feat: add Kimi OpenAI-compatible provider

Story 184.1 implementation: generic OpenAI-compatible provider with Kimi/Moonshot preset, factory aliases, dispatcher fallback support, docs/config updates, tests, and CodeRabbit feedback addressed in 2e4447b.

Author: rafaelscosta

.aiox-core/core/execution/subagent-dispatcher.js

@@ -321,6 +321,37 @@ class SubagentDispatcher extends EventEmitter {
     }
   }
 
+  /**
+   * Resolve fallback provider from configuration before using the legacy pair.
+   * @param {string} providerName - Current provider name
+   * @returns {string|null} Fallback provider name
+   */
+  getFallbackProviderName(providerName) {
+    if (AIProviderFactory?.getConfig) {
+      const config = AIProviderFactory.getConfig();
+      const configuredFallback = config.ai_providers?.fallback;
+      const configuredPrimary = config.ai_providers?.primary;
+
+      if (configuredFallback && configuredFallback !== providerName) {
+        return configuredFallback;
+      }
+
+      if (configuredPrimary && configuredPrimary !== providerName) {
+        return configuredPrimary;
+      }
+    }
+
+    if (providerName === 'claude') {
+      return 'gemini';
+    }
+
+    if (providerName === 'gemini') {
+      return 'claude';
+    }
+
+    return null;
+  }
+
   /**
    * Enrich context with memory and gotchas
    * @param {Object} task - Task being dispatched
@@ -432,8 +463,6 @@ class SubagentDispatcher extends EventEmitter {
    * @returns {Promise<Object>} - Execution result
    */
   async executeWithProvider(prompt, providerName, task) {
-    const startTime = Date.now();
-
     // Get primary provider
     const provider = this.getAIProvider(providerName);
 
@@ -450,8 +479,8 @@ class SubagentDispatcher extends EventEmitter {
       this.log('provider_not_available', { provider: providerName });
 
       // Try fallback provider
-      const fallbackName = providerName === 'claude' ? 'gemini' : 'claude';
-      const fallback = this.getAIProvider(fallbackName);
+      const fallbackName = this.getFallbackProviderName(providerName);
+      const fallback = fallbackName ? this.getAIProvider(fallbackName) : null;
 
       if (fallback && (await fallback.checkAvailability())) {
         this.log('using_fallback_provider', { original: providerName, fallback: fallbackName });
@@ -563,10 +592,10 @@ class SubagentDispatcher extends EventEmitter {
    * Select result from parallel execution based on mode
    * @param {Object|null} claudeResult - Claude result
    * @param {Object|null} geminiResult - Gemini result
-   * @param {Object} task - Original task
+   * @param {Object} _task - Original task
    * @returns {Object} - Selected result
    */
-  selectParallelResult(claudeResult, geminiResult, task) {
+  selectParallelResult(claudeResult, geminiResult, _task) {
     switch (this.parallelMode) {
       case 'race':
         // Return first successful result
@@ -698,7 +727,11 @@ class SubagentDispatcher extends EventEmitter {
 
       child.on('error', rejectOnce);
 
-      if (!child.stdin || typeof child.stdin.write !== 'function' || typeof child.stdin.end !== 'function') {
+      if (
+        !child.stdin ||
+        typeof child.stdin.write !== 'function' ||
+        typeof child.stdin.end !== 'function'
+      ) {
         rejectOnce(new Error('Claude CLI stdin is not available'));
         return;
       }

.aiox-core/infrastructure/integrations/ai-providers/README.md

@@ -1,6 +1,6 @@
 # AI Providers
 
-Multi-provider AI integration for AIOX. Supports Claude Code and Gemini CLI with automatic fallback and task-based routing.
+Multi-provider AI integration for AIOX. Supports Claude Code, Gemini CLI and OpenAI-compatible HTTP APIs with automatic fallback and task-based routing.
 
 ## Architecture
 
@@ -9,6 +9,7 @@ ai-providers/
 ├── ai-provider.js           # Base abstract class
 ├── claude-provider.js       # Claude Code implementation
 ├── gemini-provider.js       # Gemini CLI implementation
+├── openai-compatible-provider.js # OpenAI-compatible HTTP implementation
 ├── ai-provider-factory.js   # Factory with routing and fallback
 └── index.js                 # Module exports
 ```
@@ -35,13 +36,13 @@ const { getProviderForTask } = require('./ai-providers');
 
 // Get optimal provider for task type
 const provider = getProviderForTask('code_generation'); // Returns Gemini (faster)
-const provider = getProviderForTask('security');        // Returns Claude (deeper reasoning)
+const provider = getProviderForTask('security'); // Returns Claude (deeper reasoning)
 ```
 
 ### Direct Provider Access
 
 ```javascript
-const { ClaudeProvider, GeminiProvider } = require('./ai-providers');
+const { ClaudeProvider, GeminiProvider, OpenAICompatibleProvider } = require('./ai-providers');
 
 // Claude
 const claude = new ClaudeProvider({ model: 'claude-3-5-sonnet' });
@@ -50,6 +51,15 @@ const response = await claude.execute('Explain this function');
 // Gemini with JSON output
 const gemini = new GeminiProvider({ jsonOutput: true });
 const response = await gemini.executeJson('List 5 best practices');
+
+// Kimi/Moonshot through the OpenAI-compatible contract
+const kimi = new OpenAICompatibleProvider({
+  name: 'kimi',
+  baseURL: 'https://api.moonshot.ai/v1',
+  apiKeyEnv: 'MOONSHOT_API_KEY',
+  model: 'kimi-k2.5',
+});
+const response = await kimi.execute('Explain this function');
 ```
 
 ### Check Provider Status
@@ -84,16 +94,23 @@ claude:
 gemini:
   model: gemini-2.0-flash
   previewFeatures: true
+
+kimi:
+  provider: openai-compatible
+  baseURL: https://api.moonshot.ai/v1
+  endpoint: /chat/completions
+  apiKeyEnv: MOONSHOT_API_KEY
+  model: kimi-k2.5
 ```
 
 ## Provider Comparison
 
-| Feature | Claude | Gemini |
-|---------|--------|--------|
-| Best for | Complex reasoning, security | Speed, cost efficiency |
-| JSON output | Manual parsing | Native `--output-format json` |
-| Cost | Higher | ~4x cheaper (Flash) |
-| SWE-bench | ~70% | 78% (Flash) |
+| Feature     | Claude                      | Gemini                        |
+| ----------- | --------------------------- | ----------------------------- |
+| Best for    | Complex reasoning, security | Speed, cost efficiency        |
+| JSON output | Manual parsing              | Native `--output-format json` |
+| Cost        | Higher                      | ~4x cheaper (Flash)           |
+| SWE-bench   | ~70%                        | 78% (Flash)                   |
 
 ## Epic Reference