ado-sync 0.1.55 → 0.1.57

package/dist/ai/summarizer.js

@@ -14,7 +14,7 @@
  *                dependencies, works fully offline.
  *
  *   ollama     — local LLM via Ollama REST API (http://localhost:11434).
- *                Model suggestion: qwen2.5-coder:7b.
+ *                Model suggestion: gemma-4-e4b-it.
  *
  *   openai     — OpenAI Chat Completions API (requires API key).
  *
@@ -23,7 +23,7 @@
  * Usage (engine.ts / CLI):
  *   const { title, description, steps } = await summarizeTest(test, localType, {
  *     provider: 'local',
- *     model: '/path/to/qwen2.5-coder-1.5b-instruct-q4_k_m.gguf',
+ *     model: '/path/to/gemma-4-e4b-it-Q4_K_M.gguf',
  *   });
  *   test.title = title;
  *   test.description = description;
@@ -663,6 +663,9 @@ function parseAiResponse(raw, fallbackTitle) {
     }
     return { title, description, steps };
 }
+// ─── Shared dynamic import helper ────────────────────────────────────────────
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+const esmImport = new Function('m', 'return import(m)');
 const llamaSessionCache = new Map();
 async function getLlamaSession(modelPath) {
     const cached = llamaSessionCache.get(modelPath);
@@ -704,82 +707,65 @@ async function localLlamaSummary(code, fallbackTitle, modelPath, contextContent)
     }
 }
 async function ollamaSummary(code, fallbackTitle, model, baseUrl, contextContent) {
-    const res = await fetchWithRetry(`${baseUrl}/api/generate`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({ model, prompt: buildPrompt(code, contextContent), stream: false }),
-        signal: AbortSignal.timeout(60_000),
-    }, 'ollama');
-    if (!res.ok)
-        throw new Error(`Ollama ${res.status}: ${await res.text()}`);
-    const data = await res.json();
-    return parseAiResponse(data.response ?? '', fallbackTitle);
-}
-/**
- * Fetch with automatic retry for transient errors:
- *   503 — model container cold-starting (Hugging Face serverless inference)
- *   429 — rate limit exceeded
- * Retries up to `maxRetries` times with exponential backoff starting at `baseDelayMs`.
- */
-async function fetchWithRetry(url, init, provider, maxRetries = 3, baseDelayMs = 5_000) {
-    let attempt = 0;
-    while (true) {
-        const res = await fetch(url, init);
-        if (res.status === 503 || res.status === 429) {
-            if (attempt >= maxRetries)
-                return res;
-            const retryAfter = res.headers.get('retry-after');
-            const delayMs = retryAfter
-                ? parseInt(retryAfter, 10) * 1_000
-                : baseDelayMs * Math.pow(2, attempt);
-            const reason = res.status === 503 ? 'model loading' : 'rate limited';
-            process.stderr.write(`  [ai-summary] ${provider} ${reason} — retrying in ${Math.round(delayMs / 1000)}s (${attempt + 1}/${maxRetries})\n`);
-            await new Promise(r => setTimeout(r, delayMs));
-            attempt++;
-            continue;
-        }
-        return res;
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    let Ollama;
+    try {
+        ({ Ollama } = await esmImport('ollama'));
+    }
+    catch {
+        throw new Error("'ollama' provider requires the ollama package. Install it with: npm install ollama");
     }
+    const client = new Ollama({ host: baseUrl });
+    const response = await client.chat({
+        model,
+        messages: [{ role: 'user', content: buildPrompt(code, contextContent) }],
+    });
+    return parseAiResponse(response.message?.content ?? '', fallbackTitle);
 }
 async function openaiSummary(code, fallbackTitle, model, apiKey, baseUrl, contextContent) {
-    const res = await fetchWithRetry(`${baseUrl}/chat/completions`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json', Authorization: `Bearer ${apiKey}` },
-        body: JSON.stringify({
-            model,
-            messages: [{ role: 'user', content: buildPrompt(code, contextContent) }],
-            temperature: 0,
-        }),
-        signal: AbortSignal.timeout(60_000),
-    }, 'openai');
-    if (!res.ok)
-        throw new Error(`openai ${res.status}: ${await res.text()}`);
-    const data = await res.json();
-    return parseAiResponse(data.choices?.[0]?.message?.content ?? '', fallbackTitle);
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    let OpenAI;
+    try {
+        ({ OpenAI } = await esmImport('openai'));
+    }
+    catch {
+        throw new Error("'openai' provider requires the openai package. Install it with: npm install openai");
+    }
+    const client = new OpenAI({
+        apiKey,
+        ...(baseUrl ? { baseURL: baseUrl.replace(/\/$/, '') } : {}),
+    });
+    const msg = await client.chat.completions.create({
+        model,
+        messages: [{ role: 'user', content: buildPrompt(code, contextContent) }],
+        temperature: 0,
+        max_tokens: 2048,
+    });
+    return parseAiResponse(msg.choices?.[0]?.message?.content ?? '', fallbackTitle);
 }
 async function anthropicSummary(code, fallbackTitle, model, apiKey, baseUrl, contextContent) {
-    const url = `${baseUrl.replace(/\/$/, '')}/messages`;
-    const res = await fetchWithRetry(url, {
-        method: 'POST',
-        headers: {
-            'Content-Type': 'application/json',
-            'x-api-key': apiKey,
-            'anthropic-version': '2023-06-01',
-        },
-        body: JSON.stringify({
-            model,
-            max_tokens: 2048,
-            messages: [{ role: 'user', content: buildPrompt(code, contextContent) }],
-        }),
-        signal: AbortSignal.timeout(60_000),
-    }, 'anthropic');
-    if (!res.ok)
-        throw new Error(`anthropic ${res.status}: ${await res.text()}`);
-    const data = await res.json();
-    return parseAiResponse(data.content?.[0]?.text ?? '', fallbackTitle);
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    let Anthropic;
+    try {
+        ({ default: Anthropic } = await esmImport('@anthropic-ai/sdk'));
+    }
+    catch {
+        throw new Error("'anthropic' provider requires @anthropic-ai/sdk. Install it with: npm install @anthropic-ai/sdk");
+    }
+    const client = new Anthropic({
+        apiKey,
+        ...(baseUrl ? { baseURL: baseUrl.replace(/\/$/, '') } : {}),
+    });
+    const msg = await client.messages.create({
+        model,
+        max_tokens: 2048,
+        messages: [{ role: 'user', content: buildPrompt(code, contextContent) }],
+    });
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    return parseAiResponse(msg.content[0]?.text ?? '', fallbackTitle);
 }
 async function huggingfaceSummary(code, fallbackTitle, model, apiKey, contextContent) {
-    // Hugging Face OpenAI-compatible /v1 endpoint
+    // Hugging Face exposes an OpenAI-compatible /v1 endpoint — use the openai SDK with a custom baseURL
     return openaiSummary(code, fallbackTitle, model, apiKey, 'https://api-inference.huggingface.co/v1', contextContent);
 }
 // ─── Bedrock helpers (retry + credential pre-flight) ─────────────────────────
@@ -863,26 +849,57 @@ async function bedrockSummary(code, fallbackTitle, model, region, contextContent
     return parseAiResponse(raw, fallbackTitle);
 }
 async function azureaiSummary(code, fallbackTitle, model, apiKey, baseUrl, contextContent) {
-    // Azure OpenAI uses api-key header; baseUrl is the full endpoint or resource root
-    let url = baseUrl.replace(/\/$/, '');
-    if (!url.includes('/chat/completions')) {
-        url = `${url}/openai/deployments/${model}/chat/completions?api-version=2024-12-01-preview`;
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    let AzureOpenAI;
+    try {
+        ({ AzureOpenAI } = await esmImport('openai'));
     }
-    const res = await fetchWithRetry(url, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json', 'api-key': apiKey },
-        body: JSON.stringify({
+    catch {
+        throw new Error("'azureai' provider requires the openai package. Install it with: npm install openai");
+    }
+    const client = new AzureOpenAI({
+        apiKey,
+        endpoint: baseUrl.replace(/\/$/, ''),
+        apiVersion: '2024-12-01-preview',
+        deployment: model,
+    });
+    const msg = await client.chat.completions.create({
+        model,
+        messages: [{ role: 'user', content: buildPrompt(code, contextContent) }],
+        temperature: 0,
+        max_tokens: 2048,
+    });
+    return parseAiResponse(msg.choices?.[0]?.message?.content ?? '', fallbackTitle);
+}
+async function githubSummary(code, fallbackTitle, model, apiKey, contextContent) {
+    // GitHub Models is OpenAI-compatible — delegate to openaiSummary with GitHub Models endpoint
+    return openaiSummary(code, fallbackTitle, model, apiKey, 'https://models.inference.ai.azure.com', contextContent);
+}
+async function azureinferenceSummary(code, fallbackTitle, model, apiKey, endpoint, contextContent) {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    let ModelClient, AzureKeyCredential;
+    try {
+        ({ default: ModelClient } = await esmImport('@azure-rest/ai-inference'));
+        ({ AzureKeyCredential } = await esmImport('@azure/core-auth'));
+    }
+    catch {
+        throw new Error("'azureinference' provider requires @azure-rest/ai-inference and @azure/core-auth. " +
+            'Install with: npm install @azure-rest/ai-inference @azure/core-auth');
+    }
+    const client = ModelClient(endpoint.replace(/\/$/, ''), new AzureKeyCredential(apiKey));
+    const response = await client.path('/chat/completions').post({
+        body: {
+            model,
             messages: [{ role: 'user', content: buildPrompt(code, contextContent) }],
             temperature: 0,
             max_tokens: 2048,
-        }),
-        signal: AbortSignal.timeout(60_000),
-    }, 'azureai');
-    if (!res.ok)
-        throw new Error(`Azure AI ${res.status}: ${await res.text()}`);
+        },
+    });
+    if (response.status !== '200') {
+        throw new Error(`Azure AI Inference ${response.status}: ${JSON.stringify(response.body)}`);
+    }
     // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const data = await res.json();
-    return parseAiResponse(data.choices?.[0]?.message?.content ?? '', fallbackTitle);
+    return parseAiResponse(response.body.choices?.[0]?.message?.content ?? '', fallbackTitle);
 }
 function resolveEnvVar(value) {
     if (value.startsWith('$'))
@@ -942,51 +959,86 @@ async function analyzeFailure(testName, errorMessage, stackTrace, opts) {
         let raw = '';
         switch (opts.provider) {
             case 'ollama': {
-                const res = await fetchWithRetry(`${opts.baseUrl ?? 'http://localhost:11434'}/api/generate`, {
-                    method: 'POST',
-                    headers: { 'Content-Type': 'application/json' },
-                    body: JSON.stringify({ model: opts.model ?? 'qwen2.5-coder:7b', prompt, stream: false }),
-                    signal: AbortSignal.timeout(30_000),
-                }, 'ollama');
-                if (!res.ok)
-                    throw new Error(`Ollama ${res.status}`);
-                raw = (await res.json()).response ?? '';
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let OllamaFA;
+                try {
+                    ({ Ollama: OllamaFA } = await esmImport('ollama'));
+                }
+                catch {
+                    throw new Error("ollama package not installed");
+                }
+                const ollamaClient = new OllamaFA({ host: opts.baseUrl ?? 'http://localhost:11434' });
+                const ollamaResp = await ollamaClient.chat({
+                    model: opts.model ?? 'gemma-4-e4b-it',
+                    messages: [{ role: 'user', content: prompt }],
+                });
+                raw = ollamaResp.message?.content ?? '';
                 break;
             }
             case 'openai': {
-                const res = await fetchWithRetry(`${opts.baseUrl ?? 'https://api.openai.com/v1'}/chat/completions`, {
-                    method: 'POST',
-                    headers: { 'Content-Type': 'application/json', Authorization: `Bearer ${resolveEnvVar(opts.apiKey ?? '')}` },
-                    body: JSON.stringify({ model: opts.model ?? 'gpt-4o-mini', messages: [{ role: 'user', content: prompt }], temperature: 0, max_tokens: 256 }),
-                    signal: AbortSignal.timeout(30_000),
-                }, 'openai');
-                if (!res.ok)
-                    throw new Error(`openai ${res.status}`);
-                raw = (await res.json()).choices?.[0]?.message?.content ?? '';
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let OpenAI;
+                try {
+                    ({ OpenAI } = await esmImport('openai'));
+                }
+                catch {
+                    throw new Error("openai package not installed");
+                }
+                const openaiClient = new OpenAI({
+                    apiKey: resolveEnvVar(opts.apiKey ?? ''),
+                    ...(opts.baseUrl ? { baseURL: opts.baseUrl.replace(/\/$/, '') } : {}),
+                });
+                const openaiMsg = await openaiClient.chat.completions.create({
+                    model: opts.model ?? 'gpt-4o-mini',
+                    messages: [{ role: 'user', content: prompt }],
+                    temperature: 0,
+                    max_tokens: 256,
+                });
+                raw = openaiMsg.choices?.[0]?.message?.content ?? '';
                 break;
             }
             case 'anthropic': {
-                const res = await fetchWithRetry(`${(opts.baseUrl ?? 'https://api.anthropic.com/v1').replace(/\/$/, '')}/messages`, {
-                    method: 'POST',
-                    headers: { 'Content-Type': 'application/json', 'x-api-key': resolveEnvVar(opts.apiKey ?? ''), 'anthropic-version': '2023-06-01' },
-                    body: JSON.stringify({ model: opts.model ?? 'claude-haiku-4-5-20251001', max_tokens: 1024, messages: [{ role: 'user', content: prompt }] }),
-                    signal: AbortSignal.timeout(30_000),
-                }, 'anthropic');
-                if (!res.ok)
-                    throw new Error(`anthropic ${res.status}`);
-                raw = (await res.json()).content?.[0]?.text ?? '';
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let Anthropic;
+                try {
+                    ({ default: Anthropic } = await esmImport('@anthropic-ai/sdk'));
+                }
+                catch {
+                    throw new Error("@anthropic-ai/sdk package not installed");
+                }
+                const anthropicClient = new Anthropic({
+                    apiKey: resolveEnvVar(opts.apiKey ?? ''),
+                    ...(opts.baseUrl ? { baseURL: opts.baseUrl.replace(/\/$/, '') } : {}),
+                });
+                const anthropicMsg = await anthropicClient.messages.create({
+                    model: opts.model ?? 'claude-haiku-4-5-20251001',
+                    max_tokens: 1024,
+                    messages: [{ role: 'user', content: prompt }],
+                });
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                raw = anthropicMsg.content[0]?.text ?? '';
                 break;
             }
             case 'huggingface': {
-                const res = await fetchWithRetry('https://api-inference.huggingface.co/v1/chat/completions', {
-                    method: 'POST',
-                    headers: { 'Content-Type': 'application/json', Authorization: `Bearer ${resolveEnvVar(opts.apiKey ?? '')}` },
-                    body: JSON.stringify({ model: opts.model, messages: [{ role: 'user', content: prompt }], max_tokens: 256 }),
-                    signal: AbortSignal.timeout(30_000),
-                }, 'huggingface');
-                if (!res.ok)
-                    throw new Error(`huggingface ${res.status}`);
-                raw = (await res.json()).choices?.[0]?.message?.content ?? '';
+                // Hugging Face OpenAI-compatible endpoint via openai SDK
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let OpenAIhf;
+                try {
+                    ({ OpenAI: OpenAIhf } = await esmImport('openai'));
+                }
+                catch {
+                    throw new Error("openai package not installed");
+                }
+                const hfClient = new OpenAIhf({
+                    apiKey: resolveEnvVar(opts.apiKey ?? ''),
+                    baseURL: 'https://api-inference.huggingface.co/v1',
+                });
+                const hfMsg = await hfClient.chat.completions.create({
+                    model: opts.model,
+                    messages: [{ role: 'user', content: prompt }],
+                    max_tokens: 256,
+                });
+                raw = hfMsg.choices?.[0]?.message?.content ?? '';
                 break;
             }
             case 'bedrock': {
@@ -1014,19 +1066,69 @@ async function analyzeFailure(testName, errorMessage, stackTrace, opts) {
             case 'azureai': {
                 if (!opts.baseUrl)
                     break;
-                let url = opts.baseUrl.replace(/\/$/, '');
-                if (!url.includes('/chat/completions')) {
-                    url = `${url}/openai/deployments/${opts.model ?? 'gpt-4o'}/chat/completions?api-version=2024-12-01-preview`;
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let AzureOpenAI;
+                try {
+                    ({ AzureOpenAI } = await esmImport('openai'));
+                }
+                catch {
+                    throw new Error("openai package not installed");
+                }
+                const azureClient = new AzureOpenAI({
+                    apiKey: resolveEnvVar(opts.apiKey ?? ''),
+                    endpoint: opts.baseUrl.replace(/\/$/, ''),
+                    apiVersion: '2024-12-01-preview',
+                    deployment: opts.model ?? 'gpt-4o',
+                });
+                const azureMsg = await azureClient.chat.completions.create({
+                    model: opts.model ?? 'gpt-4o',
+                    messages: [{ role: 'user', content: prompt }],
+                    temperature: 0,
+                    max_tokens: 256,
+                });
+                raw = azureMsg.choices?.[0]?.message?.content ?? '';
+                break;
+            }
+            case 'github': {
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let OpenAIgh;
+                try {
+                    ({ OpenAI: OpenAIgh } = await esmImport('openai'));
+                }
+                catch {
+                    throw new Error("openai package not installed");
+                }
+                const ghClient = new OpenAIgh({
+                    apiKey: resolveEnvVar(opts.apiKey ?? process.env['GITHUB_TOKEN'] ?? ''),
+                    baseURL: 'https://models.inference.ai.azure.com',
+                });
+                const ghMsg = await ghClient.chat.completions.create({
+                    model: opts.model ?? 'gpt-4o',
+                    messages: [{ role: 'user', content: prompt }],
+                    temperature: 0,
+                    max_tokens: 256,
+                });
+                raw = ghMsg.choices?.[0]?.message?.content ?? '';
+                break;
+            }
+            case 'azureinference': {
+                if (!opts.baseUrl)
+                    break;
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                let ModelClientFA, AzureKeyCredentialFA;
+                try {
+                    ({ default: ModelClientFA } = await esmImport('@azure-rest/ai-inference'));
+                    ({ AzureKeyCredential: AzureKeyCredentialFA } = await esmImport('@azure/core-auth'));
                 }
-                const res = await fetchWithRetry(url, {
-                    method: 'POST',
-                    headers: { 'Content-Type': 'application/json', 'api-key': resolveEnvVar(opts.apiKey ?? '') },
-                    body: JSON.stringify({ messages: [{ role: 'user', content: prompt }], temperature: 0, max_tokens: 256 }),
-                    signal: AbortSignal.timeout(30_000),
-                }, 'azureai');
-                if (!res.ok)
-                    throw new Error(`azureai ${res.status}`);
-                raw = (await res.json()).choices?.[0]?.message?.content ?? '';
+                catch {
+                    throw new Error("@azure-rest/ai-inference package not installed");
+                }
+                const aiClient = ModelClientFA(opts.baseUrl.replace(/\/$/, ''), new AzureKeyCredentialFA(resolveEnvVar(opts.apiKey ?? '')));
+                const aiResp = await aiClient.path('/chat/completions').post({
+                    body: { model: opts.model ?? 'gpt-4o', messages: [{ role: 'user', content: prompt }], temperature: 0, max_tokens: 256 },
+                });
+                // eslint-disable-next-line @typescript-eslint/no-explicit-any
+                raw = aiResp.body.choices?.[0]?.message?.content ?? '';
                 break;
             }
             default:
@@ -1072,7 +1174,7 @@ async function summarizeTest(test, localType, opts) {
             case 'local':
                 return await localLlamaSummary(body, fallbackTitle, opts.model, ctx);
             case 'ollama':
-                return await ollamaSummary(body, fallbackTitle, opts.model ?? 'qwen2.5-coder:7b', opts.baseUrl ?? 'http://localhost:11434', ctx);
+                return await ollamaSummary(body, fallbackTitle, opts.model ?? 'gemma-4-e4b-it', opts.baseUrl ?? 'http://localhost:11434', ctx);
             case 'openai':
                 return await openaiSummary(body, fallbackTitle, opts.model ?? 'gpt-4o-mini', resolveEnvVar(opts.apiKey ?? ''), opts.baseUrl ?? 'https://api.openai.com/v1', ctx);
             case 'anthropic':
@@ -1087,6 +1189,12 @@ async function summarizeTest(test, localType, opts) {
                 if (!opts.baseUrl)
                     throw new Error('azureai provider requires --ai-url <azure-endpoint>');
                 return await azureaiSummary(body, fallbackTitle, opts.model ?? 'gpt-4o', resolveEnvVar(opts.apiKey ?? ''), opts.baseUrl, ctx);
+            case 'github':
+                return await githubSummary(body, fallbackTitle, opts.model ?? 'gpt-4o', resolveEnvVar(opts.apiKey ?? process.env['GITHUB_TOKEN'] ?? ''), ctx);
+            case 'azureinference':
+                if (!opts.baseUrl)
+                    throw new Error('azureinference provider requires --ai-url <endpoint>');
+                return await azureinferenceSummary(body, fallbackTitle, opts.model ?? 'gpt-4o', resolveEnvVar(opts.apiKey ?? ''), opts.baseUrl, ctx);
         }
     }
     catch (err) {