copilot-api-plus 1.2.0 → 1.2.2

package/README.en.md

@@ -49,6 +49,7 @@ English | [简体中文](README.md)
 | 🛡️ **Network Resilience** | 120s connection timeout + exponential backoff retry (2s/5s/10s) |
 | ✂️ **Context Passthrough** | Full context passthrough to upstream API; clients (e.g. Claude Code) manage compression |
 | 🔍 **Smart Model Matching** | Handles model name format differences (date suffixes, dash/dot versions, etc.) |
+| 🧠 **Thinking Chain** | Automatically enables deep thinking (thinking/reasoning) for supported models, improving code quality |
 
 ---
 
@@ -573,8 +574,8 @@ For Anthropic endpoints (`/v1/messages`), `translateModelName` also handles lega
 Each API request outputs a log line with model name, status code, and duration:
 
 ```log
-[claude-opus-4-6] 13:13:39 <-- POST /v1/messages?beta=true
-[claude-opus-4-6] 13:13:59 --> POST /v1/messages?beta=true 200 20.1s
+[claude-opus-4-6 thinking] 13:13:39 <-- POST /v1/messages?beta=true
+[claude-opus-4-6 thinking] 13:13:59 --> POST /v1/messages?beta=true 200 20.1s
 ```
 
 ### Network Resilience

package/README.md

@@ -50,6 +50,7 @@
 | 🛡️ **网络弹性** | 120s 连接超时 + 指数退避重试（2s/5s/10s） |
 | ✂️ **上下文透传** | 全量透传上下文至上游 API，由客户端（如 Claude Code）自行管理压缩 |
 | 🔍 **智能模型匹配** | 自动处理模型名格式差异（日期后缀、dash/dot 版本号等） |
+| 🧠 **Thinking 思考链** | 自动为支持的模型启用深度思考（thinking/reasoning），提升代码质量 |
 
 ---
 
@@ -736,8 +737,8 @@ Anthropic 格式的模型名（如 `claude-opus-4-6`）和 Copilot 的模型列
 每次 API 请求会输出一行日志，包含模型名、状态码和耗时：
 
 ```log
-[claude-opus-4-6] 13:13:39 <-- POST /v1/messages?beta=true
-[claude-opus-4-6] 13:13:59 --> POST /v1/messages?beta=true 200 20.1s
+[claude-opus-4-6 thinking] 13:13:39 <-- POST /v1/messages?beta=true
+[claude-opus-4-6 thinking] 13:13:59 --> POST /v1/messages?beta=true 200 20.1s
 ```
 
 ### 网络弹性

package/dist/{get-user-WQHD68Sz.js → get-user-BT7hEyDN.js}

@@ -5,10 +5,10 @@ const standardHeaders = () => ({
 	"content-type": "application/json",
 	accept: "application/json"
 });
-const COPILOT_VERSION = "0.26.7";
+const COPILOT_VERSION = "0.38.2";
 const EDITOR_PLUGIN_VERSION = `copilot-chat/${COPILOT_VERSION}`;
 const USER_AGENT = `GitHubCopilotChat/${COPILOT_VERSION}`;
-const API_VERSION = "2025-05-01";
+const API_VERSION = "2025-10-01";
 const copilotBaseUrl = (source) => {
 	if (source.copilotApiEndpoint) return source.copilotApiEndpoint;
 	return source.accountType === "individual" ? "https://api.githubcopilot.com" : `https://api.${source.accountType}.githubcopilot.com`;
@@ -21,7 +21,9 @@ const copilotHeaders = (source, vision = false) => {
 		"editor-version": `vscode/${source.vsCodeVersion}`,
 		"editor-plugin-version": EDITOR_PLUGIN_VERSION,
 		"user-agent": USER_AGENT,
-		"openai-intent": "conversation-panel",
+		"openai-intent": "conversation-agent",
+		"x-interaction-type": "conversation-agent",
+		"x-agent-task-id": randomUUID(),
 		"x-github-api-version": API_VERSION,
 		"x-request-id": randomUUID(),
 		"x-vscode-user-agent-library-version": "electron-fetch"
@@ -72,4 +74,4 @@ async function getGitHubUser(githubToken) {
 //#endregion
 export { GITHUB_BASE_URL as a, copilotHeaders as c, GITHUB_APP_SCOPES as i, githubHeaders as l, state as n, GITHUB_CLIENT_ID as o, GITHUB_API_BASE_URL as r, copilotBaseUrl as s, getGitHubUser as t, standardHeaders as u };
 
-//# sourceMappingURL=get-user-WQHD68Sz.js.map
+//# sourceMappingURL=get-user-BT7hEyDN.js.map

package/dist/get-user-BT7hEyDN.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"get-user-BT7hEyDN.js","names":[],"sources":["../src/lib/api-config.ts","../src/lib/state.ts","../src/services/github/get-user.ts"],"sourcesContent":["import { randomUUID } from \"node:crypto\"\n\nexport const standardHeaders = () => ({\n  \"content-type\": \"application/json\",\n  accept: \"application/json\",\n})\n\nconst COPILOT_VERSION = \"0.38.2\"\nconst EDITOR_PLUGIN_VERSION = `copilot-chat/${COPILOT_VERSION}`\nconst USER_AGENT = `GitHubCopilotChat/${COPILOT_VERSION}`\n\n// Updated to match latest Zed implementation - 2025-10-01 returns Claude models\nconst API_VERSION = \"2025-10-01\"\n\n/**\n * Common interface for anything that can supply Copilot/GitHub credentials.\n *\n * Both `State` and `Account` satisfy this interface, so all header/URL\n * helpers can accept either without an explicit overload.\n */\nexport interface TokenSource {\n  copilotToken?: string\n  copilotApiEndpoint?: string\n  accountType: string\n  githubToken?: string\n  vsCodeVersion?: string\n}\n\n// Re-export constants used by other modules for building headers manually\nexport { API_VERSION, EDITOR_PLUGIN_VERSION, USER_AGENT }\n\n// Use the API endpoint from token response if available, otherwise fall back to default\nexport const copilotBaseUrl = (source: TokenSource) => {\n  if (source.copilotApiEndpoint) {\n    return source.copilotApiEndpoint\n  }\n  return source.accountType === \"individual\" ?\n      \"https://api.githubcopilot.com\"\n    : `https://api.${source.accountType}.githubcopilot.com`\n}\nexport const copilotHeaders = (\n  source: TokenSource,\n  vision: boolean = false,\n) => {\n  const headers: Record<string, string> = {\n    Authorization: `Bearer ${source.copilotToken}`,\n    \"content-type\": standardHeaders()[\"content-type\"],\n    \"copilot-integration-id\": \"vscode-chat\",\n    \"editor-version\": `vscode/${source.vsCodeVersion}`,\n    \"editor-plugin-version\": EDITOR_PLUGIN_VERSION,\n    \"user-agent\": USER_AGENT,\n    \"openai-intent\": \"conversation-agent\",\n    \"x-interaction-type\": \"conversation-agent\",\n    \"x-agent-task-id\": randomUUID(),\n    \"x-github-api-version\": API_VERSION,\n    \"x-request-id\": randomUUID(),\n    \"x-vscode-user-agent-library-version\": \"electron-fetch\",\n  }\n\n  if (vision) headers[\"copilot-vision-request\"] = \"true\"\n\n  return headers\n}\n\nexport const GITHUB_API_BASE_URL = \"https://api.github.com\"\nexport const githubHeaders = (source: TokenSource) => ({\n  ...standardHeaders(),\n  authorization: `token ${source.githubToken}`,\n  \"editor-version\": `vscode/${source.vsCodeVersion}`,\n  \"editor-plugin-version\": EDITOR_PLUGIN_VERSION,\n  \"user-agent\": USER_AGENT,\n  \"x-github-api-version\": API_VERSION,\n  \"x-vscode-user-agent-library-version\": \"electron-fetch\",\n})\n\nexport const GITHUB_BASE_URL = \"https://github.com\"\nexport const GITHUB_CLIENT_ID = \"Iv1.b507a08c87ecfe98\"\nexport const GITHUB_APP_SCOPES = [\"read:user\"].join(\" \")\n","import type { ModelsResponse } from \"~/services/copilot/get-models\"\n\nexport interface State {\n  githubToken?: string\n  copilotToken?: string\n  copilotApiEndpoint?: string // API endpoint returned by token response\n\n  accountType: string\n  models?: ModelsResponse\n  vsCodeVersion?: string\n\n  manualApprove: boolean\n  rateLimitWait: boolean\n  showToken: boolean\n\n  // Rate limiting configuration\n  rateLimitSeconds?: number\n  lastRequestTimestamp?: number\n\n  // API key authentication\n  apiKeys?: Array<string>\n\n  // Multi-account mode\n  multiAccountEnabled: boolean\n\n  // Selected models (from --claude-code setup)\n  selectedModel?: string\n  selectedSmallModel?: string\n}\n\nexport const state: State = {\n  accountType: \"individual\",\n  manualApprove: false,\n  rateLimitWait: false,\n  showToken: false,\n  multiAccountEnabled: false,\n}\n","import { GITHUB_API_BASE_URL, standardHeaders } from \"~/lib/api-config\"\nimport { HTTPError } from \"~/lib/error\"\nimport { state } from \"~/lib/state\"\n\n/**\n * Fetch the GitHub user profile.\n *\n * @param githubToken  Optional explicit token.  When omitted, falls back to\n *                     the global `state.githubToken`.  Prefer passing a token\n *                     explicitly to avoid race conditions in multi-account mode.\n */\nexport async function getGitHubUser(githubToken?: string) {\n  const token = githubToken ?? state.githubToken\n  const response = await fetch(`${GITHUB_API_BASE_URL}/user`, {\n    headers: {\n      authorization: `token ${token}`,\n      ...standardHeaders(),\n    },\n  })\n\n  if (!response.ok) throw new HTTPError(\"Failed to get GitHub user\", response)\n\n  return (await response.json()) as GithubUserResponse\n}\n\n// Trimmed for the sake of simplicity\ninterface GithubUserResponse {\n  login: string\n}\n"],"mappings":";;;AAEA,MAAa,yBAAyB;CACpC,gBAAgB;CAChB,QAAQ;CACT;AAED,MAAM,kBAAkB;AACxB,MAAM,wBAAwB,gBAAgB;AAC9C,MAAM,aAAa,qBAAqB;AAGxC,MAAM,cAAc;AAoBpB,MAAa,kBAAkB,WAAwB;AACrD,KAAI,OAAO,mBACT,QAAO,OAAO;AAEhB,QAAO,OAAO,gBAAgB,eAC1B,kCACA,eAAe,OAAO,YAAY;;AAExC,MAAa,kBACX,QACA,SAAkB,UACf;CACH,MAAM,UAAkC;EACtC,eAAe,UAAU,OAAO;EAChC,gBAAgB,iBAAiB,CAAC;EAClC,0BAA0B;EAC1B,kBAAkB,UAAU,OAAO;EACnC,yBAAyB;EACzB,cAAc;EACd,iBAAiB;EACjB,sBAAsB;EACtB,mBAAmB,YAAY;EAC/B,wBAAwB;EACxB,gBAAgB,YAAY;EAC5B,uCAAuC;EACxC;AAED,KAAI,OAAQ,SAAQ,4BAA4B;AAEhD,QAAO;;AAGT,MAAa,sBAAsB;AACnC,MAAa,iBAAiB,YAAyB;CACrD,GAAG,iBAAiB;CACpB,eAAe,SAAS,OAAO;CAC/B,kBAAkB,UAAU,OAAO;CACnC,yBAAyB;CACzB,cAAc;CACd,wBAAwB;CACxB,uCAAuC;CACxC;AAED,MAAa,kBAAkB;AAC/B,MAAa,mBAAmB;AAChC,MAAa,oBAAoB,CAAC,YAAY,CAAC,KAAK,IAAI;;;AC/CxD,MAAa,QAAe;CAC1B,aAAa;CACb,eAAe;CACf,eAAe;CACf,WAAW;CACX,qBAAqB;CACtB;;;;;;;;;;ACzBD,eAAsB,cAAc,aAAsB;CACxD,MAAM,QAAQ,eAAe,MAAM;CACnC,MAAM,WAAW,MAAM,MAAM,GAAG,oBAAoB,QAAQ,EAC1D,SAAS;EACP,eAAe,SAAS;EACxB,GAAG,iBAAiB;EACrB,EACF,CAAC;AAEF,KAAI,CAAC,SAAS,GAAI,OAAM,IAAI,UAAU,6BAA6B,SAAS;AAE5E,QAAQ,MAAM,SAAS,MAAM"}

package/dist/get-user-BTN_-eOk.js

@@ -0,0 +1,3 @@
+import "./error-CvwUkoEo.js";
+import { t as getGitHubUser } from "./get-user-BT7hEyDN.js";
+export { getGitHubUser };

package/dist/main.js

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import { n as forwardError, t as HTTPError } from "./error-CvwUkoEo.js";
-import { a as stopCopilotTokenRefresh, c as cacheModels, d as isNullish, f as sleep, g as ensurePaths, h as PATHS, i as setupGitHubToken, l as cacheVSCodeVersion, m as getCopilotUsage, n as refreshCopilotToken, o as pollAccessToken, p as accountManager, r as setupCopilotToken, s as getDeviceCode, t as clearGithubToken, u as findModel } from "./token-BMHteyQY.js";
-import { a as GITHUB_BASE_URL, c as copilotHeaders, n as state, o as GITHUB_CLIENT_ID, s as copilotBaseUrl, u as standardHeaders } from "./get-user-WQHD68Sz.js";
+import { a as stopCopilotTokenRefresh, c as cacheModels, d as isNullish, f as sleep, g as ensurePaths, h as PATHS, i as setupGitHubToken, l as cacheVSCodeVersion, m as getCopilotUsage, n as refreshCopilotToken, o as pollAccessToken, p as accountManager, r as setupCopilotToken, s as getDeviceCode, t as clearGithubToken, u as findModel } from "./token-CsSZOO_Z.js";
+import { a as GITHUB_BASE_URL, c as copilotHeaders, n as state, o as GITHUB_CLIENT_ID, s as copilotBaseUrl, u as standardHeaders } from "./get-user-BT7hEyDN.js";
 import { createRequire } from "node:module";
 import { defineCommand, runMain } from "citty";
 import consola from "consola";
@@ -1320,29 +1320,47 @@ function formatDuration(ms) {
 	return `${(ms / 1e3).toFixed(1)}s`;
 }
 /**
-* Extract model name from request body
+* Extract model name and thinking status from request body.
+*
+* Thinking is considered active when:
+*  - The client explicitly passes reasoning_effort / thinking_budget / thinking
+*  - OR the backend will auto-inject thinking (default for all supported models)
+*
+* Since the proxy injects thinking for every model except those known to reject
+* it, we always mark thinking=true here. The `create-chat-completions` layer
+* will log the precise injection path separately.
 */
-async function extractModel(c) {
+async function extractModelInfo(c) {
 	try {
-		return (await c.req.raw.clone().json()).model;
+		const body = await c.req.raw.clone().json();
+		const clientThinking = Boolean(body.reasoning_effort) || Boolean(body.thinking_budget) || body.thinking?.type === "enabled";
+		return {
+			model: body.model,
+			thinking: clientThinking || true
+		};
 	} catch {
-		return;
+		return {};
 	}
 }
 /**
-* Custom logger middleware that shows model name before timestamp
+* Custom logger middleware that shows model name + thinking before timestamp
 *
 * Output format:
-* [model] HH:mm:ss <-- METHOD /path
-* [model] HH:mm:ss --> METHOD /path STATUS DURATION
+* [model thinking] HH:mm:ss <-- METHOD /path
+* [model thinking] HH:mm:ss --> METHOD /path STATUS DURATION
 */
 function modelLogger() {
 	return async (c, next) => {
 		const method = c.req.method;
 		const fullPath = `${c.req.path}${c.req.raw.url.includes("?") ? `?${c.req.raw.url.split("?")[1]}` : ""}`;
 		let model;
-		if (method === "POST" && c.req.header("content-type")?.includes("json")) model = await extractModel(c);
-		const modelPrefix = model ? `[${model}] ` : "";
+		let thinking = false;
+		if (method === "POST" && c.req.header("content-type")?.includes("json")) {
+			const info = await extractModelInfo(c);
+			model = info.model;
+			thinking = info.thinking ?? false;
+		}
+		const modelPrefix = model ? `[${model}${thinking ? " thinking" : ""}] ` : "";
 		const startTime = getTime();
 		console.log(`${modelPrefix}${startTime} <-- ${method} ${fullPath}`);
 		const start = Date.now();
@@ -1712,6 +1730,57 @@ async function* wrapGeneratorWithRelease(gen, releaseSlot) {
 		releaseSlot();
 	}
 }
+/**
+* Models that are known NOT to support the `reasoning_effort` parameter.
+* Populated at runtime: the first time a model returns 400 with
+* "Unrecognized request argument", it is added here and all future
+* requests to that model skip the injection automatically.
+*/
+const reasoningUnsupportedModels = /* @__PURE__ */ new Set();
+/**
+* Compute an appropriate thinking_budget from model capabilities.
+* Returns undefined if the model does not support thinking.
+*/
+function getThinkingBudget(model) {
+	if (!model) return void 0;
+	const { supports, limits } = model.capabilities;
+	const maxBudget = supports.max_thinking_budget;
+	if (!maxBudget || maxBudget <= 0) return void 0;
+	const maxOutput = limits.max_output_tokens ?? 0;
+	const upperBound = Math.min(maxBudget, Math.max(maxOutput - 1, 0));
+	const lowerBound = supports.min_thinking_budget ?? 1024;
+	return Math.max(upperBound, lowerBound);
+}
+/**
+* Inject thinking parameters into the payload based on model capabilities.
+*
+* Strategy (in priority order):
+*   1. If the client already set reasoning_effort or thinking_budget → keep as-is
+*   2. If model capabilities declare max_thinking_budget → inject thinking_budget
+*   3. Otherwise → inject reasoning_effort="high" (works on claude-*-4.6)
+*
+* The fallback to reasoning_effort ensures thinking works even when the
+* /models endpoint doesn't expose thinking budget fields.
+*/
+function injectThinking(payload, resolvedModel) {
+	if (payload.reasoning_effort || payload.thinking_budget) return payload;
+	const budget = getThinkingBudget(findModel(resolvedModel));
+	if (budget) return {
+		...payload,
+		thinking_budget: budget
+	};
+	if (!reasoningUnsupportedModels.has(resolvedModel)) return {
+		...payload,
+		reasoning_effort: "high"
+	};
+	return payload;
+}
+function logThinkingInjection(original, injected, resolvedModel) {
+	if (original.reasoning_effort || original.thinking_budget) consola.debug(`Thinking: client-specified (reasoning_effort=${original.reasoning_effort ?? "none"} / thinking_budget=${original.thinking_budget ?? "none"})`);
+	else if (injected.thinking_budget && injected.thinking_budget !== original.thinking_budget) consola.debug(`Thinking: injected thinking_budget=${injected.thinking_budget} for "${resolvedModel}"`);
+	else if (injected.reasoning_effort === "high") consola.debug(`Thinking: injected reasoning_effort=high for "${resolvedModel}"`);
+	else if (reasoningUnsupportedModels.has(resolvedModel)) consola.debug(`Thinking: skipped — "${resolvedModel}" does not support reasoning`);
+}
 const createChatCompletions = async (payload) => {
 	const resolvedModel = modelRouter.resolveModel(payload.model);
 	const routedPayload = resolvedModel !== payload.model ? {
@@ -1719,22 +1788,46 @@ const createChatCompletions = async (payload) => {
 		model: resolvedModel
 	} : payload;
 	if (resolvedModel !== payload.model) consola.debug(`Model routed: ${payload.model} → ${resolvedModel}`);
-	const supportsReasoning = /^claude-.+-4\.6/.test(resolvedModel);
-	const thinkingPayload = {
-		...routedPayload,
-		...!routedPayload.reasoning_effort && supportsReasoning && { reasoning_effort: "high" }
-	};
+	const thinkingPayload = injectThinking(routedPayload, resolvedModel);
+	const wasInjected = thinkingPayload.reasoning_effort !== routedPayload.reasoning_effort || thinkingPayload.thinking_budget !== routedPayload.thinking_budget;
+	logThinkingInjection(routedPayload, thinkingPayload, resolvedModel);
 	const releaseSlot = await modelRouter.acquireSlot(resolvedModel);
 	try {
-		const result = state.multiAccountEnabled && accountManager.hasAccounts() ? await createWithMultiAccount(thinkingPayload) : await createWithSingleAccount(thinkingPayload);
+		const result = await dispatchRequest(thinkingPayload);
 		if (Symbol.asyncIterator in result) return wrapGeneratorWithRelease(result, releaseSlot);
 		releaseSlot();
 		return result;
 	} catch (error) {
+		if (wasInjected && error instanceof HTTPError && error.response.status === 400 && error.message.includes("Unrecognized request argument")) {
+			reasoningUnsupportedModels.add(resolvedModel);
+			consola.info(`Model "${resolvedModel}" does not support reasoning_effort — disabled for future requests`);
+			return retryWithoutReasoning(routedPayload, releaseSlot);
+		}
 		releaseSlot();
 		throw error;
 	}
 };
+/**
+* Retry a request without reasoning_effort after the model rejected it.
+* Handles slot release for both streaming and non-streaming responses.
+*/
+async function retryWithoutReasoning(payload, releaseSlot) {
+	try {
+		const result = await dispatchRequest(payload);
+		if (Symbol.asyncIterator in result) return wrapGeneratorWithRelease(result, releaseSlot);
+		releaseSlot();
+		return result;
+	} catch (retryError) {
+		releaseSlot();
+		throw retryError;
+	}
+}
+/**
+* Dispatch request to either single-account or multi-account path.
+*/
+function dispatchRequest(payload) {
+	return state.multiAccountEnabled && accountManager.hasAccounts() ? createWithMultiAccount(payload) : createWithSingleAccount(payload);
+}
 async function createWithSingleAccount(payload) {
 	if (!state.copilotToken) throw new Error("Copilot token not found");
 	const enableVision = payload.messages.some((x) => typeof x.content !== "string" && x.content?.some((x) => x.type === "image_url"));
@@ -2246,7 +2339,8 @@ function translateToOpenAI(payload) {
 		tools: translateAnthropicToolsToOpenAI(payload.tools),
 		tool_choice: translateAnthropicToolChoiceToOpenAI(payload.tool_choice),
 		...payload.thinking && { thinking: payload.thinking },
-		...payload.thinking?.type === "enabled" && { reasoning_effort: "high" }
+		...payload.thinking?.type === "enabled" && { reasoning_effort: "high" },
+		...payload.thinking?.budget_tokens && { thinking_budget: payload.thinking.budget_tokens }
 	};
 }
 function translateModelName(model) {
@@ -2911,7 +3005,7 @@ async function validateGitHubToken(token) {
 	state.githubToken = token;
 	consola.info("Using provided GitHub token");
 	try {
-		const { getGitHubUser } = await import("./get-user-D5-qoD5J.js");
+		const { getGitHubUser } = await import("./get-user-BTN_-eOk.js");
 		const user = await getGitHubUser();
 		consola.info(`Logged in as ${user.login}`);
 	} catch (error) {
@@ -2960,7 +3054,7 @@ async function runServer(options) {
 		const { HTTPError } = await import("./error-DLqcVQL_.js");
 		if (error instanceof HTTPError && error.response.status === 401) {
 			consola.error("Failed to get Copilot token - GitHub token may be invalid or Copilot access revoked");
-			const { clearGithubToken } = await import("./token-Brc5l1C6.js");
+			const { clearGithubToken } = await import("./token-BsOH3gXZ.js");
 			await clearGithubToken();
 			consola.info("Please restart to re-authenticate");
 		}