agentic-qe 3.14.3 → 3.14.4

package/dist/shared/llm/router/agent-router-config.js

@@ -8,6 +8,7 @@
  * - Override mechanism for specific tasks
  * - Agent category detection and preferred model resolution
  */
+import { DEFAULT_OPUS_MODEL, DEFAULT_SONNET_MODEL } from '../model-registry.js';
 // ============================================================================
 // Default Model Preferences by Category
 // ============================================================================
@@ -18,7 +19,7 @@ export const DEFAULT_CATEGORY_MODELS = {
     // Security agents need best reasoning capabilities
     security: {
         provider: 'claude',
-        model: 'claude-opus-4-7',
+        model: DEFAULT_OPUS_MODEL,
         temperature: 0.1,
         maxTokens: 16000,
         priority: 100,
@@ -26,7 +27,7 @@ export const DEFAULT_CATEGORY_MODELS = {
     // Test generation needs good balance of quality and speed
     'test-generation': {
         provider: 'claude',
-        model: 'claude-sonnet-4-6',
+        model: DEFAULT_SONNET_MODEL,
         temperature: 0.3,
         maxTokens: 8000,
         priority: 90,
@@ -34,7 +35,7 @@ export const DEFAULT_CATEGORY_MODELS = {
     // Code analysis requires good understanding
     'code-analysis': {
         provider: 'claude',
-        model: 'claude-sonnet-4-6',
+        model: DEFAULT_SONNET_MODEL,
         temperature: 0.2,
         maxTokens: 8000,
         priority: 85,
@@ -58,7 +59,7 @@ export const DEFAULT_CATEGORY_MODELS = {
     // Learning/ML agents - balanced
     learning: {
         provider: 'claude',
-        model: 'claude-sonnet-4-6',
+        model: DEFAULT_SONNET_MODEL,
         temperature: 0.3,
         maxTokens: 8000,
         priority: 80,
@@ -74,7 +75,7 @@ export const DEFAULT_CATEGORY_MODELS = {
     // Simple tasks - cheapest models
     simple: {
         provider: 'openai',
-        model: 'gpt-4o-mini',
+        model: 'gpt-6-luna',
         temperature: 0.3,
         maxTokens: 2000,
         priority: 50,
@@ -82,7 +83,7 @@ export const DEFAULT_CATEGORY_MODELS = {
     // General purpose - balanced default
     general: {
         provider: 'claude',
-        model: 'claude-sonnet-4-6',
+        model: DEFAULT_SONNET_MODEL,
         temperature: 0.3,
         maxTokens: 4000,
         priority: 60,
@@ -348,7 +349,7 @@ let agentProviderOverrides = {};
  * Needed because the category defaults in `DEFAULT_CATEGORY_MODELS` pair a
  * provider with a provider-specific model id. If a user writes
  * `{ "provider": "ollama" }` and we inherited the category *model* too, the
- * agent would be routed to ollama asking for `claude-sonnet-4-6` — a model that
+ * agent would be routed to ollama asking for `claude-sonnet-5` — a model that
  * provider cannot serve. Changing provider without naming a model must fall
  * back to that provider's own default, not the previous provider's model.
  */
@@ -575,7 +576,7 @@ export const DEFAULT_ROUTING_OVERRIDES = [
         },
         modelPreference: {
             provider: 'claude',
-            model: 'claude-opus-4-7',
+            model: DEFAULT_OPUS_MODEL,
             temperature: 0.2,
             maxTokens: 16000,
             priority: 150,
@@ -590,7 +591,7 @@ export const DEFAULT_ROUTING_OVERRIDES = [
         },
         modelPreference: {
             provider: 'openai',
-            model: 'gpt-4o-mini',
+            model: 'gpt-6-luna',
             temperature: 0.3,
             maxTokens: 2000,
             priority: 40,
@@ -605,7 +606,7 @@ export const DEFAULT_ROUTING_OVERRIDES = [
         },
         modelPreference: {
             provider: 'ollama',
-            model: 'llama3.1:70b',
+            model: 'qwen3-coder:30b', // ADR-111 local QE default
             temperature: 0.2,
             maxTokens: 4000,
             priority: 200,
@@ -620,7 +621,7 @@ export const DEFAULT_ROUTING_OVERRIDES = [
         },
         modelPreference: {
             provider: 'claude',
-            model: 'claude-sonnet-4-6',
+            model: DEFAULT_SONNET_MODEL,
             temperature: 0.3,
             maxTokens: 8000,
             priority: 95,
@@ -636,49 +637,49 @@ export const DEFAULT_ROUTING_OVERRIDES = [
  */
 export const ALTERNATIVE_MODELS = {
     security: [
-        { provider: 'openai', model: 'gpt-4o', temperature: 0.1, priority: 90 },
-        { provider: 'claude', model: 'claude-sonnet-4-6', temperature: 0.1, priority: 80 },
+        { provider: 'openai', model: 'gpt-6-sol', temperature: 0.1, priority: 90 },
+        { provider: 'claude', model: DEFAULT_SONNET_MODEL, temperature: 0.1, priority: 80 },
         { provider: 'openrouter', model: 'anthropic/claude-opus-4.5', temperature: 0.1, priority: 70 },
     ],
     'test-generation': [
-        { provider: 'openai', model: 'gpt-4o', temperature: 0.3, priority: 80 },
+        { provider: 'openai', model: 'gpt-6-sol', temperature: 0.3, priority: 80 },
         { provider: 'claude', model: 'claude-haiku-4-5-20251001', temperature: 0.3, priority: 70 },
-        { provider: 'gemini', model: 'gemini-2.0-pro', temperature: 0.3, priority: 60 },
+        { provider: 'gemini', model: 'gemini-2.5-pro', temperature: 0.3, priority: 60 },
     ],
     'code-analysis': [
-        { provider: 'openai', model: 'gpt-4o', temperature: 0.2, priority: 75 },
+        { provider: 'openai', model: 'gpt-6-sol', temperature: 0.2, priority: 75 },
         { provider: 'claude', model: 'claude-haiku-4-5-20251001', temperature: 0.2, priority: 65 },
-        { provider: 'gemini', model: 'gemini-2.0-pro', temperature: 0.2, priority: 55 },
+        { provider: 'gemini', model: 'gemini-2.5-pro', temperature: 0.2, priority: 55 },
     ],
     performance: [
-        { provider: 'openai', model: 'gpt-4o-mini', temperature: 0.2, priority: 60 },
-        { provider: 'gemini', model: 'gemini-2.0-flash', temperature: 0.2, priority: 50 },
-        { provider: 'ollama', model: 'llama3.1', temperature: 0.2, priority: 40 },
+        { provider: 'openai', model: 'gpt-6-luna', temperature: 0.2, priority: 60 },
+        { provider: 'gemini', model: 'gemini-2.5-flash', temperature: 0.2, priority: 50 },
+        { provider: 'ollama', model: 'qwen3-coder:30b', temperature: 0.2, priority: 40 },
     ],
     documentation: [
-        { provider: 'openai', model: 'gpt-4o-mini', temperature: 0.4, priority: 50 },
-        { provider: 'gemini', model: 'gemini-2.0-flash', temperature: 0.4, priority: 40 },
-        { provider: 'ollama', model: 'llama3.1', temperature: 0.4, priority: 30 },
+        { provider: 'openai', model: 'gpt-6-luna', temperature: 0.4, priority: 50 },
+        { provider: 'gemini', model: 'gemini-2.5-flash', temperature: 0.4, priority: 40 },
+        { provider: 'ollama', model: 'qwen3-coder:30b', temperature: 0.4, priority: 30 },
     ],
     learning: [
-        { provider: 'openai', model: 'gpt-4o', temperature: 0.3, priority: 70 },
+        { provider: 'openai', model: 'gpt-6-sol', temperature: 0.3, priority: 70 },
         { provider: 'claude', model: 'claude-haiku-4-5-20251001', temperature: 0.3, priority: 60 },
-        { provider: 'gemini', model: 'gemini-2.0-pro', temperature: 0.3, priority: 50 },
+        { provider: 'gemini', model: 'gemini-2.5-pro', temperature: 0.3, priority: 50 },
     ],
     coordination: [
-        { provider: 'openai', model: 'gpt-4o-mini', temperature: 0.2, priority: 65 },
-        { provider: 'gemini', model: 'gemini-2.0-flash', temperature: 0.2, priority: 55 },
-        { provider: 'ollama', model: 'llama3.1', temperature: 0.2, priority: 45 },
+        { provider: 'openai', model: 'gpt-6-luna', temperature: 0.2, priority: 65 },
+        { provider: 'gemini', model: 'gemini-2.5-flash', temperature: 0.2, priority: 55 },
+        { provider: 'ollama', model: 'qwen3-coder:30b', temperature: 0.2, priority: 45 },
     ],
     simple: [
-        { provider: 'gemini', model: 'gemini-2.0-flash', temperature: 0.3, priority: 40 },
+        { provider: 'gemini', model: 'gemini-2.5-flash', temperature: 0.3, priority: 40 },
         { provider: 'claude', model: 'claude-haiku-4-5-20251001', temperature: 0.3, priority: 30 },
-        { provider: 'ollama', model: 'phi4', temperature: 0.3, priority: 20 },
+        { provider: 'ollama', model: 'qwen3-coder:30b', temperature: 0.3, priority: 20 },
     ],
     general: [
-        { provider: 'openai', model: 'gpt-4o', temperature: 0.3, priority: 50 },
+        { provider: 'openai', model: 'gpt-6-sol', temperature: 0.3, priority: 50 },
         { provider: 'claude', model: 'claude-haiku-4-5-20251001', temperature: 0.3, priority: 40 },
-        { provider: 'gemini', model: 'gemini-2.0-pro', temperature: 0.3, priority: 30 },
+        { provider: 'gemini', model: 'gemini-2.5-pro', temperature: 0.3, priority: 30 },
     ],
 };
 /**

package/dist/shared/llm/router/hybrid-router.js

@@ -13,11 +13,11 @@ import { v4 as uuidv4 } from 'uuid';
 import { createLLMError, isLLMError, } from '../interfaces.js';
 import { DEFAULT_ROUTER_CONFIG, } from './types.js';
 import { RoutingRuleEngine, DEFAULT_QE_ROUTING_RULES } from './routing-rules.js';
-import { applyCyberPin, CYBER_PIN_CHAT_FALLBACK, shouldCyberPin, isOpus47, } from '../../../routing/security/cyber-pin.js';
+import { applyCyberPin, CYBER_PIN_CHAT_FALLBACK, cyberPinFallbackFor, shouldCyberPin, isCyberGatedModel, } from '../../../routing/security/cyber-pin.js';
 import { createRouterMetricsCollector, createCostMetricsCollector, } from '../metrics/index.js';
 import { toError } from '../../error-utils.js';
 import { mapModelId, normalizeModelId, getModelMapping, } from '../model-mapping.js';
-import { modelSupportsTools } from '../model-registry.js';
+import { modelSupportsTools, DEFAULT_SONNET_MODEL } from '../model-registry.js';
 /**
  * Simple LRU-like decision cache
  */
@@ -195,20 +195,21 @@ export class HybridRouter {
         const requestId = uuidv4();
         const decision = await this.selectProvider(params);
         // ADR-093: apply Cyber Verification pin before dispatch. Security agents
-        // cannot reach Opus 4.7 until AQE_CYBER_VERIFIED=true. Covers both the
-        // canonical model field and the provider-specific id — if either targets
-        // 4.7 for a cyber-pinned agent, downgrade to Sonnet 4.6 on the same
-        // provider. Applies to direct chat() calls; MultiModelExecutor.consult()
+        // cannot reach an Opus/Fable model >= 4.7 until AQE_CYBER_VERIFIED=true.
+        // Covers both the canonical model field and the provider-specific id — if
+        // either targets a cyber-gated model for a cyber-pinned agent, downgrade
+        // to the pinned Sonnet 4.6 on the same provider (fallback candidates are
+        // pinned again in executeWithFallback). MultiModelExecutor.consult()
         // applies the same pin independently for advisor escalations.
         const agentName = params.agentType ?? '';
-        if (shouldCyberPin(agentName) && (isOpus47(decision.model) || isOpus47(decision.providerModelId))) {
+        if (shouldCyberPin(agentName) && (isCyberGatedModel(decision.model) || isCyberGatedModel(decision.providerModelId))) {
             if (params.strictModel) {
                 throw createLLMError(`Strict model request for ${decision.model} conflicts with the cyber verification pin for ${agentName}`, 'MODEL_NOT_FOUND', { provider: decision.providerType, model: decision.model, retryable: false });
             }
             const originalModel = decision.model;
             const originalProviderModelId = decision.providerModelId;
             decision.model = applyCyberPin(agentName, decision.model, CYBER_PIN_CHAT_FALLBACK);
-            decision.providerModelId = applyCyberPin(agentName, decision.providerModelId, CYBER_PIN_CHAT_FALLBACK);
+            decision.providerModelId = applyCyberPin(agentName, decision.providerModelId, cyberPinFallbackFor(decision.providerType));
             // eslint-disable-next-line no-console
             console.warn(`[aqe] ADR-093: ${agentName} pinned to ${decision.model} (was ${originalModel}/${originalProviderModelId}); ` +
                 `set AQE_CYBER_VERIFIED=true after Cyber Verification Program approval`);
@@ -498,6 +499,17 @@ export class HybridRouter {
                 }
             }
         }
+        // ADR-093: fallback candidates (default model, fallback chain) are added
+        // after chat() applied the cyber pin, so pin them here too — otherwise a
+        // failed primary could escalate a security agent to an Opus/Fable model.
+        const pinAgent = params.agentType ?? '';
+        if (shouldCyberPin(pinAgent)) {
+            for (const entry of executionOrder) {
+                if (isCyberGatedModel(entry.model)) {
+                    entry.model = cyberPinFallbackFor(entry.provider);
+                }
+            }
+        }
         // Capability-aware guard: when the request needs tools, drop any candidate
         // whose model is KNOWN to lack tool support (e.g. local llama3, codellama,
         // mistral) so we don't silently strip tool definitions and return a
@@ -951,7 +963,7 @@ export function createQERouter(providerManager) {
         mode: 'rule-based',
         rules: DEFAULT_QE_ROUTING_RULES,
         defaultProvider: 'claude',
-        defaultModel: 'claude-sonnet-4-6',
+        defaultModel: DEFAULT_SONNET_MODEL,
         enableMetrics: true,
         cacheDecisions: true,
     });

package/dist/shared/llm/router/routing-rules.js

@@ -15,6 +15,7 @@
  * - Override mechanism for specific tasks
  */
 import { getAgentRoutingCategory, getPreferredModelForAgent, getAgentCapabilityRequirements, createAllAgentRoutingRules, AGENT_CATEGORY_MAP, DEFAULT_CATEGORY_MODELS, } from './agent-router-config.js';
+import { DEFAULT_OPUS_MODEL, DEFAULT_SONNET_MODEL } from '../model-registry.js';
 // ============================================================================
 // Rule Engine
 // ============================================================================
@@ -198,7 +199,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'claude',
-            model: 'claude-opus-4-7',
+            model: DEFAULT_OPUS_MODEL,
             temperature: 0.1,
         },
         enabled: true,
@@ -215,7 +216,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'claude',
-            model: 'claude-sonnet-4-6',
+            model: DEFAULT_SONNET_MODEL,
             temperature: 0.3,
         },
         enabled: true,
@@ -232,7 +233,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'claude',
-            model: 'claude-opus-4-7',
+            model: DEFAULT_OPUS_MODEL,
             temperature: 0.2,
         },
         enabled: true,
@@ -248,7 +249,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'claude',
-            model: 'claude-sonnet-4-6',
+            model: DEFAULT_SONNET_MODEL,
             temperature: 0.2,
         },
         enabled: true,
@@ -264,7 +265,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'claude',
-            model: 'claude-sonnet-4-6',
+            model: DEFAULT_SONNET_MODEL,
             temperature: 0.1,
         },
         enabled: true,
@@ -280,7 +281,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'claude',
-            model: 'claude-sonnet-4-6',
+            model: DEFAULT_SONNET_MODEL,
             temperature: 0.3,
         },
         enabled: true,
@@ -296,7 +297,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'claude',
-            model: 'claude-sonnet-4-6',
+            model: DEFAULT_SONNET_MODEL,
             temperature: 0.3,
         },
         enabled: true,
@@ -328,7 +329,7 @@ export const DEFAULT_QE_ROUTING_RULES = [
         },
         action: {
             provider: 'ollama',
-            model: 'llama3.1',
+            model: 'qwen3-coder:30b',
             temperature: 0.3,
         },
         enabled: true,

package/dist/shared/llm/router/types.js

@@ -19,6 +19,7 @@
  * @see ADR-043 - Vendor-Independent LLM Support (this extension)
  */
 import { isBuiltinProviderType, } from '../interfaces.js';
+import { CLAUDE_TIER_MODELS, DEFAULT_SONNET_MODEL } from '../model-registry.js';
 /**
  * Type guard for the closed extended set. Use before indexing any
  * `Record<BuiltinExtendedProviderType, _>` with an open provider type, so the
@@ -84,7 +85,7 @@ export const DEFAULT_FALLBACK_BEHAVIOR = {
 export const DEFAULT_FALLBACK_CHAIN_ENTRIES = [
     {
         provider: 'claude',
-        models: ['claude-sonnet-4-6', 'claude-haiku-4-5-20251001'],
+        models: [DEFAULT_SONNET_MODEL, CLAUDE_TIER_MODELS[1]],
         enabled: true,
         priority: 100,
         maxAttempts: 2,
@@ -92,7 +93,7 @@ export const DEFAULT_FALLBACK_CHAIN_ENTRIES = [
     },
     {
         provider: 'openai',
-        models: ['gpt-4o', 'gpt-4o-mini'],
+        models: ['gpt-6-sol', 'gpt-6-luna'],
         enabled: true,
         priority: 90,
         maxAttempts: 2,
@@ -100,7 +101,7 @@ export const DEFAULT_FALLBACK_CHAIN_ENTRIES = [
     },
     {
         provider: 'openrouter',
-        models: ['anthropic/claude-sonnet-4', 'openai/gpt-4o'],
+        models: ['anthropic/claude-sonnet-5', 'openai/gpt-6-sol'],
         enabled: true,
         priority: 85,
         maxAttempts: 2,
@@ -108,7 +109,7 @@ export const DEFAULT_FALLBACK_CHAIN_ENTRIES = [
     },
     {
         provider: 'ollama',
-        models: ['llama3.1', 'mistral'],
+        models: ['qwen3-coder:30b', 'qwen3:30b-a3b'],
         enabled: true,
         priority: 80,
         maxAttempts: 2,
@@ -141,18 +142,18 @@ export const DEFAULT_ROUTER_CONFIG = {
     mode: 'rule-based',
     rules: [],
     defaultProvider: 'claude',
-    defaultModel: 'claude-sonnet-4-6',
+    defaultModel: DEFAULT_SONNET_MODEL,
     fallbackChain: DEFAULT_FALLBACK_CHAIN,
     fallbackBehavior: DEFAULT_FALLBACK_BEHAVIOR,
     providers: {
-        claude: { enabled: true, defaultModel: 'claude-sonnet-4-6' },
+        claude: { enabled: true, defaultModel: DEFAULT_SONNET_MODEL },
         // ADR-123: opt-in (issue #557 — don't default the subscription/fleet path
         // on). Enabled explicitly via AQE_LLM_PROVIDER=claude-code or disk config.
-        'claude-code': { enabled: false, defaultModel: 'claude-sonnet-4-6' },
-        openai: { enabled: true, defaultModel: 'gpt-4o' },
-        ollama: { enabled: true, defaultModel: 'llama3.1' },
-        openrouter: { enabled: true, defaultModel: 'anthropic/claude-sonnet-4' },
-        gemini: { enabled: false, defaultModel: 'gemini-pro' },
+        'claude-code': { enabled: false, defaultModel: DEFAULT_SONNET_MODEL },
+        openai: { enabled: true, defaultModel: 'gpt-6-sol' },
+        ollama: { enabled: true, defaultModel: 'qwen3-coder:30b' },
+        openrouter: { enabled: true, defaultModel: 'anthropic/claude-sonnet-5' },
+        gemini: { enabled: false, defaultModel: 'gemini-2.5-flash' },
         'azure-openai': { enabled: false },
         bedrock: { enabled: false },
         // ADR-123: opt-in; enabled when COGNITUM_API_KEY present or via override.
@@ -190,6 +191,73 @@ export const DEFAULT_ROUTER_CONFIG = {
  * This enables vendor-neutral model references
  */
 export const DEFAULT_MODEL_MAPPINGS = [
+    // Claude 5 generation (2026-09 model refresh; routing tiers 2-4)
+    {
+        canonicalId: 'claude-opus-5',
+        canonicalName: 'Claude Opus 5',
+        providerIds: {
+            claude: 'claude-opus-5',
+            openrouter: 'anthropic/claude-opus-5',
+        },
+        capabilities: {
+            supportsStreaming: true,
+            supportsTools: true,
+            supportsMCP: true,
+            supportsVision: true,
+            supportsExtendedThinking: true,
+            supportsJsonMode: true,
+            maxContextTokens: 1_000_000,
+            maxOutputTokens: 128000,
+        },
+        tier: 'flagship',
+        family: 'claude',
+        inputCostPer1M: 5,
+        outputCostPer1M: 25,
+    },
+    {
+        canonicalId: 'claude-sonnet-5',
+        canonicalName: 'Claude Sonnet 5',
+        providerIds: {
+            claude: 'claude-sonnet-5',
+            openrouter: 'anthropic/claude-sonnet-5',
+        },
+        capabilities: {
+            supportsStreaming: true,
+            supportsTools: true,
+            supportsMCP: true,
+            supportsVision: true,
+            supportsExtendedThinking: true,
+            supportsJsonMode: true,
+            maxContextTokens: 1_000_000,
+            maxOutputTokens: 128000,
+        },
+        tier: 'advanced',
+        family: 'claude',
+        inputCostPer1M: 2,
+        outputCostPer1M: 10,
+    },
+    {
+        canonicalId: 'claude-opus-5-5',
+        canonicalName: 'Claude Opus 5.5',
+        providerIds: {
+            claude: 'claude-opus-5-5',
+            openrouter: 'anthropic/claude-opus-5.5',
+        },
+        capabilities: {
+            supportsStreaming: true,
+            supportsTools: true,
+            supportsMCP: true,
+            supportsVision: true,
+            supportsExtendedThinking: true,
+            supportsJsonMode: true,
+            maxContextTokens: 1_000_000,
+            maxOutputTokens: 128000,
+        },
+        tier: 'flagship',
+        family: 'claude',
+        inputCostPer1M: 4,
+        outputCostPer1M: 20,
+    },
     // ADR-093: Claude Opus 4.7 (flagship, Apr 2026)
     {
         canonicalId: 'claude-opus-4-7',
@@ -230,8 +298,8 @@ export const DEFAULT_MODEL_MAPPINGS = [
             supportsVision: true,
             supportsExtendedThinking: true,
             supportsJsonMode: true,
-            maxContextTokens: 200000,
-            maxOutputTokens: 64000,
+            maxContextTokens: 1_000_000,
+            maxOutputTokens: 128000,
         },
         tier: 'advanced',
         family: 'claude',

package/dist/shared/llm/translation/message-formatter.js

@@ -8,6 +8,7 @@
  * - Message role mapping
  */
 import { isBuiltinExtendedProviderType } from '../router/types.js';
+import { resolveModelPricing } from '../cost-tracker.js';
 // ============================================================================
 // System Prompt Handling
 // ============================================================================
@@ -220,31 +221,11 @@ export function formatResponse(response, sourceProvider, latencyMs, cached = fal
 }
 /**
  * Calculate cost based on model pricing
- * Note: This is a simplified version - real implementation should use model-mapping pricing
  */
 function calculateCost(inputTokens, outputTokens, model) {
-    // Simplified pricing per 1M tokens (should be loaded from model-mapping)
-    const pricing = {
-        'claude-opus-4-7': { input: 15, output: 75 },
-        'claude-sonnet-4-6': { input: 3, output: 15 },
-        'claude-haiku-4-5-20251001': { input: 0.8, output: 4 },
-        'gpt-4o': { input: 2.5, output: 10 },
-        'gpt-4o-mini': { input: 0.15, output: 0.6 },
-        'gemini-pro': { input: 1.25, output: 5 },
-        'gemini-2.0-pro': { input: 1.25, output: 5 },
-    };
-    // Find matching pricing or use default
-    let modelPricing = pricing[model];
-    if (!modelPricing) {
-        // Try prefix matching
-        for (const [key, value] of Object.entries(pricing)) {
-            if (model.includes(key) || key.includes(model.split('-')[0])) {
-                modelPricing = value;
-                break;
-            }
-        }
-    }
-    modelPricing = modelPricing || { input: 1, output: 2 }; // Default fallback
+    // Shared pricing (cost tracker table, then model registry); unknown models
+    // keep the historical conservative fallback rather than being treated as free.
+    const modelPricing = resolveModelPricing(model) ?? { input: 1, output: 2 };
     const inputCost = (inputTokens / 1_000_000) * modelPricing.input;
     const outputCost = (outputTokens / 1_000_000) * modelPricing.output;
     return {

package/dist/shared/vitest-json-report.d.ts

@@ -17,19 +17,19 @@ export interface VitestJsonReport {
     /** CLI arguments to append to `vitest run --reporter=json`. */
     readonly args: readonly string[];
     /**
-     * Return the text that carries the JSON document: the report file when the
-     * runner wrote one, otherwise the captured stdout (older runners, or a run
-     * that died before reporting).
+     * Read this invocation's report. Missing reports return undefined so callers
+     * can distinguish them from a completed run; malformed reports throw.
+     * Stdout is diagnostic text, never a substitute for the requested report.
      */
-    read(stdout: string): string;
+    read(stdout: string): string | undefined;
     /** Remove the temporary report directory. Safe to call more than once. */
     cleanup(): void;
 }
 /**
- * Resolve the JSON document text for a finished Vitest run.
- * Exported separately so the precedence rule is unit-testable without spawning.
+ * Read the report owned by a finished Vitest run. The stdout parameter stays
+ * for existing callers but is never treated as result evidence.
  */
-export declare function resolveVitestJsonOutput(stdout: string, reportPath: string | undefined): string;
+export declare function resolveVitestJsonOutput(_stdout: string, reportPath: string | undefined): string | undefined;
 /**
  * Create a per-run report location. The directory is unique per call so
  * concurrent runs never read each other's output.

package/dist/shared/vitest-json-report.js

@@ -11,25 +11,34 @@
  * that spawns `vitest run --reporter=json` must go through this module
  * instead of parsing stdout directly.
  */
-import { existsSync, mkdtempSync, readFileSync, rmSync } from 'node:fs';
+import { mkdtempSync, readFileSync, rmSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 /**
- * Resolve the JSON document text for a finished Vitest run.
- * Exported separately so the precedence rule is unit-testable without spawning.
+ * Read the report owned by a finished Vitest run. The stdout parameter stays
+ * for existing callers but is never treated as result evidence.
  */
-export function resolveVitestJsonOutput(stdout, reportPath) {
-    if (reportPath && existsSync(reportPath)) {
-        try {
-            const content = readFileSync(reportPath, 'utf-8');
-            if (content.trim().length > 0)
-                return content;
-        }
-        catch {
-            // Unreadable report: fall back to stdout below.
-        }
+export function resolveVitestJsonOutput(_stdout, reportPath) {
+    if (!reportPath)
+        return undefined;
+    let content;
+    try {
+        content = readFileSync(reportPath, 'utf-8');
     }
-    return stdout;
+    catch (error) {
+        if (error.code === 'ENOENT')
+            return undefined;
+        throw new Error('Could not read the current Vitest JSON report.');
+    }
+    try {
+        const parsed = JSON.parse(content);
+        if (!parsed || !Array.isArray(parsed.testResults))
+            throw new Error('Invalid report');
+    }
+    catch {
+        throw new Error('The current Vitest JSON report is malformed.');
+    }
+    return content;
 }
 /**
  * Create a per-run report location. The directory is unique per call so

package/dist/test-scheduling/executors/vitest-executor.d.ts

@@ -25,11 +25,12 @@ export interface VitestConfig {
      * is detected historically by tracking pass/fail patterns over time.
      */
     flakyTracker?: FlakyTestTracker;
+    /** Grace period before forcefully stopping a test process tree. */
+    terminationGraceMs?: number;
 }
 export declare class VitestPhaseExecutor implements PhaseExecutor {
     private readonly config;
-    private currentProcess;
-    private isAborted;
+    private currentCommand;
     constructor(config?: VitestConfig);
     execute(phase: TestPhase, testFiles?: string[]): Promise<PhaseResult>;
     isReady(): Promise<boolean>;
@@ -38,6 +39,10 @@ export declare class VitestPhaseExecutor implements PhaseExecutor {
     private buildArgs;
     private runVitest;
     private runCommand;
+    private terminateProcessTree;
+    private signalGroup;
+    private waitForGroupExit;
+    private waitForClose;
     private parseResult;
     private getCoverageFromReport;
     private createErrorResult;