                     executionId,
                     protocolId: protocol.id,
-                    failedActions: Array.from(context.actionResults.values())
-                        .filter((r) => r.status === 'failed')
-                        .map((r) => r.actionId),
+                    failedActions,
                 }, correlationId);
             }
             else {
@@ -313,6 +326,23 @@ export class DefaultProtocolExecutor {
             const readyActions = protocol.actions.filter((action) => pending.has(action.id) &&
                 this.dependenciesSatisfied(action, completed));
             if (readyActions.length === 0 && pending.size > 0) {
+                // A failed/unknown prerequisite cannot authorize a dependent action.
+                // Mark the dependent skipped instead of misreporting a deadlock.
+                let skipped = false;
+                for (const action of protocol.actions) {
+                    if (!pending.has(action.id))
+                        continue;
+                    if (action.dependsOn?.some((dep) => {
+                        const status = actionResults.get(dep)?.status;
+                        return status !== undefined && status !== 'completed';
+                    })) {
+                        actionResults.set(action.id, { actionId: action.id, status: 'skipped' });
+                        pending.delete(action.id);
+                        skipped = true;
+                    }
+                }
+                if (skipped)
+                    continue;
                 // Deadlock - some actions can't be executed
                 throw new Error(`Deadlock detected: actions ${Array.from(pending).join(', ')} cannot proceed`);
             }
@@ -402,6 +432,26 @@ export class DefaultProtocolExecutor {
             }
             catch (error) {
                 lastError = toError(error);
+                if (error instanceof ActionTimeoutError) {
+                    // Promise.race cannot cancel the losing provider call. It may have
+                    // committed an effect already, so replay would be unsafe.
+                    const completedAt = new Date();
+                    await this.publishEvent(ProtocolEvents.ActionOutcomeUnknown, {
+                        executionId: context.execution.executionId,
+                        actionId: action.id,
+                        error: error.message,
+                        attempts,
+                    }, context.execution.correlationId);
+                    return {
+                        actionId: action.id,
+                        status: 'unknown',
+                        startedAt,
+                        completedAt,
+                        duration: completedAt.getTime() - startedAt.getTime(),
+                        error: error.message,
+                        retryAttempts: attempts > 1 ? attempts - 1 : undefined,
+                    };
+                }
                 // Retry if configured and not last attempt
                 if (attempts < maxAttempts && action.retry) {
                     const backoff = action.retry.backoffMs *
@@ -441,10 +491,19 @@ export class DefaultProtocolExecutor {
      * Execute a promise with timeout
      */
     async executeWithTimeout(promise, timeoutMs) {
-        return Promise.race([
-            promise,
-            new Promise((_, reject) => setTimeout(() => reject(new Error('Action timeout')), timeoutMs)),
-        ]);
+        let timer;
+        try {
+            return await Promise.race([
+                promise,
+                new Promise((_, reject) => {
+                    timer = setTimeout(() => reject(new ActionTimeoutError(timeoutMs)), timeoutMs);
+                }),
+            ]);
+        }
+        finally {
+            if (timer)
+                clearTimeout(timer);
+        }
     }
     /**
      * Set up event-based triggers for a protocol