@steipete/oracle 0.9.0 → 0.10.0

package/dist/src/oracle/claude.js

@@ -1,11 +1,11 @@
-const DEFAULT_CLAUDE_ENDPOINT = 'https://api.anthropic.com/v1/messages';
-const ANTHROPIC_VERSION = '2023-06-01';
+const DEFAULT_CLAUDE_ENDPOINT = "https://api.anthropic.com/v1/messages";
+const ANTHROPIC_VERSION = "2023-06-01";
 function extractPrompt(body) {
     const first = body.input?.[0]?.content?.[0];
-    if (first && first.type === 'input_text') {
-        return first.text ?? '';
+    if (first && first.type === "input_text") {
+        return first.text ?? "";
     }
-    return '';
+    return "";
 }
 async function callClaude({ apiKey, model, prompt, endpoint, stream = false, }) {
     const url = endpoint?.trim() || DEFAULT_CLAUDE_ENDPOINT;
@@ -14,34 +14,45 @@ async function callClaude({ apiKey, model, prompt, endpoint, stream = false, })
         max_tokens: 2048,
         messages: [
             {
-                role: 'user',
+                role: "user",
                 content: prompt,
             },
         ],
         stream,
     };
     return fetch(url, {
-        method: 'POST',
+        method: "POST",
         headers: {
-            'content-type': 'application/json',
-            'x-api-key': apiKey,
-            'anthropic-version': ANTHROPIC_VERSION,
+            "content-type": "application/json",
+            "x-api-key": apiKey,
+            "anthropic-version": ANTHROPIC_VERSION,
         },
         body: JSON.stringify(payload),
     });
 }
 async function parseClaudeResponse(raw) {
-    const json = (await raw.json());
+    const body = await raw.text();
+    if (!body.trim()) {
+        throw new Error(`Claude request failed (${raw.status} ${raw.statusText || "unknown status"}): empty response`);
+    }
+    let json;
+    try {
+        json = JSON.parse(body);
+    }
+    catch (error) {
+        const snippet = body.slice(0, 160).replace(/\s+/g, " ").trim();
+        throw new Error(`Claude request failed (${raw.status} ${raw.statusText || "unknown status"}): invalid JSON response${snippet ? `: ${snippet}` : ""}`, { cause: error });
+    }
     if (json.error) {
-        throw new Error(json.error.message || 'Claude request failed');
+        throw new Error(json.error.message || "Claude request failed");
     }
-    const textParts = json.content?.map((part) => part.text ?? '').filter(Boolean) ?? [];
-    const outputText = textParts.join('');
+    const textParts = json.content?.map((part) => part.text ?? "").filter(Boolean) ?? [];
+    const outputText = textParts.join("");
     return {
         id: json.id ?? `claude-${Date.now()}`,
-        status: 'completed',
+        status: "completed",
         output_text: [outputText],
-        output: [{ type: 'text', text: outputText }],
+        output: [{ type: "text", text: outputText }],
         usage: {
             input_tokens: json.usage?.input_tokens ?? 0,
             output_tokens: json.usage?.output_tokens ?? 0,
@@ -53,11 +64,17 @@ export function createClaudeClient(apiKey, modelName, resolvedModelId, baseUrl)
     const modelId = resolveClaudeModelId(resolvedModelId ?? modelName);
     const stream = async (body) => {
         const prompt = extractPrompt(body);
-        const resp = await callClaude({ apiKey, model: modelId, prompt, stream: false, endpoint: baseUrl });
+        const resp = await callClaude({
+            apiKey,
+            model: modelId,
+            prompt,
+            stream: false,
+            endpoint: baseUrl,
+        });
         const parsed = await parseClaudeResponse(resp);
         const iterator = async function* () {
             if (parsed.output_text?.[0]) {
-                yield { type: 'response.output_text.delta', delta: parsed.output_text[0] };
+                yield { type: "response.output_text.delta", delta: parsed.output_text[0] };
             }
             return;
         };
@@ -68,13 +85,19 @@ export function createClaudeClient(apiKey, modelName, resolvedModelId, baseUrl)
     };
     const create = async (body) => {
         const prompt = extractPrompt(body);
-        const resp = await callClaude({ apiKey, model: modelId, prompt, stream: false, endpoint: baseUrl });
+        const resp = await callClaude({
+            apiKey,
+            model: modelId,
+            prompt,
+            stream: false,
+            endpoint: baseUrl,
+        });
         return parseClaudeResponse(resp);
     };
     const retrieve = async (id) => ({
         id,
-        status: 'error',
-        error: { message: 'Retrieve by ID not supported for Claude API yet.' },
+        status: "error",
+        error: { message: "Retrieve by ID not supported for Claude API yet." },
     });
     return {
         responses: {
@@ -85,11 +108,16 @@ export function createClaudeClient(apiKey, modelName, resolvedModelId, baseUrl)
     };
 }
 export function resolveClaudeModelId(modelName) {
-    if (modelName === 'claude-4.5-sonnet' || modelName === 'claude-sonnet-4-5-20241022') {
-        return 'claude-sonnet-4-5';
+    if (modelName === "claude-4.6-sonnet" || modelName === "claude-sonnet-4-6") {
+        return "claude-sonnet-4-6";
+    }
+    if (modelName === "claude-4.5-sonnet" ||
+        modelName === "claude-sonnet-4-5" ||
+        modelName === "claude-sonnet-4-5-20250929") {
+        return "claude-sonnet-4-5";
     }
-    if (modelName === 'claude-4.1-opus' || modelName === 'claude-opus-4-1-20240808') {
-        return 'claude-opus-4-1';
+    if (modelName === "claude-4.1-opus" || modelName === "claude-opus-4-1-20240808") {
+        return "claude-opus-4-1";
     }
     return modelName;
 }

package/dist/src/oracle/client.js

@@ -1,19 +1,19 @@
-import OpenAI from 'openai';
-import path from 'node:path';
-import { createRequire } from 'node:module';
-import { createGeminiClient } from './gemini.js';
-import { createClaudeClient } from './claude.js';
-import { isOpenRouterBaseUrl } from './modelResolver.js';
+import OpenAI from "openai";
+import path from "node:path";
+import { createRequire } from "node:module";
+import { createGeminiClient } from "./gemini.js";
+import { createClaudeClient } from "./claude.js";
+import { isOpenRouterBaseUrl } from "./modelResolver.js";
 /**
  * Known native API base URLs that should still use their dedicated SDKs.
  * Any other custom base URL is treated as an OpenAI-compatible proxy and
  * all models are routed through the chat/completions adapter.
  */
 const NATIVE_API_HOSTS = [
-    'api.openai.com',
-    'api.anthropic.com',
-    'generativelanguage.googleapis.com',
-    'api.x.ai',
+    "api.openai.com",
+    "api.anthropic.com",
+    "generativelanguage.googleapis.com",
+    "api.x.ai",
 ];
 export function isCustomBaseUrl(baseUrl) {
     if (!baseUrl)
@@ -27,7 +27,7 @@ export function isCustomBaseUrl(baseUrl) {
     }
 }
 export function buildAzureResponsesBaseUrl(endpoint) {
-    return `${endpoint.replace(/\/+$/, '')}/openai/v1`;
+    return `${endpoint.replace(/\/+$/, "")}/openai/v1`;
 }
 export function createDefaultClientFactory() {
     const customFactory = loadCustomClientFactory();
@@ -40,17 +40,21 @@ export function createDefaultClientFactory() {
         // route ALL models through the OpenAI chat/completions adapter instead of native SDKs
         // which would reject the proxy's API key.
         if (!openRouter && !customProxy) {
-            if (options?.model?.startsWith('gemini')) {
+            if (options?.model?.startsWith("gemini")) {
                 // Gemini client uses its own SDK; allow passing the already-resolved id for transparency/logging.
                 return createGeminiClient(key, options.model, options.resolvedModelId);
             }
-            if (options?.model?.startsWith('claude')) {
+            if (options?.model?.startsWith("claude")) {
                 return createClaudeClient(key, options.model, options.resolvedModelId, options.baseUrl);
             }
         }
         let instance;
-        const defaultHeaders = openRouter ? buildOpenRouterHeaders() : undefined;
-        const httpTimeoutMs = typeof options?.httpTimeoutMs === 'number' && Number.isFinite(options.httpTimeoutMs) && options.httpTimeoutMs > 0
+        const defaultHeaders = openRouter
+            ? buildOpenRouterHeaders()
+            : undefined;
+        const httpTimeoutMs = typeof options?.httpTimeoutMs === "number" &&
+            Number.isFinite(options.httpTimeoutMs) &&
+            options.httpTimeoutMs > 0
             ? options.httpTimeoutMs
             : 20 * 60 * 1000;
         if (options?.azure?.endpoint) {
@@ -82,13 +86,15 @@ export function createDefaultClientFactory() {
 }
 function buildOpenRouterHeaders() {
     const headers = {};
-    const referer = process.env.OPENROUTER_REFERER ?? process.env.OPENROUTER_HTTP_REFERER ?? 'https://github.com/steipete/oracle';
-    const title = process.env.OPENROUTER_TITLE ?? 'Oracle CLI';
+    const referer = process.env.OPENROUTER_REFERER ??
+        process.env.OPENROUTER_HTTP_REFERER ??
+        "https://github.com/steipete/oracle";
+    const title = process.env.OPENROUTER_TITLE ?? "Oracle CLI";
     if (referer) {
-        headers['HTTP-Referer'] = referer;
+        headers["HTTP-Referer"] = referer;
     }
     if (title) {
-        headers['X-Title'] = title;
+        headers["X-Title"] = title;
     }
     return headers;
 }
@@ -97,19 +103,19 @@ function loadCustomClientFactory() {
     if (!override) {
         return null;
     }
-    if (override === 'INLINE_TEST_FACTORY') {
+    if (override === "INLINE_TEST_FACTORY") {
         return () => ({
             responses: {
-                create: async () => ({ id: 'inline-test', status: 'completed' }),
+                create: async () => ({ id: "inline-test", status: "completed" }),
                 stream: async () => ({
                     [Symbol.asyncIterator]: () => ({
                         async next() {
                             return { done: true, value: undefined };
                         },
                     }),
-                    finalResponse: async () => ({ id: 'inline-test', status: 'completed' }),
+                    finalResponse: async () => ({ id: "inline-test", status: "completed" }),
                 }),
-                retrieve: async (id) => ({ id, status: 'completed' }),
+                retrieve: async (id) => ({ id, status: "completed" }),
             },
         });
     }
@@ -117,14 +123,14 @@ function loadCustomClientFactory() {
         const require = createRequire(import.meta.url);
         const resolved = path.isAbsolute(override) ? override : path.resolve(process.cwd(), override);
         const moduleExports = require(resolved);
-        const factory = typeof moduleExports === 'function'
+        const factory = typeof moduleExports === "function"
             ? moduleExports
-            : typeof moduleExports?.default === 'function'
+            : typeof moduleExports?.default === "function"
                 ? moduleExports.default
-                : typeof moduleExports?.createClientFactory === 'function'
+                : typeof moduleExports?.createClientFactory === "function"
                     ? moduleExports.createClientFactory
                     : null;
-        if (typeof factory === 'function') {
+        if (typeof factory === "function") {
             return factory;
         }
         console.warn(`Custom client factory at ${resolved} did not export a function.`);
@@ -140,14 +146,17 @@ function buildOpenRouterCompletionClient(instance) {
     const adaptRequest = (body) => {
         const messages = [];
         if (body.instructions) {
-            messages.push({ role: 'system', content: body.instructions });
+            messages.push({ role: "system", content: body.instructions });
         }
         for (const entry of body.input) {
             const textParts = entry.content
-                .map((c) => (c.type === 'input_text' ? c.text : ''))
+                .map((c) => (c.type === "input_text" ? c.text : ""))
                 .filter((t) => t)
-                .join('\n\n');
-            messages.push({ role: entry.role ?? 'user', content: textParts });
+                .join("\n\n");
+            messages.push({
+                role: entry.role ?? "user",
+                content: textParts,
+            });
         }
         const base = {
             model: body.model,
@@ -159,7 +168,7 @@ function buildOpenRouterCompletionClient(instance) {
         return { streaming, nonStreaming };
     };
     const adaptResponse = (response) => {
-        const text = response.choices?.[0]?.message?.content ?? '';
+        const text = response.choices?.[0]?.message?.content ?? "";
         const usage = {
             input_tokens: response.usage?.prompt_tokens ?? 0,
             output_tokens: response.usage?.completion_tokens ?? 0,
@@ -167,9 +176,9 @@ function buildOpenRouterCompletionClient(instance) {
         };
         return {
             id: response.id ?? `openrouter-${Date.now()}`,
-            status: 'completed',
+            status: "completed",
             output_text: [text],
-            output: [{ type: 'text', text }],
+            output: [{ type: "text", text }],
             usage,
         };
     };
@@ -177,15 +186,15 @@ function buildOpenRouterCompletionClient(instance) {
         const { streaming } = adaptRequest(body);
         let finalUsage;
         let finalId;
-        let aggregated = '';
+        let aggregated = "";
         async function* iterator() {
             const completion = await instance.chat.completions.create(streaming);
             for await (const chunk of completion) {
                 finalId = chunk.id ?? finalId;
-                const delta = chunk.choices?.[0]?.delta?.content ?? '';
+                const delta = chunk.choices?.[0]?.delta?.content ?? "";
                 if (delta) {
                     aggregated += delta;
-                    yield { type: 'chunk', delta };
+                    yield { type: "chunk", delta };
                 }
                 if (chunk.usage) {
                     finalUsage = chunk.usage;
@@ -200,11 +209,11 @@ function buildOpenRouterCompletionClient(instance) {
             async finalResponse() {
                 return adaptResponse({
                     id: finalId ?? `openrouter-${Date.now()}`,
-                    choices: [{ message: { role: 'assistant', content: aggregated } }],
+                    choices: [{ message: { role: "assistant", content: aggregated } }],
                     usage: finalUsage ?? { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
                     created: Math.floor(Date.now() / 1000),
-                    model: '',
-                    object: 'chat.completion',
+                    model: "",
+                    object: "chat.completion",
                 });
             },
         };
@@ -219,7 +228,7 @@ function buildOpenRouterCompletionClient(instance) {
             stream,
             create,
             retrieve: async () => {
-                throw new Error('retrieve is not supported for OpenRouter chat/completions fallback.');
+                throw new Error("retrieve is not supported for OpenRouter chat/completions fallback.");
             },
         },
     };

package/dist/src/oracle/config.js

@@ -1,15 +1,45 @@
-import { countTokens as countTokensGpt5 } from 'gpt-tokenizer/model/gpt-5';
-import { countTokens as countTokensGpt5Pro } from 'gpt-tokenizer/model/gpt-5-pro';
-import { countTokens as countTokensAnthropicRaw } from '@anthropic-ai/tokenizer';
-import { stringifyTokenizerInput } from './tokenStringifier.js';
-export const DEFAULT_MODEL = 'gpt-5.4-pro';
-export const PRO_MODELS = new Set(['gpt-5.4-pro', 'gpt-5.1-pro', 'gpt-5-pro', 'gpt-5.2-pro', 'claude-4.5-sonnet', 'claude-4.1-opus']);
+import { countTokens as countTokensGpt5 } from "gpt-tokenizer/model/gpt-5";
+import { countTokens as countTokensGpt5Pro } from "gpt-tokenizer/model/gpt-5-pro";
+import { countTokens as countTokensAnthropicRaw } from "@anthropic-ai/tokenizer";
+import { stringifyTokenizerInput } from "./tokenStringifier.js";
+export const DEFAULT_MODEL = "gpt-5.5-pro";
+export const PRO_MODELS = new Set([
+    "gpt-5.5-pro",
+    "gpt-5.4-pro",
+    "gpt-5.1-pro",
+    "gpt-5-pro",
+    "gpt-5.2-pro",
+    "claude-4.6-sonnet",
+    "claude-4.1-opus",
+]);
 const countTokensAnthropic = (input) => countTokensAnthropicRaw(stringifyTokenizerInput(input));
 export const MODEL_CONFIGS = {
-    'gpt-5.1-pro': {
-        model: 'gpt-5.1-pro',
-        apiModel: 'gpt-5.4-pro',
-        provider: 'openai',
+    "gpt-5.5-pro": {
+        model: "gpt-5.5-pro",
+        provider: "openai",
+        tokenizer: countTokensGpt5Pro,
+        inputLimit: 1_050_000,
+        pricing: {
+            inputPerToken: 30 / 1_000_000,
+            outputPerToken: 180 / 1_000_000,
+        },
+        reasoning: { effort: "xhigh" },
+    },
+    "gpt-5.5": {
+        model: "gpt-5.5",
+        provider: "openai",
+        tokenizer: countTokensGpt5,
+        inputLimit: 1_050_000,
+        pricing: {
+            inputPerToken: 5 / 1_000_000,
+            outputPerToken: 30 / 1_000_000,
+        },
+        reasoning: { effort: "xhigh" },
+    },
+    "gpt-5.1-pro": {
+        model: "gpt-5.1-pro",
+        apiModel: "gpt-5.5-pro",
+        provider: "openai",
         tokenizer: countTokensGpt5Pro,
         inputLimit: 196000,
         pricing: {
@@ -18,9 +48,9 @@ export const MODEL_CONFIGS = {
         },
         reasoning: null,
     },
-    'gpt-5-pro': {
-        model: 'gpt-5-pro',
-        provider: 'openai',
+    "gpt-5-pro": {
+        model: "gpt-5-pro",
+        provider: "openai",
         tokenizer: countTokensGpt5Pro,
         inputLimit: 196000,
         pricing: {
@@ -29,65 +59,65 @@ export const MODEL_CONFIGS = {
         },
         reasoning: null,
     },
-    'gpt-5.1': {
-        model: 'gpt-5.1',
-        provider: 'openai',
+    "gpt-5.1": {
+        model: "gpt-5.1",
+        provider: "openai",
         tokenizer: countTokensGpt5,
         inputLimit: 196000,
         pricing: {
             inputPerToken: 1.25 / 1_000_000,
             outputPerToken: 10 / 1_000_000,
         },
-        reasoning: { effort: 'high' },
+        reasoning: { effort: "high" },
     },
-    'gpt-5.1-codex': {
-        model: 'gpt-5.1-codex',
-        provider: 'openai',
+    "gpt-5.1-codex": {
+        model: "gpt-5.1-codex",
+        provider: "openai",
         tokenizer: countTokensGpt5,
         inputLimit: 196000,
         pricing: {
             inputPerToken: 1.25 / 1_000_000,
             outputPerToken: 10 / 1_000_000,
         },
-        reasoning: { effort: 'high' },
+        reasoning: { effort: "high" },
     },
-    'gpt-5.4': {
-        model: 'gpt-5.4',
-        provider: 'openai',
+    "gpt-5.4": {
+        model: "gpt-5.4",
+        provider: "openai",
         tokenizer: countTokensGpt5,
         inputLimit: 196000,
         pricing: {
             inputPerToken: 2.5 / 1_000_000,
             outputPerToken: 15 / 1_000_000,
         },
-        reasoning: { effort: 'xhigh' },
+        reasoning: { effort: "xhigh" },
     },
-    'gpt-5.4-pro': {
-        model: 'gpt-5.4-pro',
-        provider: 'openai',
+    "gpt-5.4-pro": {
+        model: "gpt-5.4-pro",
+        provider: "openai",
         tokenizer: countTokensGpt5Pro,
         inputLimit: 196000,
         pricing: {
             inputPerToken: 30 / 1_000_000,
             outputPerToken: 180 / 1_000_000,
         },
-        reasoning: { effort: 'xhigh' },
+        reasoning: { effort: "xhigh" },
     },
-    'gpt-5.2': {
-        model: 'gpt-5.2',
-        provider: 'openai',
+    "gpt-5.2": {
+        model: "gpt-5.2",
+        provider: "openai",
         tokenizer: countTokensGpt5,
         inputLimit: 196000,
         pricing: {
             inputPerToken: 1.75 / 1_000_000,
             outputPerToken: 14 / 1_000_000,
         },
-        reasoning: { effort: 'xhigh' },
+        reasoning: { effort: "xhigh" },
     },
-    'gpt-5.2-instant': {
-        model: 'gpt-5.2-instant',
-        apiModel: 'gpt-5.2-chat-latest',
-        provider: 'openai',
+    "gpt-5.2-instant": {
+        model: "gpt-5.2-instant",
+        apiModel: "gpt-5.2-chat-latest",
+        provider: "openai",
         tokenizer: countTokensGpt5,
         inputLimit: 196000,
         pricing: {
@@ -96,21 +126,21 @@ export const MODEL_CONFIGS = {
         },
         reasoning: null,
     },
-    'gpt-5.2-pro': {
-        model: 'gpt-5.2-pro',
-        apiModel: 'gpt-5.4-pro',
-        provider: 'openai',
+    "gpt-5.2-pro": {
+        model: "gpt-5.2-pro",
+        apiModel: "gpt-5.5-pro",
+        provider: "openai",
         tokenizer: countTokensGpt5Pro,
         inputLimit: 196000,
         pricing: {
             inputPerToken: 30 / 1_000_000,
             outputPerToken: 180 / 1_000_000,
         },
-        reasoning: { effort: 'xhigh' },
+        reasoning: { effort: "xhigh" },
     },
-    'gemini-3.1-pro': {
-        model: 'gemini-3.1-pro',
-        provider: 'google',
+    "gemini-3.1-pro": {
+        model: "gemini-3.1-pro",
+        provider: "google",
         tokenizer: countTokensGpt5Pro,
         inputLimit: 200000,
         pricing: {
@@ -121,9 +151,9 @@ export const MODEL_CONFIGS = {
         supportsBackground: false,
         supportsSearch: true,
     },
-    'gemini-3-pro': {
-        model: 'gemini-3-pro',
-        provider: 'google',
+    "gemini-3-pro": {
+        model: "gemini-3-pro",
+        provider: "google",
         tokenizer: countTokensGpt5Pro,
         inputLimit: 200000,
         pricing: {
@@ -134,10 +164,10 @@ export const MODEL_CONFIGS = {
         supportsBackground: false,
         supportsSearch: true,
     },
-    'claude-4.5-sonnet': {
-        model: 'claude-4.5-sonnet',
-        apiModel: 'claude-sonnet-4-5',
-        provider: 'anthropic',
+    "claude-4.6-sonnet": {
+        model: "claude-4.6-sonnet",
+        apiModel: "claude-sonnet-4-6",
+        provider: "anthropic",
         tokenizer: countTokensAnthropic,
         inputLimit: 200000,
         pricing: {
@@ -148,24 +178,24 @@ export const MODEL_CONFIGS = {
         supportsBackground: false,
         supportsSearch: false,
     },
-    'claude-4.1-opus': {
-        model: 'claude-4.1-opus',
-        apiModel: 'claude-opus-4-1',
-        provider: 'anthropic',
+    "claude-4.1-opus": {
+        model: "claude-4.1-opus",
+        apiModel: "claude-opus-4-1",
+        provider: "anthropic",
         tokenizer: countTokensAnthropic,
         inputLimit: 200000,
         pricing: {
             inputPerToken: 15 / 1_000_000,
             outputPerToken: 75 / 1_000_000,
         },
-        reasoning: { effort: 'high' },
+        reasoning: { effort: "high" },
         supportsBackground: false,
         supportsSearch: false,
     },
-    'grok-4.1': {
-        model: 'grok-4.1',
-        apiModel: 'grok-4-1-fast-reasoning',
-        provider: 'xai',
+    "grok-4.1": {
+        model: "grok-4.1",
+        apiModel: "grok-4-1-fast-reasoning",
+        provider: "xai",
         tokenizer: countTokensGpt5Pro,
         inputLimit: 2_000_000,
         pricing: {
@@ -175,11 +205,11 @@ export const MODEL_CONFIGS = {
         reasoning: null,
         supportsBackground: false,
         supportsSearch: true,
-        searchToolType: 'web_search',
+        searchToolType: "web_search",
     },
 };
 export const DEFAULT_SYSTEM_PROMPT = [
-    'You are Oracle, a focused one-shot problem solver.',
-    'Emphasize direct answers and cite any files referenced.',
-].join(' ');
-export const TOKENIZER_OPTIONS = { allowedSpecial: 'all' };
+    "You are Oracle, a focused one-shot problem solver.",
+    "Emphasize direct answers and cite any files referenced.",
+].join(" ");
+export const TOKENIZER_OPTIONS = { allowedSpecial: "all" };