dsh-xdpool 2.3.0 → 2.4.1

package/CHANGELOG.md

@@ -1,6 +1,65 @@
 # Changelog
 本项目的所有重要变更都会记录在此文件。
 版本号遵循 [语义化版本](https://semver.org/lang/zh-CN/)。
+## 2.4.1 — 11128 的真正成因：system 内容指纹
+> 2.4.0 之后仍有用户撞上 `400 / 11128 Illegal API invocation from an unapproved channel`。定位下来，2.4.0 的归因是错的，这一版修的是真原因。
+### 11128 不是身份头的问题，是 system 内容里的自报句
+网关按**内容指纹**判定「这是不是官方客户端发出来的」，命中串是 system 消息里的 `You are an AI agent powered by DeepSeek Harness`。它只匹配 `role === "system"`（放进 user/tool 不管），字面子串、大小写不敏感，且带起始位置倾向——放在提示词第一行会被拦，前面垫一行普通文字就放行。
+
+这句话由宿主的第一方插件 `@deepseek-ai/dsh-system-prompt` 固定注入，**任何未改装的安装都会 100% 命中**，与账号、密钥、渠道、UA 全部无关——这也解释了为什么它在「1 个账号失败后不再重试」：每个账号都会失败，重试没有意义。
+
+修法是在发送前把 system 里的这句改写成网关放行的同义句（`powered by` → `running on`），只动命中的文本块，图片/缓存块的引用原样透传；没命中时返回同一个引用，不为「什么都没改」付深拷贝代价。改写清单做成了数组，将来网关加句只是多一行。
+
+**实测（真实国内网关，同一账号三组对照）**：
+| 请求体 | 结果 |
+|---|---|
+| 经本插件改写 | `429 / 6004`（通过渠道校验，仅受频率限制） |
+| 原样含该句（对照） | `400 / 11128` |
+| system 干净（基线） | `429 / 6004` |
+
+### 国际网关要求首条消息是 system
+`www.workbuddy.ai` 对一个以 `user` 开头的请求体回 `400 / 11128 "first message is not system prompt"`，而**同一个请求体在国内网关是被接受的**。同一个错误码，两种完全不同成因——这正是不能只按业务码分支的原因。现在国际版请求若首条不是 system，会在最前面补一条占位 system，用户自己的消息位置和措辞一个字符都不动。
+
+### 思考档位「关闭」会被两区同时拒绝
+适配层把 `thinkingLevelMap.off` 钉成字面量 `'off'`，而 pi-ai 对「没有显式指定档位」的请求发的就是它——所以选「默认」和选「关闭」产生了同一个请求体。两区都不接受这个值：国际版 GPT 族回 `400 / 11133`（`extError.param === 'reasoning.effort'`），国内版在声明了 `onlyReasoning` 的模型上报错。现在发送前把它删掉，让上游走自己的默认档。**省略**是唯一在各个区、各个模型上都实测通过的写法；换成字面量 `'none'` 不是安全替代。
+
+**实测（真实双网关，同一账号前后对照）**：
+
+| 请求 | 修复前 | 修复后 |
+|---|---|---|
+| 国际版，首条为 user | `400 / 11128 first message is not system prompt` | `200` |
+| 国际版，`reasoning_effort: "off"` | `400 / 11133 extError.param=reasoning.effort` | `200` |
+
+### 免费模型限流提示：说清配额是按会话算的
+免费车道的拒绝原文是「the provider asked to wait 74558s」——二十小时，读起来像是整个免费车道被封了。实际配额**按上游会话计**，换一个会话（或换一个模型）就另有一份额度。现在限流错误的文案补上了这一句；只有限流类错误带这句，凭据失效、模型不存在之类加上去只会引导用户去做无用重试。
+### 其他
+- 测试：新增 system 指纹改写 8 例 + 区域专属请求体规则 12 例 + 免费车道会话配额提示 1 例，全套 71 文件 932 用例。
+## 2.4.0 — 会话身份头、清单不再冻结、中文路由可存
+> 这一版收的是「用着用着就报错」的那一类问题：调用被网关判成非法渠道（11128）、国际版模型清单永远停在启动那一刻、中文名字的路由点了保存却没存下、上下文选择一选就弹回去。
+### 会话身份头：补齐官方客户端的契约
+网关对「这一轮对话属于谁」有一组身份头。补齐官方客户端的契约：
+- **会话稳定三头**（同一会话内不变）：`X-Conversation-ID`、`X-Conversation-Request-ID`、`acp-connection-id`。
+- **每请求两头**（逐次随机）：`X-Conversation-Message-ID`、`X-Request-ID`。
+- 另附 `X-Agent-Intent`、`X-Agent-Purpose`、`x-codebuddy-request`、`X-Private-Data`、`X-Product` 与 W3C/B3 追踪族。
+
+> **口径更正（见 2.4.1）**：本版曾把 `400 / 11128` 归因于「缺会话身份头」。这个归因是错的——实测国内网关根本不检查这组头（不带任何身份头、system 干净时正常放行）。身份头照官方契约补齐没有坏处，但 11128 的真正成因是 system 内容里的自报句，已在 2.4.1 修掉。
+
+顺带修掉同源的一处：`role: "developer"` 的消息在国内网关同样触发 11128，现统一上送为 `system`。
+### 国际版模型清单：不再冻结在启动那一刻
+`createProvider` 的 `models` 是在创建 provider 时快照下来的数组，而原来的 `invalidate()` 只重排了 profile、复用了同一个 provider 对象，宿主按对象身份记忆化 `current()`，于是**新拿到的模型清单永远送不到选择器**——表现是「重新获取模型」点了没反应、国际版清单停在开机那一刻。现在每次 `invalidate()` 重建 provider 与 profile 两个对象，并通过 adapter handle 的 `replace()` 重新登记路由，由宿主自己发出 `llm/adapters-updated`。
+（附带核查：国际版本来就有独立的 `modelSelectionGlobal` 键，与国内白名单分开读写，这一处无需改动。）
+### 中文路由名现在存得下
+链名 slug 原来是 `[^a-z0-9]+ → '-'`，中文名整串被压成空 id，规范化时被当作无效条目静默丢弃——于是「保存成功」但什么都没存。改为 Unicode 感知的 slug（`\p{L}`/`\p{N}`），拉丁名形态完全不变，已存的路由 id 继续可解析。
+### 上下文大小：选中态不再一选就弹回
+渠道卡把「从未保存过预算」当成了「已收窄到 200K」显示，用户点「完整」清空预算后该行又算回 200K，选中态立刻跳回第一项。现在未保存过预算就显示完整窗口；清空预算时同时保留 `nativeContextWindow`，radio 组不会在指针底下消失。
+### 思考档位：不再只认一档
+pack 兜底表把十几个模型的 `reasoningEfforts` 写成了单档（`["medium"]` / `["high"]`），于是国际版 `Deepseek-V4.1-Flash` 只能选 High。逐模型实测（两网关、五档全跑）后按真实能力补齐：凡是能调通的模型五档都接受、且各档 `reasoning_content` 长度不同。`gpt-5.3-codex` 五档皆 200 但 reasoning 恒空——档位合法而不可观测，已经在表里注明；`gemini-3.5-flash` 与 `primary-model` 的 400 是账号套餐权限（`model [X] is not allowed for user type [personal]`），与档位无关，未动。
+### 重复账号：同一个人不再显示成两个
+桌面端凭据同时带数字 `uin` 与 UUID `uid`，扫码渠道只带 `uid`，而账号 id 是按 `uin` 优先哈希的——同一个人走两条登录路径就整出两个池条目，各自带一部分模型。现在按 `uid` 建次级索引合并到同一个池条目，既不改哈希（否则全部冷却、失效标记、忽略名单、持久化快照一次性作废），也与凭据到达顺序无关。
+### 「重新检测账号」不再静默失败
+`listAllAccounts` 漏了 `await`，拿到的是一个 Promise，随后的 `.filter` 抛 `is not a function` 并被外层 try/catch 吞掉，表现是扫码账号全部并进池、重新检测永远没结果。
+### 其他
+- 测试：新增会话身份头与中文路由名的回归用例，全套 68 文件 910 用例。
 ## 2.3.0 — 扫码渠道接入账号池，免密车道修复，数据看板有了真数
 > 这一版收掉三件大事：扫码登录的账号真正归入账号池并按渠道展示；免费模型车道跟上上游的新校验、不再动辄 403；数据看板不再是一排零。
 ### 账号池：扫码渠道按渠道入池、逐模型可控

package/lib/bin.js

@@ -277,6 +277,164 @@ const CHAT_IDENTITY_HEADERS = Object.freeze({
 	"X-IDE-Type": "WorkBuddy",
 	"X-IDE-Version": "5.5.6"
 });
+/**
+* System-prompt strings the gateway treats as evidence of a non-official
+* client, and the rewriting that clears the check.
+*
+* WHY THIS EXISTS
+*
+* `code 11128 Illegal API invocation from an unapproved channel` has a SECOND
+* trigger, separate from the conversation-identity headers below: the gateway
+* matches the CONTENT of system messages against a list of strings that identify
+* other agents' clients. The first-party DSH system-prompt plugin injects
+* `You are an AI agent powered by DeepSeek Harness.` as the opening line of
+* every conversation, so every request from an unmodified install is rejected —
+* regardless of which account in the pool is used, which is why the reported
+* symptom was a 400 on the first account with no retry.
+*
+* Boundary, measured against the live domestic gateway (429 = accepted,
+* 400/11128 = blocked):
+*   - the match is a literal substring, BUT only near the START of a system
+*     message: the sentence blocked a longer prompt when it was the first line,
+*     and passed when one plain line was prepended to it ("trailing text after
+*     it" above is the same finding from the other side).
+*   - it is case-insensitive (`you are an ai agent powered by deepseek harness`
+*     is blocked too).
+*   - only `role: "system"` is inspected: the same sentence in a `user` or
+*     `tool` message passes, and it is blocked in a second system message too.
+*   - `deepseek harness` alone, or `powered by deepseek` alone, passes — the
+*     whole phrase is what matches.
+*   - replacing `powered by` with `running on` passes, which is the rewrite
+*     below: it keeps the meaning for the model and shares no substring with
+*     the fingerprint.
+*
+* A prepended blank line would equally defeat a start-anchored check, but it
+* would leave the banned string in the body: the list is server-side and may
+* lose its anchor tomorrow. Rewriting the phrase is what the report's repro
+* verified, so that is what this does.
+*
+* The list is an ARRAY because these fingerprints are appended to over time; a
+* future rejection should be one more row here, not a new code path.
+*/
+const BANNED_SYSTEM_PHRASES = Object.freeze([[/powered by DeepSeek Harness/gi, "running on DeepSeek Harness"]]);
+/**
+* Rewrite fingerprinted system content, leaving everything else byte-identical.
+*
+* System content arrives as either a plain string or an array of content blocks
+* (an OpenAI-compatible body allows both), so both shapes are handled — fixing
+* only the string form would miss every request that carries an image or a
+* cache-control block.
+*
+* Returns the SAME reference when nothing matched: a long conversation's system
+* prompt can be large, and paying for a deep copy to change nothing is waste.
+* Only matched `text` blocks are rebuilt; image and cache blocks pass through
+* untouched so the request shape survives.
+*/
+function sanitizeSystemContent(content) {
+	if (typeof content === "string") return sanitizeSystemText(content);
+	if (!Array.isArray(content)) return content;
+	let changed = false;
+	const next = content.map((block) => {
+		if (typeof block !== "object" || block === null || Array.isArray(block)) return block;
+		const record = block;
+		const text = record["text"];
+		if (typeof text !== "string") return block;
+		const cleaned = sanitizeSystemText(text);
+		if (cleaned === text) return block;
+		changed = true;
+		return {
+			...record,
+			text: cleaned
+		};
+	});
+	return changed ? next : content;
+}
+/**
+* Apply every banned-phrase rewrite to one string.
+*
+* The RegExp is rebuilt per call on purpose: a shared `g`-flagged instance
+* carries `lastIndex` between calls, so the second request in a conversation
+* could skip a match the first one caught.
+*/
+function sanitizeSystemText(text) {
+	let out = text;
+	for (const [pattern, replacement] of BANNED_SYSTEM_PHRASES) out = out.replace(new RegExp(pattern.source, pattern.flags), replacement);
+	return out;
+}
+/**
+* The conversation-identity headers the gateway's channel check demands.
+*
+* WHY THIS EXISTS
+*
+* `www.workbuddy.ai` turned "channel verification" from a soft check into a
+* hard one. A `POST /v2/chat/completions` missing ANY of `X-Conversation-ID`,
+* `X-Conversation-Request-ID` or `X-Conversation-Message-ID` is rejected with
+* business code 11128 (`msg: "request illegal"` on the international gateway,
+* `msg: "conversation id missing"` in the earlier wording). Verified by
+* elimination against real credentials: delete any one of the three and the
+* call 400s; send all three and it 200s with normal SSE. Every other header the
+* desktop sends (`acp-connection-id`, `X-Request-ID`, `X-Agent-Intent`,
+* `X-Private-Data`, the W3C/B3 tracing family) can be dropped without effect,
+* so only these three are load-bearing.
+*
+* The values are NOT validated — plain non-UUID strings pass — so the gateway
+* is checking for presence and consistency, not for a signature. That lets us
+* derive them from the conversation instead of registering state with upstream.
+*
+* 11128 IS MULTIPLEXED. The domestic gateway returns the same code with
+* `msg: "Illegal API invocation from an unapproved channel"` for a different
+* cause entirely: a `developer`-role message. `prepareChatBody()` already
+* rewrites that role; see its comment. Never branch on the business code alone.
+*/
+const CONVERSATION_STABLE_SALT = "conversation";
+/**
+* The three stable conversation ids for one logical conversation.
+*
+* Stable across every call and every account in a conversation, because the
+* upstream treats the triple as one session: regenerating them per request
+* would read as a new conversation on each turn. Derived rather than stored so
+* the identity survives a restart and needs no registry — the same reasoning
+* `deriveDeviceId` uses for the device id.
+*
+* Falls back to a random seed when the caller has no key (a one-shot compaction
+* outside any conversation), which is safe because presence is all that is
+* checked.
+*/
+function conversationIdentityOf(conversationKey, seed = randomUUID) {
+	const base = conversationKey === void 0 || conversationKey === "" ? seed() : conversationKey;
+	const hash = (suffix) => createHash("sha256").update(`${CONVERSATION_STABLE_SALT}:${suffix}:${base}`).digest("hex");
+	const asUuid = (hex) => `${hex.slice(0, 8)}-${hex.slice(8, 12)}-${hex.slice(12, 16)}-${hex.slice(16, 20)}-${hex.slice(20, 32)}`;
+	return {
+		conversationId: asUuid(hash("id")),
+		conversationRequestId: asUuid(hash("request")),
+		acpConnectionId: asUuid(hash("acp"))
+	};
+}
+/**
+* Full header set for one chat completion.
+*
+* `chatHeaders()` stays untouched on purpose: it is shared with the
+* points-automation endpoints, which have never been shown to carry
+* conversation identity, and adding it there would change how that traffic is
+* attributed. The per-request ids are fresh UUIDs every call; only the triple
+* from `conversationIdentityOf` is stable.
+*/
+function chatCompletionHeaders(credential, conversationKey) {
+	const identity = conversationIdentityOf(conversationKey);
+	return {
+		...chatHeaders(credential),
+		...CHAT_IDENTITY_HEADERS,
+		"X-Conversation-ID": identity.conversationId,
+		"X-Conversation-Request-ID": identity.conversationRequestId,
+		"acp-connection-id": identity.acpConnectionId,
+		"X-Conversation-Message-ID": randomUUID(),
+		"X-Request-ID": randomUUID(),
+		"X-Agent-Intent": "craft",
+		"X-Agent-Purpose": "conversation",
+		"X-Private-Data": "true",
+		"x-codebuddy-request": "1"
+	};
+}
 /** Refresh-endpoint headers; X-Refresh-Token appears here and nowhere else. */
 function refreshHeaders(credential) {
 	const headers = {
@@ -804,7 +962,15 @@ var WorkBuddyUpstreamClient = class {
 	* Normalize an OpenAI chat-completions body for the WorkBuddy upstream:
 	* force `stream: true` (the upstream rejects non-streaming), convert the
 	* DSH `developer` role into `system` (upstream rejects `developer` with
-	* business code 11128), and flatten `tool_choice` into its string form.
+	* business code 11128), rewrite system content the gateway fingerprints as a
+	* non-official client (also 11128), and flatten `tool_choice` into its string
+	* form.
+	*
+	* This is the REGION-BLIND half. Everything here is measured good on both
+	* gateways, so it runs before the region is known — the pool lane picks its
+	* account after this call. The two region-only rules live in
+	* {@link prepareChatBodyFor} instead; see its comment for why they cannot be
+	* folded in here.
 	*/
 	prepareChatBody(raw) {
 		let body;
@@ -823,6 +989,7 @@ var WorkBuddyUpstreamClient = class {
 			if (typeof value !== "object" || value === null || Array.isArray(value)) continue;
 			const message = value;
 			if (message["role"] === "developer") message["role"] = "system";
+			if (message["role"] === "system") message["content"] = sanitizeSystemContent(message["content"]);
 		}
 		const choice = obj["tool_choice"];
 		if (typeof choice === "string") {
@@ -849,6 +1016,61 @@ var WorkBuddyUpstreamClient = class {
 		return JSON.stringify(obj);
 	}
 	/**
+	* The region-only half of body normalisation, applied once the account is
+	* known.
+	*
+	* WHY IT IS SEPARATE
+	*
+	* The pool lane calls {@link prepareChatBody} BEFORE it picks an account —
+	* the body has to be ready for whichever account wins. So anything that
+	* depends on the region cannot live there. Two rules do:
+	*
+	* 1. **International requires a leading `system`.** Measured: a body whose
+	*    first message is `user` answers `400 / 11128 "first message is not
+	*    system prompt"` on `www.workbuddy.ai`, while the identical body is
+	*    accepted on the domestic gateway. Same code, different cause — the
+	*    domestic gateway uses 11128 for the `developer` role and for the
+	*    content fingerprint, which is why nothing may branch on the code alone.
+	*    A placeholder assistant prompt is prepended, never merged: every
+	*    caller-supplied message keeps its position and wording.
+	*
+	* 2. **Neither gateway accepts `reasoning_effort: "off"`.** The adapter pins
+	*    `thinkingLevelMap.off` to the literal `'off'` so the level stays
+	*    selectable, and pi-ai emits that value for any request carrying no
+	*    explicit level — so "Default" and "Off" produce the same body.
+	*    Measured: the international GPT family answers `400 / 11133`
+	*    (`extError.param === 'reasoning.effort'`), and the domestic gateway
+	*    rejects it on models that declare `onlyReasoning`. Omission is the only
+	*    spelling measured good everywhere; a literal `'none'` is NOT a safe
+	*    substitute (accepted by the GPT-5.6 family, rejected by others).
+	*
+	* Both rules are no-ops on a body that already complies, so this is safe to
+	* run over an already-prepared body.
+	*/
+	prepareChatBodyFor(raw, region) {
+		const prepared = this.prepareChatBody(raw);
+		let body;
+		try {
+			body = JSON.parse(prepared);
+		} catch {
+			return prepared;
+		}
+		if (typeof body !== "object" || body === null || Array.isArray(body)) return prepared;
+		const obj = body;
+		if (obj["reasoning_effort"] === "off") delete obj["reasoning_effort"];
+		if (region === "global") {
+			const messages = obj["messages"];
+			if (Array.isArray(messages)) {
+				const first = messages[0];
+				if (!(typeof first === "object" && first !== null && !Array.isArray(first) && first["role"] === "system")) messages.unshift({
+					role: "system",
+					content: "You are a helpful assistant."
+				});
+			}
+		}
+		return JSON.stringify(obj);
+	}
+	/**
 	* Parse a raw OpenAI chat body without normalising it.
 	*
 	* The compactor needs the message array as objects, while `chatStream` only
@@ -864,12 +1086,20 @@ var WorkBuddyUpstreamClient = class {
 		if (typeof body !== "object" || body === null || Array.isArray(body)) return void 0;
 		return body;
 	}
-	/** Re-serialise `base` with a rewritten `messages` array, still normalised. */
-	buildChatBody(base, messages) {
-		return this.prepareChatBody(JSON.stringify({
+	/**
+	* Re-serialise `base` with a rewritten `messages` array, still normalised.
+	*
+	* `region` is optional so the composite lanes (aggregate / free) can keep
+	* calling it without one; when supplied, the region-only rules of
+	* {@link prepareChatBodyFor} ride along. Both in-repo callers are the
+	* compaction paths, which always know their shim's region.
+	*/
+	buildChatBody(base, messages, region) {
+		const raw = JSON.stringify({
 			...base,
 			messages
-		}));
+		});
+		return region === void 0 ? this.prepareChatBody(raw) : this.prepareChatBodyFor(raw, region);
 	}
 	/**
 	* Run one NON-streaming completion and return the assistant text.
@@ -879,13 +1109,10 @@ var WorkBuddyUpstreamClient = class {
 	* single string. Throws on any failure: the compactor then falls back to
 	* plain truncation rather than failing the user's turn.
 	*/
-	async completeChat(credential, prepared, signal) {
+	async completeChat(credential, prepared, signal, conversationKey) {
 		const response = await this.fetchImpl(`${chatBase(credential)}/v2/chat/completions`, {
 			method: "POST",
-			headers: {
-				...chatHeaders(credential),
-				...CHAT_IDENTITY_HEADERS
-			},
+			headers: chatCompletionHeaders(credential, conversationKey),
 			body: prepared,
 			...signal === void 0 ? {} : { signal }
 		});
@@ -897,15 +1124,12 @@ var WorkBuddyUpstreamClient = class {
 		return await readCompletionText(response.body);
 	}
 	/** Forward one chat completion. Never throws for upstream failures. */
-	async chatStream(credential, prepared, signal) {
+	async chatStream(credential, prepared, signal, conversationKey) {
 		let response;
 		try {
 			response = await this.fetchImpl(`${chatBase(credential)}/v2/chat/completions`, {
 				method: "POST",
-				headers: {
-					...chatHeaders(credential),
-					...CHAT_IDENTITY_HEADERS
-				},
+				headers: chatCompletionHeaders(credential, conversationKey),
 				body: prepared,
 				...signal === void 0 ? {} : { signal }
 			});
@@ -3325,8 +3549,17 @@ var WorkBuddyAccountPool = class {
 				});
 			}
 		}
+		const byUid = /* @__PURE__ */ new Map();
+		for (const account of byId.values()) {
+			const uid = account.credential.uid;
+			if (typeof uid === "string" && uid !== "" && !byUid.has(uid)) byUid.set(uid, account.id);
+		}
 		for (const credential of found) {
-			const id = workbuddyAccountId(credential);
+			const byIdentity = workbuddyAccountId(credential);
+			const uid = credential.uid;
+			const knownId = typeof uid === "string" && uid !== "" ? byUid.get(uid) : void 0;
+			const id = knownId ?? byIdentity;
+			if (knownId === void 0 && typeof uid === "string" && uid !== "") byUid.set(uid, id);
 			const existing = byId.get(id);
 			if (existing === void 0) {
 				byId.set(id, {

package/lib/client.js

@@ -11674,7 +11674,7 @@ window.__ModuleLoader__.load({
 						...model,
 						...budget === void 0 ? {
 							contextBudget: void 0,
-							nativeContextWindow: void 0
+							contextWindow: model.nativeContextWindow ?? model.contextWindow
 						} : {
 							contextBudget: budget,
 							contextWindow: budget,
@@ -12079,7 +12079,7 @@ window.__ModuleLoader__.load({
 													const recuse = recommendUse(merged);
 													const native = model.nativeContextWindow ?? model.contextWindow;
 													const capped = native !== void 0 && native > CONTEXT_BUDGET_DEFAULT;
-													const budget = model.contextBudget ?? (capped ? CONTEXT_BUDGET_DEFAULT : native);
+													const budget = model.contextBudget ?? native;
 													return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("li", {
 														className: "dsm-xdpool-cpool-modelrow",
 														"data-off": !on,
@@ -12188,7 +12188,7 @@ window.__ModuleLoader__.load({
 		*
 		* @module dsh-xdpool/version
 		*/
-		const POOL_VERSION = "2.3.0";
+		const POOL_VERSION = "2.4.1";
 		//#endregion
 		//#region src/client/PoolCard.tsx
 		/**

package/lib/index.d.ts

@@ -469,9 +469,50 @@ export declare class WorkBuddyUpstreamClient {
    * Normalize an OpenAI chat-completions body for the WorkBuddy upstream:
    * force `stream: true` (the upstream rejects non-streaming), convert the
    * DSH `developer` role into `system` (upstream rejects `developer` with
-   * business code 11128), and flatten `tool_choice` into its string form.
+   * business code 11128), rewrite system content the gateway fingerprints as a
+   * non-official client (also 11128), and flatten `tool_choice` into its string
+   * form.
+   *
+   * This is the REGION-BLIND half. Everything here is measured good on both
+   * gateways, so it runs before the region is known — the pool lane picks its
+   * account after this call. The two region-only rules live in
+   * {@link prepareChatBodyFor} instead; see its comment for why they cannot be
+   * folded in here.
    */
   prepareChatBody(raw: string): string;
+  /**
+   * The region-only half of body normalisation, applied once the account is
+   * known.
+   *
+   * WHY IT IS SEPARATE
+   *
+   * The pool lane calls {@link prepareChatBody} BEFORE it picks an account —
+   * the body has to be ready for whichever account wins. So anything that
+   * depends on the region cannot live there. Two rules do:
+   *
+   * 1. **International requires a leading `system`.** Measured: a body whose
+   *    first message is `user` answers `400 / 11128 "first message is not
+   *    system prompt"` on `www.workbuddy.ai`, while the identical body is
+   *    accepted on the domestic gateway. Same code, different cause — the
+   *    domestic gateway uses 11128 for the `developer` role and for the
+   *    content fingerprint, which is why nothing may branch on the code alone.
+   *    A placeholder assistant prompt is prepended, never merged: every
+   *    caller-supplied message keeps its position and wording.
+   *
+   * 2. **Neither gateway accepts `reasoning_effort: "off"`.** The adapter pins
+   *    `thinkingLevelMap.off` to the literal `'off'` so the level stays
+   *    selectable, and pi-ai emits that value for any request carrying no
+   *    explicit level — so "Default" and "Off" produce the same body.
+   *    Measured: the international GPT family answers `400 / 11133`
+   *    (`extError.param === 'reasoning.effort'`), and the domestic gateway
+   *    rejects it on models that declare `onlyReasoning`. Omission is the only
+   *    spelling measured good everywhere; a literal `'none'` is NOT a safe
+   *    substitute (accepted by the GPT-5.6 family, rejected by others).
+   *
+   * Both rules are no-ops on a body that already complies, so this is safe to
+   * run over an already-prepared body.
+   */
+  prepareChatBodyFor(raw: string, region: WorkBuddyRegion): string;
   /**
    * Parse a raw OpenAI chat body without normalising it.
    *
@@ -479,8 +520,15 @@ export declare class WorkBuddyUpstreamClient {
    * accepts the serialised string form.
    */
   parseChatBody(raw: string): Record<string, unknown> | undefined;
-  /** Re-serialise `base` with a rewritten `messages` array, still normalised. */
-  buildChatBody(base: Record<string, unknown>, messages: readonly ChatMessage[]): string;
+  /**
+   * Re-serialise `base` with a rewritten `messages` array, still normalised.
+   *
+   * `region` is optional so the composite lanes (aggregate / free) can keep
+   * calling it without one; when supplied, the region-only rules of
+   * {@link prepareChatBodyFor} ride along. Both in-repo callers are the
+   * compaction paths, which always know their shim's region.
+   */
+  buildChatBody(base: Record<string, unknown>, messages: readonly ChatMessage[], region?: WorkBuddyRegion): string;
   /**
    * Run one NON-streaming completion and return the assistant text.
    *
@@ -489,9 +537,9 @@ export declare class WorkBuddyUpstreamClient {
    * single string. Throws on any failure: the compactor then falls back to
    * plain truncation rather than failing the user's turn.
    */
-  completeChat(credential: WorkBuddyCredential, prepared: string, signal?: AbortSignal): Promise<string>;
+  completeChat(credential: WorkBuddyCredential, prepared: string, signal?: AbortSignal, conversationKey?: string): Promise<string>;
   /** Forward one chat completion. Never throws for upstream failures. */
-  chatStream(credential: WorkBuddyCredential, prepared: string, signal?: AbortSignal): Promise<ChatStreamResult>;
+  chatStream(credential: WorkBuddyCredential, prepared: string, signal?: AbortSignal, conversationKey?: string): Promise<ChatStreamResult>;
   /** POST the token-refresh endpoint; the caller merges the outcome. */
   refreshToken(credential: WorkBuddyCredential): Promise<WorkBuddyRefreshOutcome>;
   /**

package/lib/index.js

@@ -328,6 +328,164 @@ const CHAT_IDENTITY_HEADERS = Object.freeze({
 	"X-IDE-Type": "WorkBuddy",
 	"X-IDE-Version": "5.5.6"
 });
+/**
+* System-prompt strings the gateway treats as evidence of a non-official
+* client, and the rewriting that clears the check.
+*
+* WHY THIS EXISTS
+*
+* `code 11128 Illegal API invocation from an unapproved channel` has a SECOND
+* trigger, separate from the conversation-identity headers below: the gateway
+* matches the CONTENT of system messages against a list of strings that identify
+* other agents' clients. The first-party DSH system-prompt plugin injects
+* `You are an AI agent powered by DeepSeek Harness.` as the opening line of
+* every conversation, so every request from an unmodified install is rejected —
+* regardless of which account in the pool is used, which is why the reported
+* symptom was a 400 on the first account with no retry.
+*
+* Boundary, measured against the live domestic gateway (429 = accepted,
+* 400/11128 = blocked):
+*   - the match is a literal substring, BUT only near the START of a system
+*     message: the sentence blocked a longer prompt when it was the first line,
+*     and passed when one plain line was prepended to it ("trailing text after
+*     it" above is the same finding from the other side).
+*   - it is case-insensitive (`you are an ai agent powered by deepseek harness`
+*     is blocked too).
+*   - only `role: "system"` is inspected: the same sentence in a `user` or
+*     `tool` message passes, and it is blocked in a second system message too.
+*   - `deepseek harness` alone, or `powered by deepseek` alone, passes — the
+*     whole phrase is what matches.
+*   - replacing `powered by` with `running on` passes, which is the rewrite
+*     below: it keeps the meaning for the model and shares no substring with
+*     the fingerprint.
+*
+* A prepended blank line would equally defeat a start-anchored check, but it
+* would leave the banned string in the body: the list is server-side and may
+* lose its anchor tomorrow. Rewriting the phrase is what the report's repro
+* verified, so that is what this does.
+*
+* The list is an ARRAY because these fingerprints are appended to over time; a
+* future rejection should be one more row here, not a new code path.
+*/
+const BANNED_SYSTEM_PHRASES = Object.freeze([[/powered by DeepSeek Harness/gi, "running on DeepSeek Harness"]]);
+/**
+* Rewrite fingerprinted system content, leaving everything else byte-identical.
+*
+* System content arrives as either a plain string or an array of content blocks
+* (an OpenAI-compatible body allows both), so both shapes are handled — fixing
+* only the string form would miss every request that carries an image or a
+* cache-control block.
+*
+* Returns the SAME reference when nothing matched: a long conversation's system
+* prompt can be large, and paying for a deep copy to change nothing is waste.
+* Only matched `text` blocks are rebuilt; image and cache blocks pass through
+* untouched so the request shape survives.
+*/
+function sanitizeSystemContent(content) {
+	if (typeof content === "string") return sanitizeSystemText(content);
+	if (!Array.isArray(content)) return content;
+	let changed = false;
+	const next = content.map((block) => {
+		if (typeof block !== "object" || block === null || Array.isArray(block)) return block;
+		const record = block;
+		const text = record["text"];
+		if (typeof text !== "string") return block;
+		const cleaned = sanitizeSystemText(text);
+		if (cleaned === text) return block;
+		changed = true;
+		return {
+			...record,
+			text: cleaned
+		};
+	});
+	return changed ? next : content;
+}
+/**
+* Apply every banned-phrase rewrite to one string.
+*
+* The RegExp is rebuilt per call on purpose: a shared `g`-flagged instance
+* carries `lastIndex` between calls, so the second request in a conversation
+* could skip a match the first one caught.
+*/
+function sanitizeSystemText(text) {
+	let out = text;
+	for (const [pattern, replacement] of BANNED_SYSTEM_PHRASES) out = out.replace(new RegExp(pattern.source, pattern.flags), replacement);
+	return out;
+}
+/**
+* The conversation-identity headers the gateway's channel check demands.
+*
+* WHY THIS EXISTS
+*
+* `www.workbuddy.ai` turned "channel verification" from a soft check into a
+* hard one. A `POST /v2/chat/completions` missing ANY of `X-Conversation-ID`,
+* `X-Conversation-Request-ID` or `X-Conversation-Message-ID` is rejected with
+* business code 11128 (`msg: "request illegal"` on the international gateway,
+* `msg: "conversation id missing"` in the earlier wording). Verified by
+* elimination against real credentials: delete any one of the three and the
+* call 400s; send all three and it 200s with normal SSE. Every other header the
+* desktop sends (`acp-connection-id`, `X-Request-ID`, `X-Agent-Intent`,
+* `X-Private-Data`, the W3C/B3 tracing family) can be dropped without effect,
+* so only these three are load-bearing.
+*
+* The values are NOT validated — plain non-UUID strings pass — so the gateway
+* is checking for presence and consistency, not for a signature. That lets us
+* derive them from the conversation instead of registering state with upstream.
+*
+* 11128 IS MULTIPLEXED. The domestic gateway returns the same code with
+* `msg: "Illegal API invocation from an unapproved channel"` for a different
+* cause entirely: a `developer`-role message. `prepareChatBody()` already
+* rewrites that role; see its comment. Never branch on the business code alone.
+*/
+const CONVERSATION_STABLE_SALT = "conversation";
+/**
+* The three stable conversation ids for one logical conversation.
+*
+* Stable across every call and every account in a conversation, because the
+* upstream treats the triple as one session: regenerating them per request
+* would read as a new conversation on each turn. Derived rather than stored so
+* the identity survives a restart and needs no registry — the same reasoning
+* `deriveDeviceId` uses for the device id.
+*
+* Falls back to a random seed when the caller has no key (a one-shot compaction
+* outside any conversation), which is safe because presence is all that is
+* checked.
+*/
+function conversationIdentityOf(conversationKey, seed = randomUUID) {
+	const base = conversationKey === void 0 || conversationKey === "" ? seed() : conversationKey;
+	const hash = (suffix) => createHash("sha256").update(`${CONVERSATION_STABLE_SALT}:${suffix}:${base}`).digest("hex");
+	const asUuid = (hex) => `${hex.slice(0, 8)}-${hex.slice(8, 12)}-${hex.slice(12, 16)}-${hex.slice(16, 20)}-${hex.slice(20, 32)}`;
+	return {
+		conversationId: asUuid(hash("id")),
+		conversationRequestId: asUuid(hash("request")),
+		acpConnectionId: asUuid(hash("acp"))
+	};
+}
+/**
+* Full header set for one chat completion.
+*
+* `chatHeaders()` stays untouched on purpose: it is shared with the
+* points-automation endpoints, which have never been shown to carry
+* conversation identity, and adding it there would change how that traffic is
+* attributed. The per-request ids are fresh UUIDs every call; only the triple
+* from `conversationIdentityOf` is stable.
+*/
+function chatCompletionHeaders(credential, conversationKey) {
+	const identity = conversationIdentityOf(conversationKey);
+	return {
+		...chatHeaders(credential),
+		...CHAT_IDENTITY_HEADERS,
+		"X-Conversation-ID": identity.conversationId,
+		"X-Conversation-Request-ID": identity.conversationRequestId,
+		"acp-connection-id": identity.acpConnectionId,
+		"X-Conversation-Message-ID": randomUUID(),
+		"X-Request-ID": randomUUID(),
+		"X-Agent-Intent": "craft",
+		"X-Agent-Purpose": "conversation",
+		"X-Private-Data": "true",
+		"x-codebuddy-request": "1"
+	};
+}
 /** Refresh-endpoint headers; X-Refresh-Token appears here and nowhere else. */
 function refreshHeaders(credential) {
 	const headers = {
@@ -964,7 +1122,15 @@ var WorkBuddyUpstreamClient = class {
 	* Normalize an OpenAI chat-completions body for the WorkBuddy upstream:
 	* force `stream: true` (the upstream rejects non-streaming), convert the
 	* DSH `developer` role into `system` (upstream rejects `developer` with
-	* business code 11128), and flatten `tool_choice` into its string form.
+	* business code 11128), rewrite system content the gateway fingerprints as a
+	* non-official client (also 11128), and flatten `tool_choice` into its string
+	* form.
+	*
+	* This is the REGION-BLIND half. Everything here is measured good on both
+	* gateways, so it runs before the region is known — the pool lane picks its
+	* account after this call. The two region-only rules live in
+	* {@link prepareChatBodyFor} instead; see its comment for why they cannot be
+	* folded in here.
 	*/
 	prepareChatBody(raw) {
 		let body;
@@ -983,6 +1149,7 @@ var WorkBuddyUpstreamClient = class {
 			if (typeof value !== "object" || value === null || Array.isArray(value)) continue;
 			const message = value;
 			if (message["role"] === "developer") message["role"] = "system";
+			if (message["role"] === "system") message["content"] = sanitizeSystemContent(message["content"]);
 		}
 		const choice = obj["tool_choice"];
 		if (typeof choice === "string") {
@@ -1009,6 +1176,61 @@ var WorkBuddyUpstreamClient = class {
 		return JSON.stringify(obj);
 	}
 	/**
+	* The region-only half of body normalisation, applied once the account is
+	* known.
+	*
+	* WHY IT IS SEPARATE
+	*
+	* The pool lane calls {@link prepareChatBody} BEFORE it picks an account —
+	* the body has to be ready for whichever account wins. So anything that
+	* depends on the region cannot live there. Two rules do:
+	*
+	* 1. **International requires a leading `system`.** Measured: a body whose
+	*    first message is `user` answers `400 / 11128 "first message is not
+	*    system prompt"` on `www.workbuddy.ai`, while the identical body is
+	*    accepted on the domestic gateway. Same code, different cause — the
+	*    domestic gateway uses 11128 for the `developer` role and for the
+	*    content fingerprint, which is why nothing may branch on the code alone.
+	*    A placeholder assistant prompt is prepended, never merged: every
+	*    caller-supplied message keeps its position and wording.
+	*
+	* 2. **Neither gateway accepts `reasoning_effort: "off"`.** The adapter pins
+	*    `thinkingLevelMap.off` to the literal `'off'` so the level stays
+	*    selectable, and pi-ai emits that value for any request carrying no
+	*    explicit level — so "Default" and "Off" produce the same body.
+	*    Measured: the international GPT family answers `400 / 11133`
+	*    (`extError.param === 'reasoning.effort'`), and the domestic gateway
+	*    rejects it on models that declare `onlyReasoning`. Omission is the only
+	*    spelling measured good everywhere; a literal `'none'` is NOT a safe
+	*    substitute (accepted by the GPT-5.6 family, rejected by others).
+	*
+	* Both rules are no-ops on a body that already complies, so this is safe to
+	* run over an already-prepared body.
+	*/
+	prepareChatBodyFor(raw, region) {
+		const prepared = this.prepareChatBody(raw);
+		let body;
+		try {
+			body = JSON.parse(prepared);
+		} catch {
+			return prepared;
+		}
+		if (typeof body !== "object" || body === null || Array.isArray(body)) return prepared;
+		const obj = body;
+		if (obj["reasoning_effort"] === "off") delete obj["reasoning_effort"];
+		if (region === "global") {
+			const messages = obj["messages"];
+			if (Array.isArray(messages)) {
+				const first = messages[0];
+				if (!(typeof first === "object" && first !== null && !Array.isArray(first) && first["role"] === "system")) messages.unshift({
+					role: "system",
+					content: "You are a helpful assistant."
+				});
+			}
+		}
+		return JSON.stringify(obj);
+	}
+	/**
 	* Parse a raw OpenAI chat body without normalising it.
 	*
 	* The compactor needs the message array as objects, while `chatStream` only
@@ -1024,12 +1246,20 @@ var WorkBuddyUpstreamClient = class {
 		if (typeof body !== "object" || body === null || Array.isArray(body)) return void 0;
 		return body;
 	}
-	/** Re-serialise `base` with a rewritten `messages` array, still normalised. */
-	buildChatBody(base, messages) {
-		return this.prepareChatBody(JSON.stringify({
+	/**
+	* Re-serialise `base` with a rewritten `messages` array, still normalised.
+	*
+	* `region` is optional so the composite lanes (aggregate / free) can keep
+	* calling it without one; when supplied, the region-only rules of
+	* {@link prepareChatBodyFor} ride along. Both in-repo callers are the
+	* compaction paths, which always know their shim's region.
+	*/
+	buildChatBody(base, messages, region) {
+		const raw = JSON.stringify({
 			...base,
 			messages
-		}));
+		});
+		return region === void 0 ? this.prepareChatBody(raw) : this.prepareChatBodyFor(raw, region);
 	}
 	/**
 	* Run one NON-streaming completion and return the assistant text.
@@ -1039,13 +1269,10 @@ var WorkBuddyUpstreamClient = class {
 	* single string. Throws on any failure: the compactor then falls back to
 	* plain truncation rather than failing the user's turn.
 	*/
-	async completeChat(credential, prepared, signal) {
+	async completeChat(credential, prepared, signal, conversationKey) {
 		const response = await this.fetchImpl(`${chatBase(credential)}/v2/chat/completions`, {
 			method: "POST",
-			headers: {
-				...chatHeaders(credential),
-				...CHAT_IDENTITY_HEADERS
-			},
+			headers: chatCompletionHeaders(credential, conversationKey),
 			body: prepared,
 			...signal === void 0 ? {} : { signal }
 		});
@@ -1057,15 +1284,12 @@ var WorkBuddyUpstreamClient = class {
 		return await readCompletionText(response.body);
 	}
 	/** Forward one chat completion. Never throws for upstream failures. */
-	async chatStream(credential, prepared, signal) {
+	async chatStream(credential, prepared, signal, conversationKey) {
 		let response;
 		try {
 			response = await this.fetchImpl(`${chatBase(credential)}/v2/chat/completions`, {
 				method: "POST",
-				headers: {
-					...chatHeaders(credential),
-					...CHAT_IDENTITY_HEADERS
-				},
+				headers: chatCompletionHeaders(credential, conversationKey),
 				body: prepared,
 				...signal === void 0 ? {} : { signal }
 			});
@@ -3501,8 +3725,17 @@ var WorkBuddyAccountPool = class {
 				});
 			}
 		}
+		const byUid = /* @__PURE__ */ new Map();
+		for (const account of byId.values()) {
+			const uid = account.credential.uid;
+			if (typeof uid === "string" && uid !== "" && !byUid.has(uid)) byUid.set(uid, account.id);
+		}
 		for (const credential of found) {
-			const id = workbuddyAccountId(credential);
+			const byIdentity = workbuddyAccountId(credential);
+			const uid = credential.uid;
+			const knownId = typeof uid === "string" && uid !== "" ? byUid.get(uid) : void 0;
+			const id = knownId ?? byIdentity;
+			if (knownId === void 0 && typeof uid === "string" && uid !== "") byUid.set(uid, id);
 			const existing = byId.get(id);
 			if (existing === void 0) {
 				byId.set(id, {
@@ -10043,9 +10276,22 @@ function stringList$2(value) {
 	}
 	return out;
 }
-/** The combo id a name slugifies to, or '' when the name cannot form one. */
+/**
+* The combo id a name slugifies to, or '' when the name cannot form one.
+*
+* Unicode-aware on purpose. This used to be `/[^a-z0-9]+/g`, which meant any
+* name that was not ASCII — every Chinese chain name — slugged to the empty
+* string, and `normalizeCombos` then dropped the entry with a bare `continue`.
+* The card reported "saved", the store held nothing, and the chain vanished on
+* the next read: the "new route cannot be saved" report. A name is user data and
+* may be written in any script, so the id has to survive one.
+*
+* Latin names keep the old shape (`writing-fallback`), because ids already
+* stored that way must keep resolving; only characters that used to be stripped
+* to nothing are now kept.
+*/
 function comboId(name) {
-	const slug = text$6(name).toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-+|-+$/g, "").slice(0, 48);
+	const slug = text$6(name).toLowerCase().replace(/[^\p{L}\p{N}]+/gu, "-").replace(/^-+|-+$/gu, "").slice(0, 48).replace(/-+$/u, "");
 	return slug === "" ? "" : `${COMBO_PREFIX}${slug}`;
 }
 /** True when the id names a combo row. */
@@ -10068,7 +10314,9 @@ function isComboRow(row) {
 *
 * A chain with no members is DROPPED rather than kept as an empty row: a model
 * that cannot answer anything is not a model, and showing one in the picker
-* only produces a request that fails with nothing to fix.
+* only produces a request that fails with nothing to fix. The card warns about
+* that case before the save (`route.noMembers`), which is why dropping here is
+* not a silent loss.
 */
 function normalizeCombos(raw) {
 	if (!Array.isArray(raw)) return [];
@@ -11596,7 +11844,6 @@ const NO_COST = {
 	cacheRead: 0,
 	cacheWrite: 0
 };
-/** pi-ai input modalities: images only when the catalog advertises them. */
 function modelInput(info) {
 	return info.supportsImages ? ["text", "image"] : ["text"];
 }
@@ -11773,26 +12020,44 @@ function createWorkBuddyAdapter(options) {
 		if (aggregateCatalog) return [...catalog.providerRows().map((row) => toAggregateModel(row, baseUrl, providerId)), ...catalog.comboRows().map((row) => toComboModel(row, baseUrl, providerId))];
 		return catalog.visible().map((info) => toPiModel(info, baseUrl, providerId));
 	};
-	const provider = {
-		...createProvider({
-			id: providerId,
-			name: displayName,
-			auth: { apiKey: {
-				name: "XD Pool loopback secret",
-				async resolve({ credential }) {
-					const apiKey = credential?.key;
-					return apiKey === void 0 || apiKey.length === 0 ? void 0 : {
-						auth: { apiKey },
-						source: "XD Pool"
-					};
-				}
-			} },
-			models: buildModels(),
-			api: openAICompletionsApi()
-		}),
-		getModels: () => buildModels()
+	/**
+	* Build a fresh provider object whose `models` array is the catalog as it
+	* stands right now.
+	*
+	* `createProvider` snapshots the array it is handed, and the host keeps its
+	* own snapshot keyed by the IDENTITY of the `profiles` map
+	* (`dsh-llm-pi-ai/lib/index.js` `PiAiAdapter.current()`: it rebuilds the
+	* model collection only when `this.snapshot.profiles !== profiles`, then
+	* calls `models.setProvider(profile.piProvider)`). Reusing one provider object
+	* across `invalidate()` therefore could not work: the host would recognise the
+	* unchanged profiles map, hand back its cached snapshot, and re-collect the
+	* very same provider — so a model list that changed upstream (a new model
+	* released, an account's entitlements changing) stayed frozen until restart.
+	* That is what users saw on the international gateway. Returning a brand-new
+	* provider from a brand-new profile is the only lever the adapter is given.
+	*/
+	const makeProvider = () => {
+		return {
+			...createProvider({
+				id: providerId,
+				name: displayName,
+				auth: { apiKey: {
+					name: "XD Pool loopback secret",
+					async resolve({ credential }) {
+						const apiKey = credential?.key;
+						return apiKey === void 0 || apiKey.length === 0 ? void 0 : {
+							auth: { apiKey },
+							source: "XD Pool"
+						};
+					}
+				} },
+				models: buildModels(),
+				api: openAICompletionsApi()
+			}),
+			getModels: () => buildModels()
+		};
 	};
-	const profile = {
+	const makeProfile = (piProvider) => ({
 		provider: providerId,
 		displayName,
 		streamIdleTimeoutMs: WORKBUDDY_STREAM_IDLE_TIMEOUT_MS,
@@ -11800,8 +12065,9 @@ function createWorkBuddyAdapter(options) {
 		configuredMaxTokens: /* @__PURE__ */ new Map(),
 		modelErrors: /* @__PURE__ */ new Map(),
 		...REQUEST_IMAGE_BUDGETS,
-		piProvider: provider
-	};
+		piProvider
+	});
+	const profile = makeProfile(makeProvider());
 	let profiles = /* @__PURE__ */ new Map([[providerId, profile]]);
 	return {
 		providerId,
@@ -11815,7 +12081,7 @@ function createWorkBuddyAdapter(options) {
 		}),
 		buildModels,
 		invalidate: () => {
-			profiles = /* @__PURE__ */ new Map([[providerId, profile]]);
+			profiles = /* @__PURE__ */ new Map([[providerId, makeProfile(makeProvider())]]);
 		}
 	};
 }
@@ -11968,7 +12234,7 @@ function readPackAccountPool(ctx) {
 */
 async function readPackCredentials(pool) {
 	if (pool === void 0) return [];
-	const entries = pool.listAllAccounts().filter((entry) => {
+	const entries = (await pool.listAllAccounts()).filter((entry) => {
 		const provider = text$3(entry.provider);
 		return provider !== void 0 && PACK_REGION_CHANNELS.includes(provider);
 	});
@@ -13330,6 +13596,24 @@ function freeFetch(url, init = {}) {
 	});
 }
 /**
+* Add the one fact a 429 on this lane can never convey on its own.
+*
+* The free tier meters quota PER UPSTREAM SESSION, not per key and not per
+* account — the gateway's own sentence is "the provider asked to wait 74558s",
+* i.e. twenty hours, which reads like a ban on the whole provider. It is not:
+* a different conversation maps to a different session and is served
+* immediately. Saying so costs one clause and turns a message that looks like
+* "this lane is over" into one that tells the user what to do next.
+*
+* Only a rate class gets the hint: a dead credential or a missing model has
+* nothing to do with the session budget, and adding the clause there would
+* invite the user to retry something that cannot succeed.
+*/
+function withSessionQuotaHint(error, retryAfterMs) {
+	if (!(error instanceof AggregateError) || error.kind !== "soft_rate" || retryAfterMs === void 0 || error.message.includes("per conversation")) return error;
+	return new AggregateError(error.kind, `${error.message} — this quota is counted per conversation on the free tier, so a new conversation (or another model) starts with its own budget`, error.status);
+}
+/**
 * POST one streaming turn and hand back chat-shaped deltas.
 *
 * The shape of the ANSWER decides how it is read, never its Content-Type: a front
@@ -13373,7 +13657,7 @@ async function postFreeStreamed({ modelId, body, session, requestId, hooks, plan
 	const retryAfter = retryAfterMs(response);
 	if (!response.ok) {
 		const refusal = await response.text().catch(() => "");
-		throw classifyUpstream(response.status, refusal, retryAfter, "");
+		throw withSessionQuotaHint(classifyUpstream(response.status, refusal, retryAfter, ""), retryAfter);
 	}
 	if (response.body === null) throw new AggregateError("server", "the free lane returned no body", response.status);
 	let events;
@@ -13406,7 +13690,7 @@ async function postFreeStreamed({ modelId, body, session, requestId, hooks, plan
 		} catch {
 			continue;
 		}
-		if (isRecord$1(decoded) && (decoded.error !== void 0 || decoded.type === "error")) throw classifyUpstream(response.status, event, retryAfter, "");
+		if (isRecord$1(decoded) && (decoded.error !== void 0 || decoded.type === "error")) throw withSessionQuotaHint(classifyUpstream(response.status, event, retryAfter, ""), retryAfter);
 		const translated = chatChunksFor(wire, decoded, tools);
 		for (const delta of translated.frames) hooks.onDelta(delta);
 		if (translated.usage !== void 0) usage = translated.usage;
@@ -18204,6 +18488,7 @@ function createWorkBuddyShim(options) {
 		}
 		const attempts = [];
 		let last;
+		const conversationKey = conversationKeyOf(raw);
 		for (const member of members) {
 			if (signal.aborted) return;
 			attempts.push(member.modelId);
@@ -18218,8 +18503,8 @@ function createWorkBuddyShim(options) {
 					};
 					continue;
 				}
-				const prepared = client.prepareChatBody(raw);
-				const result = await client.chatStream(account.credential, prepared, signal);
+				const prepared = client.prepareChatBodyFor(raw, region ?? "cn");
+				const result = await client.chatStream(account.credential, prepared, signal, conversationKey);
 				if (result.ok) {
 					const turnId = randomBytes(8).toString("hex");
 					await serveSuccessfulStream(res, account, result, logger, refreshBalance, pool, usageRecorder === void 0 ? void 0 : {
@@ -18342,7 +18627,7 @@ function createWorkBuddyShim(options) {
 			writeOpenAIError(res, 503, `${lane}_unavailable`, lane === "free" ? `${modelId} is not available right now; the free lane's model listing is refreshed on the XD Pool card` : `${modelId} is not available right now; sign in to the co-paid lane on the XD Pool card`);
 			return;
 		}
-		const prepared = client.prepareChatBody(raw);
+		const prepared = client.prepareChatBodyFor(raw, region ?? "cn");
 		const conversationKey = conversationKeyOf(raw);
 		const tried = [];
 		let last;
@@ -18381,7 +18666,7 @@ function createWorkBuddyShim(options) {
 				return;
 			}
 			tried.push(account.label);
-			const result = await client.chatStream(account.credential, prepared, controller.signal);
+			const result = await client.chatStream(account.credential, prepared, controller.signal, conversationKey);
 			if (result.ok) {
 				await serveSuccessfulStream(res, account, result, logger, refreshBalance, pool, usageContext);
 				return;
@@ -18639,8 +18924,8 @@ async function recoverFromContextOverrun(options) {
 					...parsed,
 					stream: true,
 					max_tokens: Math.max(256, Math.floor(budget / 2))
-				}, request);
-				return await client.completeChat(summariser.credential, body, signal ?? controller.signal);
+				}, request, region ?? "cn");
+				return await client.completeChat(summariser.credential, body, signal ?? controller.signal, conversationKeyOf(raw));
 			} }, controller.signal);
 			compacted = outcome.messages;
 			summary = outcome.summary;
@@ -18654,7 +18939,7 @@ async function recoverFromContextOverrun(options) {
 		ok: false,
 		detail: compactionDetail === "" ? "compaction could not reduce the prompt" : compactionDetail
 	};
-	const retryBody = client.buildChatBody(parsed, compacted);
+	const retryBody = client.buildChatBody(parsed, compacted, region ?? "cn");
 	const tried = [];
 	for (let attempt = 0; attempt < maxAttempts; attempt += 1) {
 		if (controller.signal.aborted) return {
@@ -18667,7 +18952,7 @@ async function recoverFromContextOverrun(options) {
 			detail: "no account available after compaction"
 		};
 		tried.push(account.label);
-		const result = await client.chatStream(account.credential, retryBody, controller.signal);
+		const result = await client.chatStream(account.credential, retryBody, controller.signal, conversationKeyOf(raw));
 		if (!result.ok) usage?.recorder.record({
 			region: usage.region,
 			model: usage.model,
@@ -20415,6 +20700,19 @@ function apply(ctx, config = {}) {
 				adaptersByRegion.global.invalidate();
 				adaptersByRegion.aggregate.invalidate();
 				adaptersByRegion.free.invalidate();
+				for (const [handle, routes] of [
+					[releaseAdapterCn, [POOL_PROVIDER_BY_REGION.cn]],
+					[releaseAdapterGlobal, [POOL_PROVIDER_BY_REGION.global]],
+					[releaseAdapterAggregate, [XDPOOL_AGGREGATE_PROVIDER]],
+					[releaseAdapterFree, [XDPOOL_FREE_PROVIDER]]
+				]) {
+					if (handle === void 0) continue;
+					try {
+						handle.replace([...routes]);
+					} catch (error) {
+						ctx.logger.warn("dsh-xdpool: could not re-announce adapter routes", error);
+					}
+				}
 			};
 			aggregateCatalog.onChange(() => {
 				invalidateCatalog();

package/package.json

@@ -2,7 +2,7 @@
   "name": "dsh-xdpool",
   "displayName": "DSH XD Pool",
   "description": "One DSH model pool for every free quota you have: merge all locally signed-in WorkBuddy accounts with auto-failover, sign in to 13 vendors by QR/web, aggregate 76 vendors' free tiers, plus a usage dashboard, model routing, a local OpenAI-compatible gateway and a points-automation suite.",
-  "version": "2.3.0",
+  "version": "2.4.1",
   "license": "MIT",
   "author": "XDTrees",
   "homepage": "https://github.com/XDTrees/dsh-xdpool#readme",

package/vendor/channel-pack/pack.js

@@ -34239,7 +34239,11 @@ var CODEBUDDY_FALLBACK_MODELS = [
     name: "GLM-5.1",
     contextWindow: 2e5,
     supportsImages: true,
-    reasoningEfforts: ["medium"],
+    // Measured 2026-10-21 against the live gateway: all five levels answered
+    // 200 with a distinct reasoning_content length per level. The single level
+    // this used to declare was a guess, and it hid four working levels from
+    // the picker.
+    reasoningEfforts: ["low", "medium", "high", "xhigh", "max"],
     maxOutputTokens: 48e3
   },
   {
@@ -34247,7 +34251,11 @@ var CODEBUDDY_FALLBACK_MODELS = [
     name: "GLM-5V-Turbo",
     contextWindow: 2e5,
     supportsImages: true,
-    reasoningEfforts: ["medium"],
+    // Measured 2026-10-21 against the live gateway: all five levels answered
+    // 200 with a distinct reasoning_content length per level. The single level
+    // this used to declare was a guess, and it hid four working levels from
+    // the picker.
+    reasoningEfforts: ["low", "medium", "high", "xhigh", "max"],
     maxOutputTokens: 64e3
   },
   {
@@ -34255,7 +34263,11 @@ var CODEBUDDY_FALLBACK_MODELS = [
     name: "Kimi-K3-1",
     contextWindow: 1e6,
     supportsImages: true,
-    reasoningEfforts: ["medium"],
+    // Measured 2026-10-21 against the live gateway: all five levels answered
+    // 200 with a distinct reasoning_content length per level. The single level
+    // this used to declare was a guess, and it hid four working levels from
+    // the picker.
+    reasoningEfforts: ["low", "medium", "high", "xhigh", "max"],
     maxOutputTokens: 32e3
   },
   {
@@ -34274,7 +34286,11 @@ var CODEBUDDY_FALLBACK_MODELS = [
     name: "Kimi-K2.7",
     contextWindow: 256e3,
     supportsImages: true,
-    reasoningEfforts: ["medium"],
+    // Measured 2026-10-21 against the live gateway: all five levels answered
+    // 200 with a distinct reasoning_content length per level. The single level
+    // this used to declare was a guess, and it hid four working levels from
+    // the picker.
+    reasoningEfforts: ["low", "medium", "high", "xhigh", "max"],
     maxOutputTokens: 32e3
   },
   {
@@ -34282,7 +34298,11 @@ var CODEBUDDY_FALLBACK_MODELS = [
     name: "Kimi-K2.6",
     contextWindow: 256e3,
     supportsImages: true,
-    reasoningEfforts: ["medium"],
+    // Measured 2026-10-21 against the live gateway: all five levels answered
+    // 200 with a distinct reasoning_content length per level. The single level
+    // this used to declare was a guess, and it hid four working levels from
+    // the picker.
+    reasoningEfforts: ["low", "medium", "high", "xhigh", "max"],
     maxOutputTokens: 32e3
   },
   {
@@ -34290,7 +34310,11 @@ var CODEBUDDY_FALLBACK_MODELS = [
     name: "MiniMax-M3",
     contextWindow: 512e3,
     supportsImages: true,
-    reasoningEfforts: ["medium"],
+    // Measured 2026-10-21 against the live gateway: all five levels answered
+    // 200 with a distinct reasoning_content length per level. The single level
+    // this used to declare was a guess, and it hid four working levels from
+    // the picker.
+    reasoningEfforts: ["low", "medium", "high", "xhigh", "max"],
     maxOutputTokens: 64e3
   }
 ];
@@ -34316,8 +34340,8 @@ var WORKBUDDY_FALLBACK_MODELS = [
   // （`node scripts/dump-max-output.mjs`）。该值就是用户在 IDE 里实际拿到的
   // 单次输出额度，远端不可用时由本表顶替。远端未下发的模型保持 undefined。
   { id: "default-model", name: "Auto", contextWindow: 176e3, supportsImages: true, maxOutputTokens: 24e3 },
-  { id: "fast-model", name: "Fast", contextWindow: 2e5, supportsImages: true, reasoningEfforts: ["medium"], maxOutputTokens: 32e3 },
-  { id: "balanced-model", name: "Balanced", contextWindow: 256e3, supportsImages: true, reasoningEfforts: ["medium"], maxOutputTokens: 32e3 },
+  { id: "fast-model", name: "Fast", contextWindow: 2e5, supportsImages: true, reasoningEfforts: ["low", "medium", "high", "xhigh", "max"], maxOutputTokens: 32e3 },
+  { id: "balanced-model", name: "Balanced", contextWindow: 256e3, supportsImages: true, reasoningEfforts: ["low", "medium", "high", "xhigh", "max"], maxOutputTokens: 32e3 },
   { id: "primary-model", name: "Primary", contextWindow: 272e3, supportsImages: true, reasoningEfforts: ["high"], maxOutputTokens: 72e3 },
   { id: "deep-model", name: "Deep", contextWindow: 176e3, supportsImages: true, maxOutputTokens: 24e3 },
   {
@@ -34349,14 +34373,14 @@ var WORKBUDDY_FALLBACK_MODELS = [
     defaultReasoningEffort: "high",
     maxOutputTokens: 64e3
   },
-  { id: "deepseek-v4.1-flash", name: "Deepseek-V4.1-Flash", contextWindow: 1e6, supportsImages: true, reasoningEfforts: ["high"], defaultReasoningEffort: "high", maxOutputTokens: 128e3 },
+  { id: "deepseek-v4.1-flash", name: "Deepseek-V4.1-Flash", contextWindow: 1e6, supportsImages: true, reasoningEfforts: ["low", "medium", "high", "xhigh", "max"], defaultReasoningEffort: "high", maxOutputTokens: 128e3 },
   {
     // 2026-09 补录：新加坡区的同代模型（-sg 后缀），远端下发且实测能看图。
     id: "deepseek-v4.1-flash-sg",
     name: "Deepseek-V4.1-Flash",
     contextWindow: 1e6,
     supportsImages: true,
-    reasoningEfforts: ["high"],
+    reasoningEfforts: ["low", "medium", "high", "xhigh", "max"],
     defaultReasoningEffort: "high",
     maxOutputTokens: 128e3
   },
@@ -34416,7 +34440,11 @@ var WORKBUDDY_FALLBACK_MODELS = [
   },
   // gpt-5.3-codex：远端未下发 maxOutputTokens，故不填（保持 undefined，
   // 交由网关默认），不臆造数值。
-  { id: "gpt-5.3-codex", name: "GPT-5.3-Codex", contextWindow: 272e3, supportsImages: true, reasoningEfforts: ["medium"] },
+  // gpt-5.3-codex answered 200 at every level, but returned an EMPTY
+  // reasoning_content at all five — the levels are legal yet not observably
+  // distinct. Declaring the ladder states what may be sent; it does not
+  // claim the model thinks.
+  { id: "gpt-5.3-codex", name: "GPT-5.3-Codex", contextWindow: 272e3, supportsImages: true, reasoningEfforts: ["low", "medium", "high", "xhigh", "max"] },
   { id: "gemini-3.5-flash", name: "Gemini-3.5-Flash", contextWindow: 1e6, supportsImages: true, reasoningEfforts: ["medium"], maxOutputTokens: 65536 },
   {
     id: "glm-5.3",
@@ -34436,7 +34464,7 @@ var WORKBUDDY_FALLBACK_MODELS = [
     defaultReasoningEffort: "high",
     maxOutputTokens: 48e3
   },
-  { id: "kimi-k3", name: "Kimi-K3", contextWindow: 1e6, supportsImages: true, reasoningEfforts: ["medium"], maxOutputTokens: 32e3 },
+  { id: "kimi-k3", name: "Kimi-K3", contextWindow: 1e6, supportsImages: true, reasoningEfforts: ["low", "medium", "high", "xhigh", "max"], maxOutputTokens: 32e3 },
   {
     // 2026-09 补录：远端 /v3/config 的 cli agent 白名单里有它，兜底表原先漏了。
     id: "kimi-k2.8-preview",
@@ -34447,7 +34475,7 @@ var WORKBUDDY_FALLBACK_MODELS = [
     defaultReasoningEffort: "high",
     maxOutputTokens: 32e3
   },
-  { id: "kimi-k2.6", name: "Kimi-K2.6", contextWindow: 256e3, supportsImages: true, reasoningEfforts: ["medium"], maxOutputTokens: 32e3 }
+  { id: "kimi-k2.6", name: "Kimi-K2.6", contextWindow: 256e3, supportsImages: true, reasoningEfforts: ["low", "medium", "high", "xhigh", "max"], maxOutputTokens: 32e3 }
 ];
 var WORKBUDDY = {
   id: "workbuddy",