dsh-llm-adapters 1.2.1 → 1.2.2

package/lib/index.js

@@ -788,6 +788,7 @@ function codeartsRecord(value) {
 		...text$5(raw.codeVerifier) === void 0 ? {} : { codeVerifier: text$5(raw.codeVerifier) },
 		...key === void 0 ? {} : { dpopPrivateJwk: key },
 		...text$5(raw.userName) === void 0 ? {} : { userName: text$5(raw.userName) },
+		...text$5(raw.otSessionId) === void 0 ? {} : { otSessionId: text$5(raw.otSessionId) },
 		...noBenefit.length === 0 ? {} : { noBenefit }
 	};
 }
@@ -1425,6 +1426,18 @@ function readUserProfile(token) {
 		...domainId === void 0 ? {} : { domainId }
 	};
 }
+/**
+* The session id this account should chat under, derived from who the account is.
+*
+* 32 lowercase hex, the shape the IDE's own kernel sends. Nothing but stability
+* matters here — the gateway uses the value as a key, not as an identity — so it
+* is hashed from the principal id rather than kept in a file.
+*/
+function otSessionIdOf(profile) {
+	const seed = profile?.userId ?? profile?.userName;
+	if (seed === void 0) return void 0;
+	return createHash("sha256").update(`codearts-ot-session:${seed}`).digest("hex").slice(0, 32);
+}
 /** The portal URL the browser opens; the callback is on loopback. */
 function buildAuthorizeUrl(port, pkce, ticketId) {
 	return `${CODEARTS_PORTAL_AUTHORIZE}?theme=2&locale=zh-cn&uri_scheme=${CODEARTS_CLIENT_ID}&client_id=${CODEARTS_CLIENT_ID}&port=${port}&code_challenge=${pkce.codeChallenge}&code_challenge_method=SHA-256&ticket_id=${ticketId}&plugin-name=snap_AIIDE&plugin-version=5.2.0`;
@@ -1475,25 +1488,30 @@ async function exchangeRefreshToken(refreshToken, codeVerifier, keyPair, fetcher
 		grant_type: "refresh_token",
 		refresh_token: refreshToken
 	}, keyPair, fetcher), codeVerifier, keyPair);
-	const userName = credential.userName ?? readUserProfile(refreshToken)?.userName;
+	const previous = readUserProfile(refreshToken);
+	const userName = credential.userName ?? previous?.userName;
+	const otSessionId = credential.otSessionId ?? otSessionIdOf(previous);
 	return {
 		...credential,
 		refreshToken: credential.refreshToken ?? refreshToken,
-		...userName === void 0 ? {} : { userName }
+		...userName === void 0 ? {} : { userName },
+		...otSessionId === void 0 ? {} : { otSessionId }
 	};
 }
 function credentialFrom(token, codeVerifier, keyPair) {
 	const credentials = token.credentials ?? {};
 	const expiration = credentials.expiration;
 	const expiresAt = expiration === void 0 ? void 0 : Date.parse(expiration);
-	const userName = readUserProfile(token.refresh_token)?.userName;
+	const profile = readUserProfile(token.refresh_token);
+	const otSessionId = otSessionIdOf(profile);
 	return {
 		accessKeyId: credentials.access_key_id ?? "",
 		secretAccessKey: credentials.secret_access_key ?? "",
 		securityToken: credentials.security_token ?? "",
 		...expiresAt === void 0 || Number.isNaN(expiresAt) ? {} : { expiresAt },
 		...token.refresh_token === void 0 ? {} : { refreshToken: token.refresh_token },
-		...userName === void 0 ? {} : { userName },
+		...profile?.userName === void 0 ? {} : { userName: profile.userName },
+		...otSessionId === void 0 ? {} : { otSessionId },
 		codeVerifier,
 		dpopPrivateJwk: keyPair.privateKeyJwk
 	};
@@ -2224,7 +2242,8 @@ function decodeJwtFacts(token) {
 			...typeof record.name === "string" ? { name: record.name } : {},
 			...typeof record.phone === "string" ? { phone: record.phone } : {},
 			...typeof record.owner_type === "string" ? { ownerType: record.owner_type } : {},
-			...typeof record.sid === "string" ? { sessionId: record.sid } : {}
+			...typeof record.sid === "string" ? { sessionId: record.sid } : {},
+			...typeof record.sub === "string" ? { subject: record.sub } : {}
 		};
 	} catch {
 		return {};
@@ -2557,7 +2576,8 @@ var CodeArtsAuth = class {
 					securityToken: next.securityToken,
 					...next.expiresAt === void 0 ? {} : { expiresAt: next.expiresAt },
 					...next.refreshToken === void 0 ? {} : { refreshToken: next.refreshToken },
-					...next.userName === void 0 ? {} : { userName: next.userName }
+					...next.userName === void 0 ? {} : { userName: next.userName },
+					...next.otSessionId === void 0 ? {} : { otSessionId: next.otSessionId }
 				});
 				this.lastError = void 0;
 				return next;
@@ -2577,6 +2597,7 @@ var CodeArtsAuth = class {
 	}
 };
 function credentialOf(record) {
+	const otSessionId = record.otSessionId ?? otSessionIdOf(readUserProfile(record.refreshToken));
 	return {
 		accessKeyId: record.accessKeyId,
 		secretAccessKey: record.secretAccessKey,
@@ -2584,7 +2605,8 @@ function credentialOf(record) {
 		...record.expiresAt === void 0 ? {} : { expiresAt: record.expiresAt },
 		...record.refreshToken === void 0 ? {} : { refreshToken: record.refreshToken },
 		...record.codeVerifier === void 0 ? {} : { codeVerifier: record.codeVerifier },
-		...record.dpopPrivateJwk === void 0 ? {} : { dpopPrivateJwk: record.dpopPrivateJwk }
+		...record.dpopPrivateJwk === void 0 ? {} : { dpopPrivateJwk: record.dpopPrivateJwk },
+		...otSessionId === void 0 ? {} : { otSessionId }
 	};
 }
 // src/config.ts
@@ -2631,8 +2653,7 @@ const Config$1 = z.object({
 	defaultMaxTokens: z.number().step(1).min(1).default(32e3).description("模型目录未声明输出上限时使用的输出上限。").volatile(),
 	modelCacheTtlSeconds: z.number().step(1).min(10).max(86400).default(120).description("模型列表缓存时长（秒）。调小可更快看到账号侧变化。").volatile(),
 	catalogFetchTimeoutSeconds: z.number().step(1).min(1).max(300).default(20).description("读取模型目录、额度等请求的时限（秒）。").volatile(),
-	openRouterSyncMinutes: z.number().step(1).min(0).max(43200).default(60).description("OpenRouter 模型元数据的自动同步间隔（分钟）。0 表示关闭自动同步；缓存保存在本机，连不上时继续使用上次成功同步的数据。").volatile(),
-	raccoonCheckinMinutes: z.number().step(1).min(0).max(1440).default(30).description("小浣熊每日积分自动监测间隔（分钟）：到点核对积分流水，未到账时按网页方式尝试触发。0 表示关闭自动监测；「打开网页签到」入口始终保留。").volatile()
+	openRouterSyncMinutes: z.number().step(1).min(0).max(43200).default(60).description("OpenRouter 模型元数据的自动同步间隔（分钟）。0 表示关闭自动同步；缓存保存在本机，连不上时继续使用上次成功同步的数据。").volatile()
 });
 function liveConfig(config) {
 	const source = config ?? {};
@@ -2673,8 +2694,7 @@ function liveConfig(config) {
 		defaultMaxTokens: () => positive(field("defaultMaxTokens"), 32e3),
 		modelCacheTtlMs: () => positive(field("modelCacheTtlSeconds"), 120) * 1e3,
 		catalogFetchTimeoutMs: () => positive(field("catalogFetchTimeoutSeconds"), 20) * 1e3,
-		openRouterSyncMinutes: () => nonNegative(field("openRouterSyncMinutes"), 60),
-		raccoonCheckinMinutes: () => nonNegative(field("raccoonCheckinMinutes"), 30)
+		openRouterSyncMinutes: () => nonNegative(field("openRouterSyncMinutes"), 60)
 	};
 }
 function resolveConfig(config) {
@@ -2712,8 +2732,7 @@ function resolveConfig(config) {
 		defaultMaxTokens: live.defaultMaxTokens(),
 		modelCacheTtlMs: live.modelCacheTtlMs(),
 		catalogFetchTimeoutMs: live.catalogFetchTimeoutMs(),
-		openRouterSyncMinutes: live.openRouterSyncMinutes(),
-		raccoonCheckinMinutes: live.raccoonCheckinMinutes()
+		openRouterSyncMinutes: live.openRouterSyncMinutes()
 	};
 }
 function text$4(value) {
@@ -3056,117 +3075,6 @@ function headersForFetch(signed) {
 	}
 	return out;
 }
-// src/codearts-window.ts
-/**
-* The trial plan's chat budget.
-*
-* The gateway opens one "session" per chat request and allows **three per
-* rolling minute** for the whole account. The window is shared by every model,
-* it is not keyed by `session-id` (re-sending the same id does not buy a slot),
-* and a request rejected with `InferHub.4004.200` spends a slot just like a
-* successful one. Measured behaviour, three slots then a lockout:
-*
-* ```
-* t+0s   ok ok ok cap cap cap     burst of six
-* t+10s  cap                      …
-* t+51s  cap
-* t+61s  ok ok ok                 the burst ages out of the window
-* ```
-*
-* So a turn cannot be made to "go faster" by retrying: the only way a request
-* stops failing is to spend the budget deliberately — wait for the oldest start
-* to age out instead of firing into a wall. This class owns that bookkeeping and
-* is shared by every provider instance in the process, because a hot reload
-* must not reset the count the gateway still remembers.
-*
-* @module dsh-llm-adapters/codearts-window
-*/
-/** Sessions the trial plan allows inside one window. */
-const CODEARTS_SESSION_LIMIT = 3;
-/** How long a start keeps its slot. Measured: the burst frees at ~60 s. */
-const CODEARTS_SESSION_WINDOW_MS = 6e4;
-/** A little slack so a request never lands a few milliseconds early. */
-const MARGIN_MS = 1500;
-function sleep$2(ms, signal) {
-	return new Promise((resolve, reject) => {
-		if (signal?.aborted === true) {
-			reject(signal.reason instanceof Error ? signal.reason : /* @__PURE__ */ new Error("aborted"));
-			return;
-		}
-		const onAbort = () => {
-			clearTimeout(timer);
-			reject(signal?.reason instanceof Error ? signal.reason : /* @__PURE__ */ new Error("aborted"));
-		};
-		const timer = setTimeout(() => {
-			signal?.removeEventListener("abort", onAbort);
-			resolve();
-		}, ms);
-		signal?.addEventListener("abort", onAbort, { once: true });
-	});
-}
-var CodeArtsSessionWindow = class {
-	limit;
-	windowMs;
-	now;
-	starts = [];
-	/** Serialises waiters so two callers cannot both take the last slot. */
-	tail = Promise.resolve();
-	constructor(limit = 3, windowMs = CODEARTS_SESSION_WINDOW_MS, now = Date.now) {
-		this.limit = limit;
-		this.windowMs = windowMs;
-		this.now = now;
-	}
-	/** Starts recorded inside the current window. */
-	get size() {
-		this.trim();
-		return this.starts.length;
-	}
-	/** Milliseconds until another start fits; `0` when the window has room. */
-	waitMs() {
-		this.trim();
-		if (this.starts.length < this.limit) return 0;
-		const oldest = this.starts[0] ?? 0;
-		return Math.max(0, this.windowMs - (this.now() - oldest));
-	}
-	/**
-	* Claim a slot, waiting for the window to roll over when it is full.
-	*
-	* The slot is recorded on return, not on completion: the gateway counts the
-	* start, and holds it for a fixed time whether the answer was short or long.
-	*/
-	async open(signal) {
-		const run = this.tail.then(async () => {
-			for (;;) {
-				const wait = this.waitMs();
-				if (wait === 0) {
-					const start = this.now();
-					this.starts.push(start);
-					return { release: () => {
-						const index = this.starts.indexOf(start);
-						if (index >= 0) this.starts.splice(index, 1);
-					} };
-				}
-				await sleep$2(wait + MARGIN_MS, signal);
-			}
-		});
-		this.tail = run.then(() => void 0, () => void 0);
-		return await run;
-	}
-	/** Forget every start (tests, and a fresh login on another account). */
-	reset() {
-		this.starts.length = 0;
-	}
-	trim() {
-		const now = this.now();
-		while (this.starts.length > 0 && now - (this.starts[0] ?? 0) >= this.windowMs) this.starts.shift();
-	}
-};
-let shared;
-/** The process-wide window; a plugin reload reuses it on purpose. */
-function sharedSessionWindow() {
-	shared ??= new CodeArtsSessionWindow();
-	return shared;
-}
 // src/openrouter.ts
 /**
 * Live capability lookup from OpenRouter's public model catalog.
@@ -3692,18 +3600,20 @@ const sleep$1 = (ms, signal) => new Promise((resolve, reject) => {
 	}, ms);
 	signal?.addEventListener("abort", onAbort, { once: true });
 });
-/** Retries for a gateway that is busy for reasons the window cannot predict. */
+/** Retries for a gateway that is busy for reasons the caller cannot predict. */
 const QUEUE_ATTEMPTS = 3;
-/** Minimum pause before retrying a rejected call; the window adds its own. */
+/** Pause before retrying a rejected call. */
 const QUEUE_BACKOFF_MS = 2e4;
 /**
 * The message a user sees once waiting is hopeless.
 *
-* The raw gateway text (`TM.00001041`) reads like a bug; the limit is the
-* trial plan's, and the only fix on this side is to say so.
+* The raw gateway text (`TM.00001041`) reads like a bug. The three concurrent
+* sessions are the trial plan's, and one plugin chatting under a single session
+* id is not meant to reach them: the slots are being held by something else on
+* the same account.
 */
 function queueExhausted() {
-	return new ProviderHttpError("CodeArts 体验版每分钟只允许 3 个会话，当前已被占满。请稍等片刻再发，或关闭其它 CodeArts 会话。", 429);
+	return new ProviderHttpError("CodeArts 这个账号的 3 个并发会话已被占满（通常是码道 IDE 或另一台设备），请稍后再发，或关闭其它 CodeArts 会话。", 429);
 }
 /** Image transcripts live beside accounts.json, so a restart does not re-bill. */
 const IMAGE_CACHE_FILE = "codearts-image-cache.json";
@@ -3736,7 +3646,6 @@ var CodeArtsProvider = class {
 	/** Models that answered `InferHub.4004.200`: no allowance pack to bill. */
 	noBenefit = /* @__PURE__ */ new Set();
 	noBenefitLoaded = false;
-	window;
 	openRouter;
 	sessionId = randomUUID();
 	/**
@@ -3751,7 +3660,6 @@ var CodeArtsProvider = class {
 	imageCacheWrite = Promise.resolve();
 	constructor(options) {
 		this.options = options;
-		this.window = options.window ?? sharedSessionWindow();
 		this.openRouter = options.openRouter ?? new OpenRouterCatalog(options.fetchImpl ?? fetch);
 		this.catalog = new ModelCatalog(() => this.loadModels(), options.ttlMs);
 		this.login = new CodeArtsLogin({
@@ -3760,7 +3668,6 @@ var CodeArtsProvider = class {
 			...options.logger === void 0 ? {} : { logger: options.logger },
 			save: async (account) => {
 				await options.store.saveCodearts(account);
-				this.window.reset();
 				this.noBenefit.clear();
 				this.invalidateModels();
 				this.invalidateQuota();
@@ -4089,7 +3996,7 @@ var CodeArtsProvider = class {
 			}
 		}
 	}
-	/** Ask the vision model to describe one image, within the same chat budget. */
+	/** Ask the vision model to describe one image; another turn on the same session. */
 	async describeImage(dataUrl, signal) {
 		const payload = JSON.stringify({
 			model: CODEARTS_VISION_MODEL,
@@ -4110,7 +4017,6 @@ var CodeArtsProvider = class {
 		let authRefreshed = false;
 		let queueAttempts = 0;
 		for (;;) {
-			const slot = await this.window.open(signal);
 			let response;
 			try {
 				response = await this.send(chatId, payload, CODEARTS_VISION_MODEL, false, signal);
@@ -4121,10 +4027,9 @@ var CodeArtsProvider = class {
 					continue;
 				}
 				if (error instanceof CodeArtsRetry && error.kind === "queue") {
-					slot.release();
 					if (queueAttempts < QUEUE_ATTEMPTS) {
 						queueAttempts++;
-						await sleep$1(Math.max(this.window.waitMs(), QUEUE_BACKOFF_MS), signal);
+						await sleep$1(QUEUE_BACKOFF_MS, signal);
 						continue;
 					}
 					throw queueExhausted();
@@ -4145,10 +4050,9 @@ var CodeArtsProvider = class {
 				})) if (chunk.type === "text-delta") transcript += chunk.text;
 			} catch (error) {
 				if (error instanceof CodeArtsRetry && error.kind === "queue") {
-					slot.release();
 					if (queueAttempts < QUEUE_ATTEMPTS) {
 						queueAttempts++;
-						await sleep$1(Math.max(this.window.waitMs(), QUEUE_BACKOFF_MS), signal);
+						await sleep$1(QUEUE_BACKOFF_MS, signal);
 						continue;
 					}
 					throw queueExhausted();
@@ -4187,7 +4091,6 @@ var CodeArtsProvider = class {
 		let benefitDropped = false;
 		let queueAttempts = 0;
 		for (;;) {
-			const slot = await this.window.open(options.signal);
 			let response;
 			try {
 				response = await this.send(chatId, payload, model, benefitDropped, options.signal);
@@ -4210,8 +4113,7 @@ var CodeArtsProvider = class {
 					}
 					if (error.kind === "queue" && queueAttempts < QUEUE_ATTEMPTS) {
 						queueAttempts++;
-						slot.release();
-						await sleep$1(Math.max(this.window.waitMs(), QUEUE_BACKOFF_MS), options.signal);
+						await sleep$1(QUEUE_BACKOFF_MS, options.signal);
 						continue;
 					}
 					if (error.kind === "queue") throw queueExhausted();
@@ -4238,8 +4140,7 @@ var CodeArtsProvider = class {
 			} catch (error) {
 				if (!produced && error instanceof CodeArtsRetry && error.kind === "queue" && queueAttempts < QUEUE_ATTEMPTS) {
 					queueAttempts++;
-					slot.release();
-					await sleep$1(Math.max(this.window.waitMs(), QUEUE_BACKOFF_MS), options.signal);
+					await sleep$1(QUEUE_BACKOFF_MS, options.signal);
 					continue;
 				}
 				if (error instanceof CodeArtsRetry && error.kind === "queue") throw queueExhausted();
@@ -4259,13 +4160,17 @@ var CodeArtsProvider = class {
 	async send(chatId, payload, model, benefitDropped, signal) {
 		const credential = await this.credential();
 		const benefit = !benefitDropped && this.isBenefitModel(model);
+		const otSessionId = credential.otSessionId ?? this.sessionId.replace(/-/g, "");
 		const headers = {
 			...headersForFetch(await signRequestHuawei(credential.accessKeyId, credential.secretAccessKey, credential.securityToken, "POST", CODEARTS_CHAT_URL, payload, benefit ? { maas_type: "benefit" } : void 0)),
 			"content-type": "application/json",
 			accept: "text/event-stream",
 			"chat-id": chatId,
 			"session-id": this.sessionId,
-			lang: "en"
+			lang: "en",
+			"x-ot-session-id": otSessionId,
+			"user-session-id": otSessionId,
+			"x-ot-trace-id": randomUUID().replace(/-/g, "")
 		};
 		const timeout = AbortSignal.timeout(Math.max(this.options.timeoutMs(), 6e4));
 		const response = await this.fetcher(CODEARTS_CHAT_URL, {
@@ -4295,7 +4200,6 @@ var CodeArtsProvider = class {
 	}
 	async logoutAccount() {
 		await this.options.auth.logout();
-		this.window.reset();
 		this.invalidateModels();
 		this.invalidateQuota();
 	}
@@ -6359,6 +6263,100 @@ function raccoonHeaders(token) {
 		"x-client-version": RACCOON_CLIENT_VERSION
 	};
 }
+/** Wording the gateway has used alongside that code. */
+const RATE_LIMIT_WORDS = [
+	"user_model_rate_limited",
+	"rate_limit",
+	"频率超限"
+];
+/**
+* Whether a rejection is the token wall rather than a generic failure.
+*
+* Matched on the status, the code and the wording, because the same condition
+* arrives as an HTTP 429 and as an error frame inside a 200 stream.
+*/
+function isRaccoonRateLimited(status, text) {
+	if (status === 429) return true;
+	const lower = text.toLowerCase();
+	return RATE_LIMIT_WORDS.some((word) => lower.includes(word.toLowerCase()));
+}
+/**
+* The wall, as an error the caller can tell apart from a generic 429.
+*
+* A subclass rather than a flag: the in-band path and the HTTP path would
+* otherwise both be caught by the same handler, and the second pass would wrap
+* the message the first pass had already rewritten.
+*/
+var RaccoonRateLimitedError = class extends ProviderHttpError {
+	constructor(message) {
+		super(message, 429);
+		this.name = "RaccoonRateLimitedError";
+	}
+};
+/**
+* The message a user sees once the wall is hit.
+*
+* The raw text says "retry later" and gives no reason, which is why this reads
+* as the plugin being broken. The honest answer names the cause and the one
+* lever the user actually has on this side.
+*/
+function raccoonRateLimited(detail) {
+	return new RaccoonRateLimitedError(`小浣熊账号的模型吞吐已达上限（不是每分钟请求数，而是按账号的 token 速率）。一轮对话带的上下文越大越快撞到，撞到后额度按吞吐缓慢回血，上游也没有给恢复时间。稍等一会儿再发；把该模型的上下文窗口调小，同一会话能撑得更久。上游原文：${detail}`);
+}
+/** The `error` object a compatible gateway puts in an SSE frame. */
+function frameFailure(frame) {
+	const error = frame.error;
+	if (error === void 0) return void 0;
+	if (typeof error === "string") return { message: error };
+	if (error === null || typeof error !== "object") return void 0;
+	const record = error;
+	const code = typeof record.code === "string" ? record.code : void 0;
+	const message = typeof record.message === "string" ? record.message : void 0;
+	return {
+		...code === void 0 ? {} : { code },
+		...message === void 0 ? {} : { message }
+	};
+}
+/**
+* One conversation's session id, stable across the turns of that conversation.
+*
+* The official client never opens a session per request — measured on its own
+* logs, 125 calls over 26 minutes all carried one `sess-0-53927d74` — while an
+* implementation that mints a fresh id per request tells the gateway that one
+* account opened hundreds of sessions within minutes.
+*
+* The id is a pure function of (account, model) rather than a random value, so a
+* restarted or hot-reloaded process continues the same conversation instead of
+* presenting a new session, and two models stay two conversations. The principal
+* itself never leaves the process — only a truncated digest of it does.
+*/
+var RaccoonSessionIds = class {
+	principal;
+	/**
+	* @param principal - a value that is stable for one account and identifies it
+	*   without being the token itself (the JWT's `sub`).
+	*/
+	constructor(principal) {
+		this.principal = principal;
+	}
+	async for(model) {
+		const principal = await this.principal();
+		if (principal === void 0) return void 0;
+		return raccoonSessionId(principal, model);
+	}
+};
+/**
+* `sess-<sequence>-<8 hex>`, the shape the official client sends.
+*
+* The sequence is the official client's process-local counter and is pinned to 0
+* here on purpose: an incrementing value would make one conversation look like a
+* different session after a restart, which is the one thing this id exists to
+* prevent. Identity lives in the digest, which is truncated on purpose — the id
+* must separate accounts without naming them on the wire.
+*/
+function raccoonSessionId(principal, model) {
+	return `sess-0-${createHash("sha256").update(`${principal}\0${model}`).digest("hex").slice(0, 8)}`;
+}
 /** Whether the gateway accepts a token harvested from the browser. */
 async function raccoonAccessTokenValid(token, fetchImpl = fetch) {
 	try {
@@ -6416,12 +6414,16 @@ var RaccoonProvider = class {
 	catalog;
 	quotaCache;
 	quotaInflight;
-	checkinInflight;
 	loginFlow;
 	openRouter;
+	sessionIds;
 	constructor(options) {
 		this.options = options;
 		this.openRouter = options.openRouter ?? new OpenRouterCatalog(options.fetchImpl ?? fetch);
+		this.sessionIds = options.sessionIds ?? new RaccoonSessionIds(async () => {
+			const stored = await options.store.raccoon();
+			return stored === void 0 ? void 0 : decodeJwtFacts(stored.accessToken).subject;
+		});
 		this.catalog = new ModelCatalog(() => this.fetchModels(), options.ttlMs);
 	}
 	async status() {
@@ -6475,6 +6477,21 @@ var RaccoonProvider = class {
 	headers(token) {
 		return raccoonHeaders(token);
 	}
+	/**
+	* The chat headers, including the session id the gateway counts sessions by.
+	*
+	* Only chat carries it: the catalogue, quota and check-in routes are plain
+	* business calls the official client sends outside any session, so leaving
+	* them alone keeps this change to the one route it was measured on.
+	*/
+	async chatHeaders(token, model) {
+		const base = raccoonHeaders(token);
+		const sessionId = await this.sessionIds.for(model).catch(() => void 0);
+		return sessionId === void 0 ? base : {
+			...base,
+			"x-raccoon-session-id": sessionId
+		};
+	}
 	/** Run one JSON call; a rejected token means this login needs a new scan. */
 	async authorizedJson(path, init = {}, extraHeaders = {}) {
 		const token = await this.options.auth.accessToken();
@@ -6590,6 +6607,15 @@ var RaccoonProvider = class {
 			};
 		}
 	}
+	/**
+	* Read today's grant. **Read-only on purpose.**
+	*
+	* The gateway has no claim endpoint: points are issued on the day's first
+	* website visit. Replaying that visit (`setting_info`) can trigger the grant
+	* from here, but doing so takes a state-changing action on the account without
+	* the user asking for it in that moment, so the plugin only reports. Claiming
+	* is the website's job, and the card links there.
+	*/
 	async checkinStatus() {
 		const grant = await this.dailyGrant();
 		if (grant !== void 0) return {
@@ -6609,46 +6635,6 @@ var RaccoonProvider = class {
 			url: RACCOON_CHECKIN_URL
 		};
 	}
-	async claimCheckin() {
-		await this.authorizedJson("/office/v3/setting_info", {}, {
-			"x-client-platform": "web",
-			"x-client-version": "131.0.0.0"
-		});
-		const grant = await this.dailyGrant();
-		if (grant === void 0) return {
-			available: true,
-			claimed: false,
-			deviceClaimed: false,
-			message: "每日积分尚未到账；请打开网页签到",
-			url: RACCOON_CHECKIN_URL
-		};
-		this.invalidateQuota();
-		return {
-			available: true,
-			claimed: true,
-			deviceClaimed: false,
-			...grant.points === void 0 ? {} : { reward: grant.points },
-			...grant.grantedAt === void 0 ? {} : { grantedAt: grant.grantedAt },
-			message: "每日积分已到账",
-			url: RACCOON_CHECKIN_URL
-		};
-	}
-	/**
-	* One automatic check-in pass: report today's grant, or ask the website path
-	* to issue it. Shared so overlapping timers cannot spend two call slots.
-	*/
-	async ensureDailyCheckin() {
-		if (this.checkinInflight !== void 0) return this.checkinInflight;
-		this.checkinInflight = (async () => {
-			const status = await this.checkinStatus();
-			return status.claimed ? status : this.claimCheckin();
-		})();
-		try {
-			return await this.checkinInflight;
-		} finally {
-			this.checkinInflight = void 0;
-		}
-	}
 	async *stream(options, model, readImage, onUnauthorized) {
 		const token = await this.options.auth.accessToken();
 		const body = buildBody(options, model, await buildMessages(options, readImage), buildTools(options));
@@ -6665,7 +6651,7 @@ var RaccoonProvider = class {
 		try {
 			yield* openAiStream({
 				url: `${RACCOON_BASE}/llm/v2/chat/completions`,
-				headers: this.headers(token),
+				headers: await this.chatHeaders(token, model),
 				body,
 				signal: options.signal,
 				fetchImpl: this.options.fetchImpl,
@@ -6673,10 +6659,17 @@ var RaccoonProvider = class {
 					this.invalidateModels();
 					this.invalidateQuota();
 					onUnauthorized?.();
+				},
+				onFrame: (frame) => {
+					const failure = frameFailure(frame);
+					if (failure === void 0) return;
+					const detail = failure.message ?? failure.code ?? "上游返回错误";
+					throw isRaccoonRateLimited(void 0, `${failure.code ?? ""} ${detail}`) ? raccoonRateLimited(detail) : new ProviderHttpError(detail, 502);
 				}
 			});
 		} catch (error) {
 			if (error instanceof ProviderHttpError && (error.status === 401 || error.status === 403)) throw new Error("小浣熊登录已失效，请在设置页点「打开扫码登录」。");
+			if (error instanceof ProviderHttpError && !(error instanceof RaccoonRateLimitedError) && error.status === 429) throw raccoonRateLimited(error.message);
 			throw error;
 		}
 	}
@@ -8883,23 +8876,6 @@ function apply(ctx, config) {
 	syncTimer.unref?.();
 	syncOpenRouter();
 	ctx.effect(() => () => clearInterval(syncTimer), "llm-adapters: openrouter sync");
-	let lastRaccoonCheckin = Date.now();
-	const checkinMonitor = () => {
-		if (!live.raccoonEnabled()) return;
-		const minutes = live.raccoonCheckinMinutes();
-		if (minutes <= 0) return;
-		if (Date.now() - lastRaccoonCheckin < minutes * 6e4) return;
-		lastRaccoonCheckin = Date.now();
-		(async () => {
-			if (!await raccoonProvider.signedIn()) return;
-			await raccoonProvider.ensureDailyCheckin();
-		})().catch((error) => {
-			ctx.logger?.warn(`llm-adapters: 小浣熊签到监测失败：${error instanceof Error ? error.message : String(error)}`);
-		});
-	};
-	const checkinTimer = setInterval(checkinMonitor, 6e4);
-	checkinTimer.unref?.();
-	ctx.effect(() => () => clearInterval(checkinTimer), "llm-adapters: raccoon check-in monitor");
 	ctx.inject(["attachments"], (scope) => {
 		attachments = scope.attachments;
 		scope.effect(() => () => {
@@ -8965,4 +8941,4 @@ function apply(ctx, config) {
 		ctx.logger?.info(`llm-adapters: 设置接口已挂载于 ${API_PREFIX}`);
 	});
 }
-export { API_PREFIX, AccountStore, BRANDS, CODEARTS_BENEFIT_FALLBACK, CODEARTS_BRAND, CODEARTS_CHAT_URL, CODEARTS_CONTEXT_WINDOWS, CODEARTS_DEFAULT_EFFORT, CODEARTS_EFFORTS, CODEARTS_EFFORT_LEVELS, CODEARTS_FALLBACK_CONTEXT_WINDOW, CODEARTS_MAX_OUTPUT_TOKENS, CODEARTS_PROVIDER, CODEARTS_SESSION_LIMIT, CODEARTS_SESSION_WINDOW_MS, CODEARTS_SNAP_BASE, CODEARTS_VISION_MODELS, CodeArtsAdapter, CodeArtsAuth, CodeArtsLogin, CodeArtsProvider, CodeArtsSessionWindow, Config, EMPTY_OPENROUTER_INDEX, LOOMY_ACCOUNT_BASE, LOOMY_BASE, LOOMY_BRAND, LOOMY_PROVIDER, LOOMY_VERSION, LOOMY_WECHAT_APP_ID, LOOMY_WECHAT_REDIRECT_URI, LocalAdapter, LoomyAdapter, LoomyAuth, LoomyProvider, LoomyQrLogin, ModelCatalog, OpenRouterCatalog, ProviderHttpError, RACCOON_BASE, RACCOON_BRAND, RACCOON_CLIENT_VERSION, RACCOON_IAM_BASE, RACCOON_PROVIDER, RACCOON_QR_APP_NAME, RACCOON_QR_ORIGIN, RaccoonAdapter, RaccoonAuth, RaccoonIamError, RaccoonProvider, RaccoonQrLogin, RefreshTokenExpiredError, TRAE_APP_ID, TRAE_BRAND, TRAE_CHAT_FUNCTIONS, TRAE_CHECKIN_REQ_SOURCE, TRAE_DIRECTORY_FUNCTIONS, TRAE_PROVIDER, TRAE_REGION_PROFILES, TRAE_SOLO_CHAT_PATH, TRAE_WEB_USER_AGENT, TraeAdapter, TraeAuth, TraeProvider, TraeWebLogin, apply, buildAuthorizeUrl, buildLoomyWechatAuthUrl, buildMessages, buildOpenRouterIndex, buildQrMatrix, buildRows, buildTools, clampMaxTokens, codeartsContextWindow, codeartsEfforts, consumeOpenAiStream, createRaccoonQr, decodeJwtFacts, defaultEffortOf, effortLabel, exchangeAuthorizationCode, exchangeRefreshToken, extractWechatUuid, fetchLoomyWechatUuid, generateDpopKeyPair, generatePkcePair, headersForFetch, inject, isBrandId, liveConfig, loomyGatewaySessionValid, machineIdOf, mergeTraeModels, modelKey, name, normalizeCodeArtsModelId, normalizeEffort, normalizeTraeRegion, openAiStream, openTraeStream, parseCodeArtsPackage, parseCodeArtsRoster, parseLoomyCatalog, parseModelName, parseOpenRouterModels, parseRaccoonCatalog, parseTraeCheckin, parseTraeClientStorage, parseTraeGlobalUsage, parseTraeRemoteModels, parseTraeUsage, parseTraeWireModels, pickActiveCredit, pollRaccoonQr, priceFactorOf, priceFactorOfLabel, raccoonAccessTokenValid, raccoonQrContent, raccoonRefreshCandidate, readJwtExpiry, readOpenRouter, readTraeClientLogin, renderQrSvg, resolveCapabilities, resolveConfig, resolveModelId, sharedSessionWindow, signDpopJws, signRequestHuawei, sseEvents, startLoginPage, traeClientDataDirs, traeMessages, traeTools };
+export { API_PREFIX, AccountStore, BRANDS, CODEARTS_BENEFIT_FALLBACK, CODEARTS_BRAND, CODEARTS_CHAT_URL, CODEARTS_CONTEXT_WINDOWS, CODEARTS_DEFAULT_EFFORT, CODEARTS_EFFORTS, CODEARTS_EFFORT_LEVELS, CODEARTS_FALLBACK_CONTEXT_WINDOW, CODEARTS_MAX_OUTPUT_TOKENS, CODEARTS_PROVIDER, CODEARTS_SNAP_BASE, CODEARTS_VISION_MODELS, CodeArtsAdapter, CodeArtsAuth, CodeArtsLogin, CodeArtsProvider, Config, EMPTY_OPENROUTER_INDEX, LOOMY_ACCOUNT_BASE, LOOMY_BASE, LOOMY_BRAND, LOOMY_PROVIDER, LOOMY_VERSION, LOOMY_WECHAT_APP_ID, LOOMY_WECHAT_REDIRECT_URI, LocalAdapter, LoomyAdapter, LoomyAuth, LoomyProvider, LoomyQrLogin, ModelCatalog, OpenRouterCatalog, ProviderHttpError, RACCOON_BASE, RACCOON_BRAND, RACCOON_CLIENT_VERSION, RACCOON_IAM_BASE, RACCOON_PROVIDER, RACCOON_QR_APP_NAME, RACCOON_QR_ORIGIN, RaccoonAdapter, RaccoonAuth, RaccoonIamError, RaccoonProvider, RaccoonQrLogin, RefreshTokenExpiredError, TRAE_APP_ID, TRAE_BRAND, TRAE_CHAT_FUNCTIONS, TRAE_CHECKIN_REQ_SOURCE, TRAE_DIRECTORY_FUNCTIONS, TRAE_PROVIDER, TRAE_REGION_PROFILES, TRAE_SOLO_CHAT_PATH, TRAE_WEB_USER_AGENT, TraeAdapter, TraeAuth, TraeProvider, TraeWebLogin, apply, buildAuthorizeUrl, buildLoomyWechatAuthUrl, buildMessages, buildOpenRouterIndex, buildQrMatrix, buildRows, buildTools, clampMaxTokens, codeartsContextWindow, codeartsEfforts, consumeOpenAiStream, createRaccoonQr, decodeJwtFacts, defaultEffortOf, effortLabel, exchangeAuthorizationCode, exchangeRefreshToken, extractWechatUuid, fetchLoomyWechatUuid, generateDpopKeyPair, generatePkcePair, headersForFetch, inject, isBrandId, liveConfig, loomyGatewaySessionValid, machineIdOf, mergeTraeModels, modelKey, name, normalizeCodeArtsModelId, normalizeEffort, normalizeTraeRegion, openAiStream, openTraeStream, otSessionIdOf, parseCodeArtsPackage, parseCodeArtsRoster, parseLoomyCatalog, parseModelName, parseOpenRouterModels, parseRaccoonCatalog, parseTraeCheckin, parseTraeClientStorage, parseTraeGlobalUsage, parseTraeRemoteModels, parseTraeUsage, parseTraeWireModels, pickActiveCredit, pollRaccoonQr, priceFactorOf, priceFactorOfLabel, raccoonAccessTokenValid, raccoonQrContent, raccoonRefreshCandidate, readJwtExpiry, readOpenRouter, readTraeClientLogin, renderQrSvg, resolveCapabilities, resolveConfig, resolveModelId, signDpopJws, signRequestHuawei, sseEvents, startLoginPage, traeClientDataDirs, traeMessages, traeTools };

package/lib/types/account-store.d.ts

@@ -83,12 +83,19 @@ export interface CodeArtsAccountRecord {
         d: string;
     };
     userName?: string;
+    /**
+     * The `x-ot-session-id` this account chats under: one stable value instead of
+     * one per request, because the gateway counts concurrent sessions by it.
+     *
+     * A record written before the field existed derives the same value from the
+     * refresh token's profile, so an existing sign-in needs no new login.
+     */
+    otSessionId?: string;
     /**
      * Models the account answered `InferHub.4004.200` for.
      *
-     * The verdict is worth persisting: asking with `maas_type: benefit` spends
-     * one of the three chat slots the plan allows per minute, even though the
-     * call cannot succeed.
+     * Worth persisting: the account has no allowance pack for them, so without the
+     * verdict every call would pay for a request that cannot succeed.
      */
     noBenefit?: string[];
     loginAt: number;