scai 0.1.178 → 1.0.1

package/dist/lib/generate.js

@@ -1,5 +1,24 @@
-// File: lib/generate.ts
+/**
+ * Sends one model request through Ollama and, when enabled, saves exact call inputs for integration review.
+ *
+ * Why this file exists:
+ * - all normal model calls go through one gateway
+ * - recovery and retry behavior should stay consistent across callers
+ * - integration tracing belongs here so callers only provide caller name and exact input context
+ */
 import { Config, readConfig } from '../config.js';
+import { OLLAMA_CHAT_URL, OLLAMA_GENERATE_URL, probeOllamaHealth, restartOllamaServer, } from '../utils/ollamaService.js';
+import { nowLocalIso, startTimer } from '../utils/time.js';
+import { writeLlmTraceRecord } from '../testing/llmTraceSession.js';
+import { getOllamaModelPolicy, } from './ollamaModelPolicy.js';
+class OllamaRequestError extends Error {
+    constructor(message, options) {
+        super(message);
+        this.name = 'OllamaRequestError';
+        this.status = options?.status;
+        this.responseBody = options?.responseBody;
+    }
+}
 /**
  * The generate module uses the local model API to produce output
  * based on the `content` of the input. It returns the response
@@ -7,7 +26,7 @@ import { Config, readConfig } from '../config.js';
  *
  * Assumes the model is already running. Does NOT attempt to start or restart it.
  */
-export async function generate(input) {
+export async function generate(input, traceInput) {
     const model = Config.getModel();
     const { contextLength, maxOutputTokens } = readConfig();
     // Safely build prompt
@@ -16,39 +35,264 @@ export async function generate(input) {
         ? JSON.stringify(input.content, null, 2)
         : input.content || '';
     const prompt = `${queryPart}${contentPart}`.trim();
-    const data = await doGenerate(prompt, model, contextLength, maxOutputTokens);
-    return { query: input.query, data };
+    const policy = getOllamaModelPolicy(model);
+    const initialRequest = buildOllamaRequest({
+        model,
+        prompt,
+        contextLength,
+        maxOutputTokens,
+        policy,
+        transport: policy.transport,
+    });
+    const startedAt = nowLocalIso();
+    const stop = startTimer();
+    try {
+        const result = await doGenerate(prompt, model, contextLength, maxOutputTokens);
+        const timing = finalizeTiming(startedAt, stop, result.attemptCount);
+        const trace = writeGenerateTrace({
+            traceInput,
+            input,
+            model,
+            requestBody: result.requestBody,
+            prompt,
+            timing,
+        });
+        return { query: input.query, data: result.data, trace };
+    }
+    catch (err) {
+        const error = err instanceof Error ? err : new Error(String(err));
+        const tracedError = error;
+        const timing = finalizeTiming(startedAt, stop, inferAttemptCount(error));
+        const trace = writeGenerateTrace({
+            traceInput,
+            input,
+            model,
+            requestBody: tracedError.requestBody ?? initialRequest.body,
+            prompt,
+            timing,
+            errorMessage: error.message,
+        });
+        tracedError.trace = trace ?? timing;
+        throw tracedError;
+    }
 }
 async function doGenerate(prompt, model, contextLength, maxOutputTokens) {
     const REQUEST_TIMEOUT_MS = 270000;
+    const policy = getOllamaModelPolicy(model);
+    let activeRequest = buildOllamaRequest({
+        model,
+        prompt,
+        contextLength,
+        maxOutputTokens,
+        policy,
+        transport: policy.transport,
+    });
+    let attemptCount = 1;
+    try {
+        const data = await sendOllamaRequest(activeRequest, REQUEST_TIMEOUT_MS);
+        return { data, attemptCount, requestBody: activeRequest.body };
+    }
+    catch (err) {
+        if (shouldFallbackToGenerate(policy, activeRequest.transport, err)) {
+            activeRequest = buildOllamaRequest({
+                model,
+                prompt,
+                contextLength,
+                maxOutputTokens,
+                policy,
+                transport: 'generate',
+            });
+            try {
+                const data = await sendOllamaRequest(activeRequest, REQUEST_TIMEOUT_MS);
+                return { data, attemptCount, requestBody: activeRequest.body };
+            }
+            catch (fallbackErr) {
+                err = fallbackErr;
+            }
+        }
+        if (!isRecoverableGenerateError(err)) {
+            attachAttemptCount(err, attemptCount);
+            attachRequestBody(err, activeRequest.body);
+            throw err;
+        }
+        const health = await probeOllamaHealth();
+        if (health.status === 'unreachable') {
+            console.log('⚠️  Ollama is not running. SCAI is restarting the Ollama server and retrying once.');
+        }
+        else {
+            console.log('⚠️  Ollama is not responding. SCAI is restarting the Ollama server and retrying once.');
+        }
+        const restart = await restartOllamaServer();
+        if (!restart.restarted || !restart.ready) {
+            throw new Error(buildRecoveryFailureMessage(health.status, restart.health.detail, true));
+        }
+        console.log('✅ Ollama restarted. Retrying the model request once.');
+        attemptCount += 1;
+        try {
+            const data = await sendOllamaRequest(activeRequest, REQUEST_TIMEOUT_MS);
+            return { data, attemptCount, requestBody: activeRequest.body };
+        }
+        catch (retryErr) {
+            const error = new Error(buildRecoveryFailureMessage(classifyGenerateError(retryErr), retryErr?.message, false));
+            attachAttemptCount(error, attemptCount);
+            attachRequestBody(error, activeRequest.body);
+            throw error;
+        }
+    }
+}
+async function sendOllamaRequest(request, timeoutMs) {
     let res;
     try {
-        res = await fetch('http://localhost:11434/api/generate', {
+        res = await fetch(request.url, {
             method: 'POST',
             headers: { 'Content-Type': 'application/json' },
-            signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
-            body: JSON.stringify({
-                model,
-                prompt,
-                stream: false,
-                options: {
-                    // context window
-                    num_ctx: contextLength,
-                    // limit output to avoid exceeding context
-                    max_output_tokens: maxOutputTokens,
-                },
-            }),
+            signal: AbortSignal.timeout(timeoutMs),
+            body: JSON.stringify(request.body),
         });
     }
     catch (err) {
         if (err?.name === 'TimeoutError' || err?.name === 'AbortError') {
-            throw new Error(`Model request timed out after ${REQUEST_TIMEOUT_MS}ms. Verify Ollama is responsive or retry with a smaller prompt.`);
+            throw new Error(`Model request timed out after ${timeoutMs}ms. Verify Ollama is responsive or retry with a smaller prompt.`);
         }
         throw err;
     }
     if (!res.ok) {
-        throw new Error(`Model request failed with status ${res.status}`);
+        const responseBody = await safeReadResponseText(res);
+        throw new OllamaRequestError(`Model request failed with status ${res.status}`, {
+            status: res.status,
+            responseBody,
+        });
     }
     const data = await res.json();
-    return data.response?.trim() ?? '';
+    return extractResponseText(data, request.transport);
+}
+function isRecoverableGenerateError(err) {
+    return classifyGenerateError(err) !== 'non_recoverable';
+}
+function classifyGenerateError(err) {
+    const message = typeof err?.message === 'string' ? err.message.toLowerCase() : '';
+    if (err?.name === 'TimeoutError' || err?.name === 'AbortError' || message.includes('timed out')) {
+        return 'timeout';
+    }
+    if (message.includes('fetch failed') || message.includes('econnrefused') || message.includes('network')) {
+        return 'unreachable';
+    }
+    if (err instanceof TypeError) {
+        return 'transport';
+    }
+    return 'non_recoverable';
+}
+function buildOllamaRequest(args) {
+    const common = {
+        model: args.model,
+        stream: false,
+        options: {
+            num_ctx: args.contextLength,
+            max_output_tokens: args.maxOutputTokens,
+        },
+    };
+    if (args.policy.sendThinkFlag) {
+        common.think = false;
+    }
+    if (args.transport === 'chat') {
+        return {
+            url: OLLAMA_CHAT_URL,
+            transport: 'chat',
+            body: {
+                ...common,
+                messages: [{ role: 'user', content: args.prompt }],
+            },
+        };
+    }
+    return {
+        url: OLLAMA_GENERATE_URL,
+        transport: 'generate',
+        body: {
+            ...common,
+            prompt: args.prompt,
+        },
+    };
+}
+function shouldFallbackToGenerate(policy, transport, err) {
+    if (transport !== 'chat' || !policy.fallbackToGenerate)
+        return false;
+    if (!(err instanceof OllamaRequestError))
+        return false;
+    if (err.status === 404 || err.status === 405 || err.status === 501) {
+        return true;
+    }
+    if (err.status !== 400) {
+        return false;
+    }
+    const body = err.responseBody?.toLowerCase() ?? '';
+    return body.includes('messages') || body.includes('chat') || body.includes('unsupported');
+}
+function extractResponseText(data, transport) {
+    if (transport === 'chat') {
+        return typeof data?.message?.content === 'string' ? data.message.content.trim() : '';
+    }
+    return typeof data?.response === 'string' ? data.response.trim() : '';
+}
+async function safeReadResponseText(res) {
+    try {
+        const text = await res.text();
+        return text.length > 0 ? text : undefined;
+    }
+    catch {
+        return undefined;
+    }
+}
+function buildRecoveryFailureMessage(kind, detail, beforeRetry) {
+    const suffix = ' Run `scai model status` to inspect the server or `scai model restart` to retry manually.';
+    if (beforeRetry) {
+        if (kind === 'unreachable') {
+            return `Ollama was not running, and SCAI could not restart it. ${detail ?? ''}`.trimEnd() + suffix;
+        }
+        return `Ollama is not responding, and SCAI could not recover it automatically. ${detail ?? ''}`.trimEnd() + suffix;
+    }
+    if (kind === 'timeout') {
+        return `Ollama restarted, but the model request still timed out. ${detail ?? ''}`.trimEnd() + suffix;
+    }
+    return `Ollama restarted, but the model request still failed. ${detail ?? ''}`.trimEnd() + suffix;
+}
+function finalizeTiming(startedAt, stop, attemptCount) {
+    return {
+        startedAt,
+        finishedAt: nowLocalIso(),
+        durationMs: stop(),
+        attemptCount,
+    };
+}
+function writeGenerateTrace(input) {
+    if (!input.traceInput) {
+        return input.timing;
+    }
+    const trace = writeLlmTraceRecord({
+        caller: input.traceInput.caller,
+        query: input.input.query,
+        model: input.model,
+        requestBody: input.requestBody,
+        promptText: input.prompt,
+        inputContext: input.traceInput.inputContext,
+        ...input.timing,
+        ...(input.errorMessage ? { errorMessage: input.errorMessage } : {}),
+    });
+    return trace ? { callId: trace.callId, ...input.timing } : input.timing;
+}
+function attachAttemptCount(error, attemptCount) {
+    if (error && typeof error === "object") {
+        error.attemptCount = attemptCount;
+    }
+}
+function attachRequestBody(error, requestBody) {
+    if (error && typeof error === "object") {
+        error.requestBody = requestBody;
+    }
+}
+function inferAttemptCount(error) {
+    if (error && typeof error === "object" && typeof error.attemptCount === "number") {
+        return error.attemptCount;
+    }
+    return 1;
 }
+//# sourceMappingURL=generate.js.map

package/dist/lib/generate.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"generate.js","sourceRoot":"","sources":["../../src/lib/generate.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AACH,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAElD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,iBAAiB,EACjB,mBAAmB,GACpB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC3D,OAAO,EAAE,mBAAmB,EAAE,MAAM,+BAA+B,CAAC;AACpE,OAAO,EACL,oBAAoB,GAGrB,MAAM,wBAAwB,CAAC;AAyChC,MAAM,kBAAmB,SAAQ,KAAK;IAIpC,YAAY,OAAe,EAAE,OAAoD;QAC/E,KAAK,CAAC,OAAO,CAAC,CAAC;QACf,IAAI,CAAC,IAAI,GAAG,oBAAoB,CAAC;QACjC,IAAI,CAAC,MAAM,GAAG,OAAO,EAAE,MAAM,CAAC;QAC9B,IAAI,CAAC,YAAY,GAAG,OAAO,EAAE,YAAY,CAAC;IAC5C,CAAC;CACF;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,KAAe,EAAE,UAA+B;IAC7E,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,EAAE,CAAC;IAChC,MAAM,EAAE,aAAa,EAAE,eAAe,EAAE,GAAG,UAAU,EAAE,CAAC;IAExD,sBAAsB;IACtB,MAAM,SAAS,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC,gBAAgB,KAAK,CAAC,KAAK,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC;IACvE,MAAM,WAAW,GACf,KAAK,CAAC,OAAO,IAAI,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ;QAChD,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACxC,CAAC,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC;IAE1B,MAAM,MAAM,GAAG,GAAG,SAAS,GAAG,WAAW,EAAE,CAAC,IAAI,EAAE,CAAC;IACnD,MAAM,MAAM,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAC3C,MAAM,cAAc,GAAG,kBAAkB,CAAC;QACxC,KAAK;QACL,MAAM;QACN,aAAa;QACb,eAAe;QACf,MAAM;QACN,SAAS,EAAE,MAAM,CAAC,SAAS;KAC5B,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,WAAW,EAAE,CAAC;IAChC,MAAM,IAAI,GAAG,UAAU,EAAE,CAAC;IAE1B,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,MAAM,EAAE,KAAK,EAAE,aAAa,EAAE,eAAe,CAAC,CAAC;QAC/E,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,EAAE,IAAI,EAAE,MAAM,CAAC,YAAY,CAAC,CAAC;QACpE,MAAM,KAAK,GAAG,kBAAkB,CAAC;YAC/B,UAAU;YACV,KAAK;YACL,KAAK;YACL,WAAW,EAAE,MAAM,CAAC,WAAW;YAC/B,MAAM;YACN,MAAM;SACP,CAAC,CAAC;QAEH,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC;IAC1D,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,KAAK,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC;QAClE,MAAM,WAAW,GAAG,KAA+B,CAAC;QACpD,MAAM,MAAM,GAAG,cAAc,CAAC,SAAS,EAAE,IAAI,EAAE,iBAAiB,CAAC,KAAK,CAAC,CAAC,CAAC;QACzE,MAAM,KAAK,GAAG,kBAAkB,CAAC;YAC/B,UAAU;YACV,KAAK;YACL,KAAK;YACL,WAAW,EAAE,WAAW,CAAC,WAAW,IAAI,cAAc,CAAC,IAAI;YAC3D,MAAM;YACN,MAAM;YACN,YAAY,EAAE,KAAK,CAAC,OAAO;SAC5B,CAAC,CAAC;QAEH,WAAW,CAAC,KAAK,GAAG,KAAK,IAAI,MAAM,CAAC;QACpC,MAAM,WAAW,CAAC;IACpB,CAAC;AACH,CAAC;AAED,KAAK,UAAU,UAAU,CACvB,MAAc,EACd,KAAa,EACb,aAAqB,EACrB,eAAuB;IAEvB,MAAM,kBAAkB,GAAG,MAAO,CAAC;IACnC,MAAM,MAAM,GAAG,oBAAoB,CAAC,KAAK,CAAC,CAAC;IAC3C,IAAI,aAAa,GAAG,kBAAkB,CAAC;QACrC,KAAK;QACL,MAAM;QACN,aAAa;QACb,eAAe;QACf,MAAM;QACN,SAAS,EAAE,MAAM,CAAC,SAAS;KAC5B,CAAC,CAAC;IACH,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,iBAAiB,CAAC,aAAa,EAAE,kBAAkB,CAAC,CAAC;QACxE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,WAAW,EAAE,aAAa,CAAC,IAAI,EAAE,CAAC;IACjE,CAAC;IAAC,OAAO,GAAQ,EAAE,CAAC;QAClB,IAAI,wBAAwB,CAAC,MAAM,EAAE,aAAa,CAAC,SAAS,EAAE,GAAG,CAAC,EAAE,CAAC;YACnE,aAAa,GAAG,kBAAkB,CAAC;gBACjC,KAAK;gBACL,MAAM;gBACN,aAAa;gBACb,eAAe;gBACf,MAAM;gBACN,SAAS,EAAE,UAAU;aACtB,CAAC,CAAC;YAEH,IAAI,CAAC;gBACH,MAAM,IAAI,GAAG,MAAM,iBAAiB,CAAC,aAAa,EAAE,kBAAkB,CAAC,CAAC;gBACxE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,WAAW,EAAE,aAAa,CAAC,IAAI,EAAE,CAAC;YACjE,CAAC;YAAC,OAAO,WAAgB,EAAE,CAAC;gBAC1B,GAAG,GAAG,WAAW,CAAC;YACpB,CAAC;QACH,CAAC;QAED,IAAI,CAAC,0BAA0B,CAAC,GAAG,CAAC,EAAE,CAAC;YACrC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;YACtC,iBAAiB,CAAC,GAAG,EAAE,aAAa,CAAC,IAAI,CAAC,CAAC;YAC3C,MAAM,GAAG,CAAC;QACZ,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,iBAAiB,EAAE,CAAC;QACzC,IAAI,MAAM,CAAC,MAAM,KAAK,aAAa,EAAE,CAAC;YACpC,OAAO,CAAC,GAAG,CAAC,oFAAoF,CAAC,CAAC;QACpG,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,GAAG,CAAC,uFAAuF,CAAC,CAAC;QACvG,CAAC;QAED,MAAM,OAAO,GAAG,MAAM,mBAAmB,EAAE,CAAC;QAC5C,IAAI,CAAC,OAAO,CAAC,SAAS,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;YACzC,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC;QAC3F,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,sDAAsD,CAAC,CAAC;QACpE,YAAY,IAAI,CAAC,CAAC;QAElB,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,iBAAiB,CAAC,aAAa,EAAE,kBAAkB,CAAC,CAAC;YACxE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,WAAW,EAAE,aAAa,CAAC,IAAI,EAAE,CAAC;QACjE,CAAC;QAAC,OAAO,QAAa,EAAE,CAAC;YACvB,MAAM,KAAK,GAAG,IAAI,KAAK,CAAC,2BAA2B,CAAC,qBAAqB,CAAC,QAAQ,CAAC,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC;YAChH,kBAAkB,CAAC,KAAK,EAAE,YAAY,CAAC,CAAC;YACxC,iBAAiB,CAAC,KAAK,EAAE,aAAa,CAAC,IAAI,CAAC,CAAC;YAC7C,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;AACH,CAAC;AAED,KAAK,UAAU,iBAAiB,CAAC,OAAsB,EAAE,SAAiB;IACxE,IAAI,GAAa,CAAC;IAClB,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,KAAK,CAAC,OAAO,CAAC,GAAG,EAAE;YAC7B,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,SAAS,CAAC;YACtC,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC;SACnC,CAAC,CAAC;IACL,CAAC;IAAC,OAAO,GAAQ,EAAE,CAAC;QAClB,IAAI,GAAG,EAAE,IAAI,KAAK,cAAc,IAAI,GAAG,EAAE,IAAI,KAAK,YAAY,EAAE,CAAC;YAC/D,MAAM,IAAI,KAAK,CACb,iCAAiC,SAAS,iEAAiE,CAC5G,CAAC;QACJ,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;IAED,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;QACZ,MAAM,YAAY,GAAG,MAAM,oBAAoB,CAAC,GAAG,CAAC,CAAC;QACrD,MAAM,IAAI,kBAAkB,CAAC,oCAAoC,GAAG,CAAC,MAAM,EAAE,EAAE;YAC7E,MAAM,EAAE,GAAG,CAAC,MAAM;YAClB,YAAY;SACb,CAAC,CAAC;IACL,CAAC;IAED,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC;IAC9B,OAAO,mBAAmB,CAAC,IAAI,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC;AACtD,CAAC;AAED,SAAS,0BAA0B,CAAC,GAAY;IAC9C,OAAO,qBAAqB,CAAC,GAAG,CAAC,KAAK,iBAAiB,CAAC;AAC1D,CAAC;AAED,SAAS,qBAAqB,CAAC,GAAQ;IACrC,MAAM,OAAO,GAAG,OAAO,GAAG,EAAE,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IAElF,IAAI,GAAG,EAAE,IAAI,KAAK,cAAc,IAAI,GAAG,EAAE,IAAI,KAAK,YAAY,IAAI,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAC,EAAE,CAAC;QAChG,OAAO,SAAS,CAAC;IACnB,CAAC;IAED,IAAI,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,EAAE,CAAC;QACxG,OAAO,aAAa,CAAC;IACvB,CAAC;IAED,IAAI,GAAG,YAAY,SAAS,EAAE,CAAC;QAC7B,OAAO,WAAW,CAAC;IACrB,CAAC;IAED,OAAO,iBAAiB,CAAC;AAC3B,CAAC;AAED,SAAS,kBAAkB,CAAC,IAO3B;IACC,MAAM,MAAM,GAA4B;QACtC,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE;YACP,OAAO,EAAE,IAAI,CAAC,aAAa;YAC3B,iBAAiB,EAAE,IAAI,CAAC,eAAe;SACxC;KACF,CAAC;IAEF,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE,CAAC;QAC9B,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC;IACvB,CAAC;IAED,IAAI,IAAI,CAAC,SAAS,KAAK,MAAM,EAAE,CAAC;QAC9B,OAAO;YACL,GAAG,EAAE,eAAe;YACpB,SAAS,EAAE,MAAM;YACjB,IAAI,EAAE;gBACJ,GAAG,MAAM;gBACT,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC;aACnD;SACF,CAAC;IACJ,CAAC;IAED,OAAO;QACL,GAAG,EAAE,mBAAmB;QACxB,SAAS,EAAE,UAAU;QACrB,IAAI,EAAE;YACJ,GAAG,MAAM;YACT,MAAM,EAAE,IAAI,CAAC,MAAM;SACpB;KACF,CAAC;AACJ,CAAC;AAED,SAAS,wBAAwB,CAC/B,MAAyB,EACzB,SAA0B,EAC1B,GAAY;IAEZ,IAAI,SAAS,KAAK,MAAM,IAAI,CAAC,MAAM,CAAC,kBAAkB;QAAE,OAAO,KAAK,CAAC;IACrE,IAAI,CAAC,CAAC,GAAG,YAAY,kBAAkB,CAAC;QAAE,OAAO,KAAK,CAAC;IAEvD,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;QACnE,OAAO,IAAI,CAAC;IACd,CAAC;IAED,IAAI,GAAG,CAAC,MAAM,KAAK,GAAG,EAAE,CAAC;QACvB,OAAO,KAAK,CAAC;IACf,CAAC;IAED,MAAM,IAAI,GAAG,GAAG,CAAC,YAAY,EAAE,WAAW,EAAE,IAAI,EAAE,CAAC;IACnD,OAAO,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,CAAC;AAC5F,CAAC;AAED,SAAS,mBAAmB,CAAC,IAAS,EAAE,SAA0B;IAChE,IAAI,SAAS,KAAK,MAAM,EAAE,CAAC;QACzB,OAAO,OAAO,IAAI,EAAE,OAAO,EAAE,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACvF,CAAC;IAED,OAAO,OAAO,IAAI,EAAE,QAAQ,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;AACxE,CAAC;AAED,KAAK,UAAU,oBAAoB,CAAC,GAAa;IAC/C,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC;QAC9B,OAAO,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IAC5C,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED,SAAS,2BAA2B,CAClC,IAAuF,EACvF,MAA0B,EAC1B,WAAoB;IAEpB,MAAM,MAAM,GAAG,2FAA2F,CAAC;IAE3G,IAAI,WAAW,EAAE,CAAC;QAChB,IAAI,IAAI,KAAK,aAAa,EAAE,CAAC;YAC3B,OAAO,0DAA0D,MAAM,IAAI,EAAE,EAAE,CAAC,OAAO,EAAE,GAAG,MAAM,CAAC;QACrG,CAAC;QAED,OAAO,0EAA0E,MAAM,IAAI,EAAE,EAAE,CAAC,OAAO,EAAE,GAAG,MAAM,CAAC;IACrH,CAAC;IAED,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;QACvB,OAAO,4DAA4D,MAAM,IAAI,EAAE,EAAE,CAAC,OAAO,EAAE,GAAG,MAAM,CAAC;IACvG,CAAC;IAED,OAAO,yDAAyD,MAAM,IAAI,EAAE,EAAE,CAAC,OAAO,EAAE,GAAG,MAAM,CAAC;AACpG,CAAC;AAED,SAAS,cAAc,CAAC,SAAiB,EAAE,IAAkB,EAAE,YAAoB;IACjF,OAAO;QACL,SAAS;QACT,UAAU,EAAE,WAAW,EAAE;QACzB,UAAU,EAAE,IAAI,EAAE;QAClB,YAAY;KACb,CAAC;AACJ,CAAC;AAED,SAAS,kBAAkB,CAAC,KAA8B;IACxD,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,CAAC;QACtB,OAAO,KAAK,CAAC,MAAM,CAAC;IACtB,CAAC;IAED,MAAM,KAAK,GAAG,mBAAmB,CAAC;QAChC,MAAM,EAAE,KAAK,CAAC,UAAU,CAAC,MAAM;QAC/B,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK;QACxB,KAAK,EAAE,KAAK,CAAC,KAAK;QAClB,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,UAAU,EAAE,KAAK,CAAC,MAAM;QACxB,YAAY,EAAE,KAAK,CAAC,UAAU,CAAC,YAAY;QAC3C,GAAG,KAAK,CAAC,MAAM;QACf,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,KAAK,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpE,CAAC,CAAC;IAEH,OAAO,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,MAAM,EAAE,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC;AAC1E,CAAC;AAED,SAAS,kBAAkB,CAAC,KAAc,EAAE,YAAoB;IAC9D,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QACtC,KAAmC,CAAC,YAAY,GAAG,YAAY,CAAC;IACnE,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAc,EAAE,WAAoC;IAC7E,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QACtC,KAAmD,CAAC,WAAW,GAAG,WAAW,CAAC;IACjF,CAAC;AACH,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAc;IACvC,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,OAAQ,KAAoC,CAAC,YAAY,KAAK,QAAQ,EAAE,CAAC;QACjH,OAAQ,KAAkC,CAAC,YAAY,CAAC;IAC1D,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC","sourcesContent":["/**\n * Sends one model request through Ollama and, when enabled, saves exact call inputs for integration review.\n *\n * Why this file exists:\n * - all normal model calls go through one gateway\n * - recovery and retry behavior should stay consistent across callers\n * - integration tracing belongs here so callers only provide caller name and exact input context\n */\nimport { Config, readConfig } from '../config.js';\nimport { ModuleIO } from '../types.js';\nimport {\n  OLLAMA_CHAT_URL,\n  OLLAMA_GENERATE_URL,\n  probeOllamaHealth,\n  restartOllamaServer,\n} from '../utils/ollamaService.js';\nimport { nowLocalIso, startTimer } from '../utils/time.js';\nimport { writeLlmTraceRecord } from '../testing/llmTraceSession.js';\nimport {\n  getOllamaModelPolicy,\n  type OllamaModelPolicy,\n  type OllamaTransport,\n} from './ollamaModelPolicy.js';\n\nexport type GenerateTraceInput = {\n  caller: string;\n  inputContext: unknown;\n};\n\ntype GenerateTiming = {\n  startedAt: string;\n  finishedAt: string;\n  durationMs: number;\n  attemptCount: number;\n};\n\ntype GenerateResult = {\n  data: string;\n  attemptCount: number;\n  requestBody: Record<string, unknown>;\n};\n\ntype GenerateErrorWithTrace = Error & {\n  trace?: ModuleIO[\"trace\"];\n  requestBody?: Record<string, unknown>;\n};\n\ntype GenerateTraceWriteInput = {\n  traceInput?: GenerateTraceInput;\n  input: ModuleIO;\n  model: string;\n  requestBody: Record<string, unknown>;\n  prompt: string;\n  timing: GenerateTiming;\n  errorMessage?: string;\n};\n\ntype OllamaRequest = {\n  url: string;\n  body: Record<string, unknown>;\n  transport: OllamaTransport;\n};\n\nclass OllamaRequestError extends Error {\n  status?: number;\n  responseBody?: string;\n\n  constructor(message: string, options?: { status?: number; responseBody?: string }) {\n    super(message);\n    this.name = 'OllamaRequestError';\n    this.status = options?.status;\n    this.responseBody = options?.responseBody;\n  }\n}\n\n/**\n * The generate module uses the local model API to produce output\n * based on the `content` of the input. It returns the response\n * on the `data` property of ModuleIO.\n *\n * Assumes the model is already running. Does NOT attempt to start or restart it.\n */\nexport async function generate(input: ModuleIO, traceInput?: GenerateTraceInput): Promise<ModuleIO> {\n  const model = Config.getModel();\n  const { contextLength, maxOutputTokens } = readConfig();\n\n  // Safely build prompt\n  const queryPart = input.query ? `User query:\\n${input.query}\\n\\n` : '';\n  const contentPart =\n    input.content && typeof input.content !== 'string'\n      ? JSON.stringify(input.content, null, 2)\n      : input.content || '';\n\n  const prompt = `${queryPart}${contentPart}`.trim();\n  const policy = getOllamaModelPolicy(model);\n  const initialRequest = buildOllamaRequest({\n    model,\n    prompt,\n    contextLength,\n    maxOutputTokens,\n    policy,\n    transport: policy.transport,\n  });\n\n  const startedAt = nowLocalIso();\n  const stop = startTimer();\n\n  try {\n    const result = await doGenerate(prompt, model, contextLength, maxOutputTokens);\n    const timing = finalizeTiming(startedAt, stop, result.attemptCount);\n    const trace = writeGenerateTrace({\n      traceInput,\n      input,\n      model,\n      requestBody: result.requestBody,\n      prompt,\n      timing,\n    });\n\n    return { query: input.query, data: result.data, trace };\n  } catch (err) {\n    const error = err instanceof Error ? err : new Error(String(err));\n    const tracedError = error as GenerateErrorWithTrace;\n    const timing = finalizeTiming(startedAt, stop, inferAttemptCount(error));\n    const trace = writeGenerateTrace({\n      traceInput,\n      input,\n      model,\n      requestBody: tracedError.requestBody ?? initialRequest.body,\n      prompt,\n      timing,\n      errorMessage: error.message,\n    });\n\n    tracedError.trace = trace ?? timing;\n    throw tracedError;\n  }\n}\n\nasync function doGenerate(\n  prompt: string,\n  model: string,\n  contextLength: number,\n  maxOutputTokens: number\n): Promise<GenerateResult> {\n  const REQUEST_TIMEOUT_MS = 270_000;\n  const policy = getOllamaModelPolicy(model);\n  let activeRequest = buildOllamaRequest({\n    model,\n    prompt,\n    contextLength,\n    maxOutputTokens,\n    policy,\n    transport: policy.transport,\n  });\n  let attemptCount = 1;\n\n  try {\n    const data = await sendOllamaRequest(activeRequest, REQUEST_TIMEOUT_MS);\n    return { data, attemptCount, requestBody: activeRequest.body };\n  } catch (err: any) {\n    if (shouldFallbackToGenerate(policy, activeRequest.transport, err)) {\n      activeRequest = buildOllamaRequest({\n        model,\n        prompt,\n        contextLength,\n        maxOutputTokens,\n        policy,\n        transport: 'generate',\n      });\n\n      try {\n        const data = await sendOllamaRequest(activeRequest, REQUEST_TIMEOUT_MS);\n        return { data, attemptCount, requestBody: activeRequest.body };\n      } catch (fallbackErr: any) {\n        err = fallbackErr;\n      }\n    }\n\n    if (!isRecoverableGenerateError(err)) {\n      attachAttemptCount(err, attemptCount);\n      attachRequestBody(err, activeRequest.body);\n      throw err;\n    }\n\n    const health = await probeOllamaHealth();\n    if (health.status === 'unreachable') {\n      console.log('⚠️  Ollama is not running. SCAI is restarting the Ollama server and retrying once.');\n    } else {\n      console.log('⚠️  Ollama is not responding. SCAI is restarting the Ollama server and retrying once.');\n    }\n\n    const restart = await restartOllamaServer();\n    if (!restart.restarted || !restart.ready) {\n      throw new Error(buildRecoveryFailureMessage(health.status, restart.health.detail, true));\n    }\n\n    console.log('✅ Ollama restarted. Retrying the model request once.');\n    attemptCount += 1;\n\n    try {\n      const data = await sendOllamaRequest(activeRequest, REQUEST_TIMEOUT_MS);\n      return { data, attemptCount, requestBody: activeRequest.body };\n    } catch (retryErr: any) {\n      const error = new Error(buildRecoveryFailureMessage(classifyGenerateError(retryErr), retryErr?.message, false));\n      attachAttemptCount(error, attemptCount);\n      attachRequestBody(error, activeRequest.body);\n      throw error;\n    }\n  }\n}\n\nasync function sendOllamaRequest(request: OllamaRequest, timeoutMs: number): Promise<string> {\n  let res: Response;\n  try {\n    res = await fetch(request.url, {\n      method: 'POST',\n      headers: { 'Content-Type': 'application/json' },\n      signal: AbortSignal.timeout(timeoutMs),\n      body: JSON.stringify(request.body),\n    });\n  } catch (err: any) {\n    if (err?.name === 'TimeoutError' || err?.name === 'AbortError') {\n      throw new Error(\n        `Model request timed out after ${timeoutMs}ms. Verify Ollama is responsive or retry with a smaller prompt.`\n      );\n    }\n    throw err;\n  }\n\n  if (!res.ok) {\n    const responseBody = await safeReadResponseText(res);\n    throw new OllamaRequestError(`Model request failed with status ${res.status}`, {\n      status: res.status,\n      responseBody,\n    });\n  }\n\n  const data = await res.json();\n  return extractResponseText(data, request.transport);\n}\n\nfunction isRecoverableGenerateError(err: unknown): boolean {\n  return classifyGenerateError(err) !== 'non_recoverable';\n}\n\nfunction classifyGenerateError(err: any): 'unreachable' | 'timeout' | 'transport' | 'non_recoverable' {\n  const message = typeof err?.message === 'string' ? err.message.toLowerCase() : '';\n\n  if (err?.name === 'TimeoutError' || err?.name === 'AbortError' || message.includes('timed out')) {\n    return 'timeout';\n  }\n\n  if (message.includes('fetch failed') || message.includes('econnrefused') || message.includes('network')) {\n    return 'unreachable';\n  }\n\n  if (err instanceof TypeError) {\n    return 'transport';\n  }\n\n  return 'non_recoverable';\n}\n\nfunction buildOllamaRequest(args: {\n  model: string;\n  prompt: string;\n  contextLength: number;\n  maxOutputTokens: number;\n  policy: OllamaModelPolicy;\n  transport: OllamaTransport;\n}): OllamaRequest {\n  const common: Record<string, unknown> = {\n    model: args.model,\n    stream: false,\n    options: {\n      num_ctx: args.contextLength,\n      max_output_tokens: args.maxOutputTokens,\n    },\n  };\n\n  if (args.policy.sendThinkFlag) {\n    common.think = false;\n  }\n\n  if (args.transport === 'chat') {\n    return {\n      url: OLLAMA_CHAT_URL,\n      transport: 'chat',\n      body: {\n        ...common,\n        messages: [{ role: 'user', content: args.prompt }],\n      },\n    };\n  }\n\n  return {\n    url: OLLAMA_GENERATE_URL,\n    transport: 'generate',\n    body: {\n      ...common,\n      prompt: args.prompt,\n    },\n  };\n}\n\nfunction shouldFallbackToGenerate(\n  policy: OllamaModelPolicy,\n  transport: OllamaTransport,\n  err: unknown\n): boolean {\n  if (transport !== 'chat' || !policy.fallbackToGenerate) return false;\n  if (!(err instanceof OllamaRequestError)) return false;\n\n  if (err.status === 404 || err.status === 405 || err.status === 501) {\n    return true;\n  }\n\n  if (err.status !== 400) {\n    return false;\n  }\n\n  const body = err.responseBody?.toLowerCase() ?? '';\n  return body.includes('messages') || body.includes('chat') || body.includes('unsupported');\n}\n\nfunction extractResponseText(data: any, transport: OllamaTransport): string {\n  if (transport === 'chat') {\n    return typeof data?.message?.content === 'string' ? data.message.content.trim() : '';\n  }\n\n  return typeof data?.response === 'string' ? data.response.trim() : '';\n}\n\nasync function safeReadResponseText(res: Response): Promise<string | undefined> {\n  try {\n    const text = await res.text();\n    return text.length > 0 ? text : undefined;\n  } catch {\n    return undefined;\n  }\n}\n\nfunction buildRecoveryFailureMessage(\n  kind: 'healthy' | 'unreachable' | 'timeout' | 'error' | 'transport' | 'non_recoverable',\n  detail: string | undefined,\n  beforeRetry: boolean\n): string {\n  const suffix = ' Run `scai model status` to inspect the server or `scai model restart` to retry manually.';\n\n  if (beforeRetry) {\n    if (kind === 'unreachable') {\n      return `Ollama was not running, and SCAI could not restart it. ${detail ?? ''}`.trimEnd() + suffix;\n    }\n\n    return `Ollama is not responding, and SCAI could not recover it automatically. ${detail ?? ''}`.trimEnd() + suffix;\n  }\n\n  if (kind === 'timeout') {\n    return `Ollama restarted, but the model request still timed out. ${detail ?? ''}`.trimEnd() + suffix;\n  }\n\n  return `Ollama restarted, but the model request still failed. ${detail ?? ''}`.trimEnd() + suffix;\n}\n\nfunction finalizeTiming(startedAt: string, stop: () => number, attemptCount: number): GenerateTiming {\n  return {\n    startedAt,\n    finishedAt: nowLocalIso(),\n    durationMs: stop(),\n    attemptCount,\n  };\n}\n\nfunction writeGenerateTrace(input: GenerateTraceWriteInput): ModuleIO[\"trace\"] | undefined {\n  if (!input.traceInput) {\n    return input.timing;\n  }\n\n  const trace = writeLlmTraceRecord({\n    caller: input.traceInput.caller,\n    query: input.input.query,\n    model: input.model,\n    requestBody: input.requestBody,\n    promptText: input.prompt,\n    inputContext: input.traceInput.inputContext,\n    ...input.timing,\n    ...(input.errorMessage ? { errorMessage: input.errorMessage } : {}),\n  });\n\n  return trace ? { callId: trace.callId, ...input.timing } : input.timing;\n}\n\nfunction attachAttemptCount(error: unknown, attemptCount: number): void {\n  if (error && typeof error === \"object\") {\n    (error as { attemptCount?: number }).attemptCount = attemptCount;\n  }\n}\n\nfunction attachRequestBody(error: unknown, requestBody: Record<string, unknown>): void {\n  if (error && typeof error === \"object\") {\n    (error as { requestBody?: Record<string, unknown> }).requestBody = requestBody;\n  }\n}\n\nfunction inferAttemptCount(error: unknown): number {\n  if (error && typeof error === \"object\" && typeof (error as { attemptCount?: unknown }).attemptCount === \"number\") {\n    return (error as { attemptCount: number }).attemptCount;\n  }\n  return 1;\n}\n"]}

package/dist/lib/generateFolderCapsules.js

@@ -107,3 +107,4 @@ ${JSON.stringify(inputPayload, null, 2)}
         console.log(`📦 Folder capsule generated for ${folder.path}`);
     }
 }
+//# sourceMappingURL=generateFolderCapsules.js.map

package/dist/lib/generateFolderCapsules.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"generateFolderCapsules.js","sourceRoot":"","sources":["../../src/lib/generateFolderCapsules.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAC;AAC/C,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAE,MAAM,sCAAsC,CAAC;AAGrE,OAAO,IAAI,MAAM,MAAM,CAAC;AAWxB,MAAM,CAAC,KAAK,UAAU,sBAAsB;IACxC,MAAM,EAAE,GAAG,YAAY,EAAE,CAAC;IAE1B,kCAAkC;IAClC,MAAM,KAAK,GAAG,EAAE;SACX,OAAO,CAAC,wBAAwB,CAAC;SACjC,GAAG,EAAmB,CAAC;IAE5B,gCAAgC;IAChC,MAAM,OAAO,GAAG,IAAI,GAAG,EAAuB,CAAC;IAE/C,KAAK,MAAM,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,KAAK,EAAE,CAAC;QACrC,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,IAAI,GAAG,CAAC;QAC1C,MAAM,KAAK,GAAG,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC;QAEtD,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YACpB,OAAO,CAAC,GAAG,CAAC,GAAG,EAAE;gBACb,IAAI,EAAE,GAAG;gBACT,KAAK;gBACL,KAAK,EAAE,EAAE;aACZ,CAAC,CAAC;QACP,CAAC;QAED,OAAO,CAAC,GAAG,CAAC,GAAG,CAAE,CAAC,KAAK,CAAC,IAAI,CAAC;YACzB,IAAI,EAAE,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC;YAC7B,GAAG,EAAE,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,CAAC,IAAI,SAAS;SAC5D,CAAC,CAAC;IACP,CAAC;IAED,0CAA0C;IAC1C,KAAK,MAAM,MAAM,IAAI,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC;QACpC,MAAM,MAAM,GAA2B,EAAE,CAAC;QAC1C,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;YAC3B,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;QAC7C,CAAC;QAED,MAAM,YAAY,GAAG;YACjB,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,SAAS,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM;YAC9B,MAAM;YACN,KAAK,EAAE,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;SACzC,CAAC;QAEF,MAAM,KAAK,GAAa;YACpB,KAAK,EAAE,gCAAgC,MAAM,CAAC,IAAI,EAAE;YACpD,OAAO,EAAE;;;;;;;;;;;;;;;;;;;;;;;EAuBnB,IAAI,CAAC,SAAS,CAAC,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC;OAChC,CAAC,IAAI,EAAE;SACL,CAAC;QAEF,MAAM,QAAQ,GAAG,MAAM,QAAQ,CAAC,KAAK,CAAC,CAAC;QAEvC,+BAA+B;QAC/B,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC,GAAG,CAAC;YACpC,KAAK,EAAE,KAAK,CAAC,KAAK;YAClB,OAAO,EAAE,QAAQ,CAAC,OAAO;SAC5B,CAAC,CAAC;QAEH,IAAI,OAAO,GAAyB,IAAI,CAAC;QAEzC,IAAI,CAAC;YACD,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QAClD,CAAC;QAAC,MAAM,CAAC;YACL,OAAO,CAAC,IAAI,CAAC,wCAAwC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;YACpE,SAAS;QACb,CAAC;QAED,EAAE,CAAC,OAAO,CAAC;;;;;;;;;;;;;;;;;KAiBd,CAAC,CAAC,GAAG,CAAC;YACC,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC;YACrC,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI,IAAI;YACvC,iBAAiB,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM;SACzC,CAAC,CAAC;QAEH,OAAO,CAAC,GAAG,CAAC,mCAAmC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;IAClE,CAAC;AACL,CAAC","sourcesContent":["import { getDbForRepo } from \"../db/client.js\";\nimport { generate } from \"../lib/generate.js\";\nimport { cleanupModule } from \"../pipeline/modules/cleanupModule.js\";\n\nimport { ModuleIO, FolderCapsule, FilePathRow } from \"../types.js\";\nimport path from \"path\";\n\ntype FolderFacts = {\n    path: string;\n    depth: number;\n    files: {\n        name: string;\n        ext: string;\n    }[];\n};\n\nexport async function generateFolderCapsules() {\n    const db = getDbForRepo();\n\n    // --- Load files (paths only) ---\n    const files = db\n        .prepare(`SELECT path FROM files`)\n        .all() as FilePathRow[];\n\n    // --- Group files by folder ---\n    const folders = new Map<string, FolderFacts>();\n\n    for (const { path: filePath } of files) {\n        const dir = path.dirname(filePath) || \"/\";\n        const depth = dir === \"/\" ? 0 : dir.split(\"/\").length;\n\n        if (!folders.has(dir)) {\n            folders.set(dir, {\n                path: dir,\n                depth,\n                files: [],\n            });\n        }\n\n        folders.get(dir)!.files.push({\n            name: path.basename(filePath),\n            ext: path.extname(filePath).replace(\".\", \"\") || \"unknown\",\n        });\n    }\n\n    // --- Generate one capsule per folder ---\n    for (const folder of folders.values()) {\n        const byType: Record<string, number> = {};\n        for (const f of folder.files) {\n            byType[f.ext] = (byType[f.ext] ?? 0) + 1;\n        }\n\n        const inputPayload = {\n            path: folder.path,\n            depth: folder.depth,\n            fileCount: folder.files.length,\n            byType,\n            files: folder.files.map((f) => f.name),\n        };\n\n        const input: ModuleIO = {\n            query: `Generate a FolderCapsule for ${folder.path}`,\n            content: `\nYou are generating a structured FolderCapsule describing the architectural role of a folder.\n\nReturn ONLY valid JSON matching this TypeScript interface:\n\ninterface FolderCapsule {\n  path: string;\n  depth: number;\n  stats: {\n    fileCount: number;\n    byType: Record<string, number>;\n  };\n  roles: string[];\n  concerns: string[];\n  keyFiles: { path: string; reason: string }[];\n  dependencies: {\n    importsFrom: string[];\n    usedBy: string[];\n  };\n  confidence: number;\n}\n\nFolder facts:\n${JSON.stringify(inputPayload, null, 2)}\n      `.trim(),\n        };\n\n        const response = await generate(input);\n\n        // --- Cleanup model output ---\n        const cleaned = await cleanupModule.run({\n            query: input.query,\n            content: response.content,\n        });\n\n        let capsule: FolderCapsule | null = null;\n\n        try {\n            capsule = JSON.parse(String(cleaned.content));\n        } catch {\n            console.warn(`⚠️ Failed to parse FolderCapsule for ${folder.path}`);\n            continue;\n        }\n\n        db.prepare(`\n      INSERT INTO folder_capsules\n        (path, depth, capsule_json, confidence, last_generated, source_file_count)\n      VALUES\n        (\n          @path,\n          @depth,\n          @capsule_json,\n          @confidence,\n          datetime('now'),\n          @source_file_count\n        )\n      ON CONFLICT(path) DO UPDATE SET\n        capsule_json = excluded.capsule_json,\n        confidence = excluded.confidence,\n        last_generated = datetime('now'),\n        source_file_count = excluded.source_file_count\n    `).run({\n            path: folder.path,\n            depth: folder.depth,\n            capsule_json: JSON.stringify(capsule),\n            confidence: capsule?.confidence ?? null,\n            source_file_count: folder.files.length,\n        });\n\n        console.log(`📦 Folder capsule generated for ${folder.path}`);\n    }\n}\n"]}

package/dist/lib/ollamaModelPolicy.js

@@ -0,0 +1,59 @@
+/**
+ * Resolves one shared Ollama request policy from a model tag.
+ *
+ * Why this file exists:
+ * - all model-family request behavior should stay in one place
+ * - transport, thinking defaults, and fallback rules should not be duplicated
+ * - table-driven matching keeps future model-family updates small and testable
+ */
+const DEFAULT_OLLAMA_MODEL_POLICY = {
+    transport: "chat",
+    fallbackToGenerate: true,
+    thinkingDefault: "unsupported",
+    sendThinkFlag: false,
+};
+const OLLAMA_MODEL_POLICY_RULES = [
+    {
+        prefixes: [
+            "qwen3-coder:",
+            "qwen3:",
+            "qwen3.5:",
+            "qwen3.6:",
+            "qwen3-next:",
+            "qwen3-vl:",
+            "aratan/qwen3.5-agent-multimodal:",
+            "gpt-oss:",
+            "nemotron-cascade-2:",
+            "gemma4:",
+            "granite4.1:",
+        ],
+        transport: "chat",
+        fallbackToGenerate: true,
+        thinkingDefault: "off",
+    },
+    {
+        prefixes: [
+            "glm-4.7-flash:",
+        ],
+        transport: "chat",
+        fallbackToGenerate: true,
+        thinkingDefault: "unsupported",
+    },
+];
+export function getOllamaModelPolicy(model) {
+    const normalized = normalizeModelTag(model);
+    const matchedRule = OLLAMA_MODEL_POLICY_RULES.find((rule) => rule.prefixes.some((prefix) => normalized.startsWith(prefix)));
+    if (!matchedRule) {
+        return { ...DEFAULT_OLLAMA_MODEL_POLICY };
+    }
+    return {
+        transport: matchedRule.transport,
+        fallbackToGenerate: matchedRule.fallbackToGenerate,
+        thinkingDefault: matchedRule.thinkingDefault,
+        sendThinkFlag: matchedRule.thinkingDefault === "off",
+    };
+}
+function normalizeModelTag(model) {
+    return String(model ?? "").trim().toLowerCase();
+}
+//# sourceMappingURL=ollamaModelPolicy.js.map

package/dist/lib/ollamaModelPolicy.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ollamaModelPolicy.js","sourceRoot":"","sources":["../../src/lib/ollamaModelPolicy.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAmBH,MAAM,2BAA2B,GAAsB;IACrD,SAAS,EAAE,MAAM;IACjB,kBAAkB,EAAE,IAAI;IACxB,eAAe,EAAE,aAAa;IAC9B,aAAa,EAAE,KAAK;CACrB,CAAC;AAEF,MAAM,yBAAyB,GAA4B;IACzD;QACE,QAAQ,EAAE;YACR,cAAc;YACd,QAAQ;YACR,UAAU;YACV,UAAU;YACV,aAAa;YACb,WAAW;YACX,kCAAkC;YAClC,UAAU;YACV,qBAAqB;YACrB,SAAS;YACT,aAAa;SACd;QACD,SAAS,EAAE,MAAM;QACjB,kBAAkB,EAAE,IAAI;QACxB,eAAe,EAAE,KAAK;KACvB;IACD;QACE,QAAQ,EAAE;YACR,gBAAgB;SACjB;QACD,SAAS,EAAE,MAAM;QACjB,kBAAkB,EAAE,IAAI;QACxB,eAAe,EAAE,aAAa;KAC/B;CACF,CAAC;AAEF,MAAM,UAAU,oBAAoB,CAAC,KAAa;IAChD,MAAM,UAAU,GAAG,iBAAiB,CAAC,KAAK,CAAC,CAAC;IAC5C,MAAM,WAAW,GAAG,yBAAyB,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAC1D,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,CAC9D,CAAC;IAEF,IAAI,CAAC,WAAW,EAAE,CAAC;QACjB,OAAO,EAAE,GAAG,2BAA2B,EAAE,CAAC;IAC5C,CAAC;IAED,OAAO;QACL,SAAS,EAAE,WAAW,CAAC,SAAS;QAChC,kBAAkB,EAAE,WAAW,CAAC,kBAAkB;QAClD,eAAe,EAAE,WAAW,CAAC,eAAe;QAC5C,aAAa,EAAE,WAAW,CAAC,eAAe,KAAK,KAAK;KACrD,CAAC;AACJ,CAAC;AAED,SAAS,iBAAiB,CAAC,KAAa;IACtC,OAAO,MAAM,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AAClD,CAAC","sourcesContent":["/**\n * Resolves one shared Ollama request policy from a model tag.\n *\n * Why this file exists:\n * - all model-family request behavior should stay in one place\n * - transport, thinking defaults, and fallback rules should not be duplicated\n * - table-driven matching keeps future model-family updates small and testable\n */\n\nexport type OllamaTransport = \"chat\" | \"generate\";\nexport type OllamaThinkingDefault = \"off\" | \"unsupported\";\n\nexport type OllamaModelPolicy = {\n  transport: OllamaTransport;\n  fallbackToGenerate: boolean;\n  thinkingDefault: OllamaThinkingDefault;\n  sendThinkFlag: boolean;\n};\n\ntype OllamaModelPolicyRule = {\n  prefixes: string[];\n  transport: OllamaTransport;\n  fallbackToGenerate: boolean;\n  thinkingDefault: OllamaThinkingDefault;\n};\n\nconst DEFAULT_OLLAMA_MODEL_POLICY: OllamaModelPolicy = {\n  transport: \"chat\",\n  fallbackToGenerate: true,\n  thinkingDefault: \"unsupported\",\n  sendThinkFlag: false,\n};\n\nconst OLLAMA_MODEL_POLICY_RULES: OllamaModelPolicyRule[] = [\n  {\n    prefixes: [\n      \"qwen3-coder:\",\n      \"qwen3:\",\n      \"qwen3.5:\",\n      \"qwen3.6:\",\n      \"qwen3-next:\",\n      \"qwen3-vl:\",\n      \"aratan/qwen3.5-agent-multimodal:\",\n      \"gpt-oss:\",\n      \"nemotron-cascade-2:\",\n      \"gemma4:\",\n      \"granite4.1:\",\n    ],\n    transport: \"chat\",\n    fallbackToGenerate: true,\n    thinkingDefault: \"off\",\n  },\n  {\n    prefixes: [\n      \"glm-4.7-flash:\",\n    ],\n    transport: \"chat\",\n    fallbackToGenerate: true,\n    thinkingDefault: \"unsupported\",\n  },\n];\n\nexport function getOllamaModelPolicy(model: string): OllamaModelPolicy {\n  const normalized = normalizeModelTag(model);\n  const matchedRule = OLLAMA_MODEL_POLICY_RULES.find((rule) =>\n    rule.prefixes.some((prefix) => normalized.startsWith(prefix))\n  );\n\n  if (!matchedRule) {\n    return { ...DEFAULT_OLLAMA_MODEL_POLICY };\n  }\n\n  return {\n    transport: matchedRule.transport,\n    fallbackToGenerate: matchedRule.fallbackToGenerate,\n    thinkingDefault: matchedRule.thinkingDefault,\n    sendThinkFlag: matchedRule.thinkingDefault === \"off\",\n  };\n}\n\nfunction normalizeModelTag(model: string): string {\n  return String(model ?? \"\").trim().toLowerCase();\n}\n"]}

package/dist/lib/spinner.js

@@ -1,3 +1,10 @@
+/**
+ * Renders one in-place CLI spinner line without leaking stale text between updates.
+ *
+ * Why this exists:
+ * - durable agent messages should pause cleanly while long-running work is active
+ * - spinner redraws must clear the current row first so shorter status text does not leave old tails behind
+ */
 export class Spinner {
     constructor(message) {
         this.frames = ['⠋', '⠙', '⠹', '⠸', '⠼', '⠴', '⠦', '⠧', '⠇', '⠏'];
@@ -9,29 +16,30 @@ export class Spinner {
     }
     start() {
         process.stdout.write('\x1b[?25l'); // hide cursor
-        if (this.text) {
-            process.stdout.write(` ${this.text}`);
-        }
+        this.render();
         this.interval = setInterval(() => {
-            const text = this.text ?? '';
-            process.stdout.write('\r');
-            process.stdout.write(`${this.frames[this.i]} ${text}`);
-            this.i = (this.i + 1) % this.frames.length;
+            this.render();
         }, 80);
     }
+    isActive() {
+        return this.interval !== null;
+    }
     update(text) {
         this.text = text;
+        if (this.interval) {
+            this.render();
+        }
     }
     succeed(msg) {
         this.stop();
-        process.stdout.write('\r'); // go to start
+        process.stdout.write('\r\x1b[K'); // go to start + clear line
         if (msg) {
             console.log(`✅ ${msg}`);
         }
     }
     fail(msg) {
         this.stop();
-        process.stdout.write('\r'); // go to start
+        process.stdout.write('\r\x1b[K'); // go to start + clear line
         console.log(`❌ ${msg}`);
     }
     stop() {
@@ -39,6 +47,18 @@ export class Spinner {
             clearInterval(this.interval);
             this.interval = null;
         }
+        process.stdout.write('\r\x1b[K'); // clear any partial spinner line before normal output resumes
         process.stdout.write('\x1b[?25h'); // show cursor
     }
+    /**
+     * Repaints the full spinner row.
+     * Example: switching from "long-module-name.ts" to "a.ts" clears the stale tail first.
+     */
+    render() {
+        const text = this.text ?? '';
+        process.stdout.write('\r\x1b[K');
+        process.stdout.write(`${this.frames[this.i]} ${text}`);
+        this.i = (this.i + 1) % this.frames.length;
+    }
 }
+//# sourceMappingURL=spinner.js.map

package/dist/lib/spinner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"spinner.js","sourceRoot":"","sources":["../../src/lib/spinner.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,MAAM,OAAO,OAAO;IAMlB,YAAY,OAAgB;QALpB,WAAM,GAAG,CAAC,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;QAC5D,aAAQ,GAA0C,IAAI,CAAC;QAEvD,MAAC,GAAG,CAAC,CAAC;QAGZ,IAAI,OAAO,EAAE,CAAC;YACZ,IAAI,CAAC,IAAI,GAAG,OAAO,CAAC;QACtB,CAAC;IACH,CAAC;IAED,KAAK;QACH,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,cAAc;QACjD,IAAI,CAAC,MAAM,EAAE,CAAC;QAEd,IAAI,CAAC,QAAQ,GAAG,WAAW,CAAC,GAAG,EAAE;YAC/B,IAAI,CAAC,MAAM,EAAE,CAAC;QAChB,CAAC,EAAE,EAAE,CAAC,CAAC;IACT,CAAC;IAED,QAAQ;QACN,OAAO,IAAI,CAAC,QAAQ,KAAK,IAAI,CAAC;IAChC,CAAC;IAGD,MAAM,CAAC,IAAY;QACjB,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAClB,IAAI,CAAC,MAAM,EAAE,CAAC;QAChB,CAAC;IACH,CAAC;IAED,OAAO,CAAC,GAAY;QAClB,IAAI,CAAC,IAAI,EAAE,CAAC;QACZ,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,2BAA2B;QAE7D,IAAI,GAAG,EAAE,CAAC;YACR,OAAO,CAAC,GAAG,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC;QAC1B,CAAC;IACH,CAAC;IAED,IAAI,CAAC,GAAW;QACd,IAAI,CAAC,IAAI,EAAE,CAAC;QACZ,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,2BAA2B;QAC7D,OAAO,CAAC,GAAG,CAAC,KAAK,GAAG,EAAE,CAAC,CAAC;IAC1B,CAAC;IAED,IAAI;QACF,IAAI,IAAI,CAAC,QAAQ,EAAE,CAAC;YAClB,aAAa,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC7B,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACvB,CAAC;QACD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,8DAA8D;QAChG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,cAAc;IACnD,CAAC;IAED;;;OAGG;IACK,MAAM;QACZ,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,EAAE,CAAC;QAC7B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;QACjC,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;QACvD,IAAI,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC;IAC7C,CAAC;CACF","sourcesContent":["/**\n * Renders one in-place CLI spinner line without leaking stale text between updates.\n *\n * Why this exists:\n * - durable agent messages should pause cleanly while long-running work is active\n * - spinner redraws must clear the current row first so shorter status text does not leave old tails behind\n */\nexport class Spinner {\n  private frames = ['⠋', '⠙', '⠹', '⠸', '⠼', '⠴', '⠦', '⠧', '⠇', '⠏'];\n  private interval: ReturnType<typeof setInterval> | null = null;\n  private text?: string;\n  private i = 0;\n\r\n  constructor(message?: string) {\r\n    if (message) {\r\n      this.text = message;\r\n    }\r\n  }\r\n\r\n  start() {\n    process.stdout.write('\\x1b[?25l'); // hide cursor\n    this.render();\n\n    this.interval = setInterval(() => {\n      this.render();\n    }, 80);\n  }\n\n  isActive() {\n    return this.interval !== null;\n  }\n\n\n  update(text: string) {\n    this.text = text;\n    if (this.interval) {\n      this.render();\n    }\n  }\n\r\n  succeed(msg?: string) {\n    this.stop();\n    process.stdout.write('\\r\\x1b[K'); // go to start + clear line\n\n    if (msg) {\n      console.log(`✅ ${msg}`);\n    }\n  }\n\n  fail(msg: string) {\n    this.stop();\n    process.stdout.write('\\r\\x1b[K'); // go to start + clear line\n    console.log(`❌ ${msg}`);\n  }\n\r\n  stop() {\n    if (this.interval) {\n      clearInterval(this.interval);\n      this.interval = null;\n    }\n    process.stdout.write('\\r\\x1b[K'); // clear any partial spinner line before normal output resumes\n    process.stdout.write('\\x1b[?25h'); // show cursor\n  }\n\n  /**\n   * Repaints the full spinner row.\n   * Example: switching from \"long-module-name.ts\" to \"a.ts\" clears the stale tail first.\n   */\n  private render() {\n    const text = this.text ?? '';\n    process.stdout.write('\\r\\x1b[K');\n    process.stdout.write(`${this.frames[this.i]} ${text}`);\n    this.i = (this.i + 1) % this.frames.length;\n  }\n}\n"]}