@mars-sea/dsh-commandcode-provider 0.16.2 → 0.16.4

package/CHANGELOG.md

@@ -4,6 +4,27 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.16.4] - 2026-11-02
+
+### Added
+
+- **新增免费模型 Ling 3.1 Flash 1019（`inclusionai/ling-3.1-flash-1019:free`，服务端目录先行）。** `?view=client` 的 `catalogVersion` 由 `7f405e24…` 变为 `f3a51355…`，默认目录 94 → 95、client 视图 106 → 107，唯一新增行是 Ling 3.1 Flash 1019：1,000,000 上下文、262,144 输出上限、纯文本输入、推理档位 `low`／`medium`／`high`、Go 档、ZDR 覆盖、`/chat/completions` 端点；100% 免费促销自 `2026-10-29T16:00Z` 起至 `2026-11-03T16:00Z`（北京时间 11-04 00:00 到期），每账号每天 100 次请求。官方 models、pricing 与 go／goat／pro／max 计划页均已上架；`1.88.2` 产物注册表尚未收录该行（审计报 `absent from CLI registry`，与 `gpt-6.1-terra` 的目录先行形态一致）。快照 +1：`KNOWN_PLANS` 94 → 95、`KNOWN_EFFORTS` 67 → 68、`KNOWN_DEALS` 4 → 5（带 `expiresAt`，到期后 FREE 徽标自动消失，不跟随下架）；`KNOWN_IMAGE_MODELS`（73）、`KNOWN_THINKING_MODELS`（16）、`KNOWN_NON_ZDR_MODELS`（18）、订阅表（9）、峰值定价（5）、内置价格行（94）与输出上限表（79）均不变——免费行照旧不进价格表，`MODEL_OUTPUT_TOKEN_LIMITS` 也不含 Ling 行（走 chat/completions）。`COMMAND_CODE_CLI_VERSION` 保持 `1.88.2`。**验证**：`typecheck` 退出码 0，`npm test` 1156／1156（扩展既有快照断言，未新增用例），`build` 两段 `Build complete`；本轮仅公开 GET 采集（14 个源全部 HTTP 200），未做真实账号登录、付费生成与服务端协议验证。
+
+### Fixed
+
+- **账户余额为 0 时提示指向错误方向（2026-10-31 实测）。** 上游对 `POST /provider/v1/chat/completions` 返回 `400`，正文为 `You have insufficient credits to make this request…`、结构化代码只有 `BAD_REQUEST`；插件已能据此把账号轮换掉，但单账号场景没有可换对象，最终只显示通用文案「服务端拒绝了本次请求，请检查模型、套餐和请求设置」——按这三条查一条也查不到点上。现在该形状在通用分支之前被识别，提示为「账户额度不足，无法完成本次请求——请充值额度，或切换到其他账号」。判定复用账号轮换的同一组 pattern（新增 `isInsufficientCreditsText()` 与 `INSUFFICIENT_CREDITS`／`PREMIUM_CREDITS_EXHAUSTED` 代码集），两处不会漂移；失败码仍取 `httpErrorCode(status)`，重试预算与轮换行为不变。**验证**：`typecheck` 退出码 0，`npm test` 1156／1156（新增 3 条：实测正文、仅结构化代码、普通 400 不被吞）。
+
+## [0.16.3] - 2026-10-31
+
+### Changed
+
+- **同步 Command Code CLI `1.88.1` → `1.88.2`（纯请求头同步）。** npm `latest` 于 `2026-10-29T02:05Z` 前进到 `1.88.2`。CLI 的变更日志在 npm 包内 `CHANGELOG.md`（官网 `/changelog` 只覆盖 Desktop 应用），该版本只有一条 patch：「修复全新安装或更新后启动崩溃」。产物级比对：包内 80 个文件只有 `CHANGELOG.md`、`dist/cli.mjs`、`package.json`（仅版本号）与 VSIX 变化，依赖零变化；请求路径（193）、自定义请求头（13）、环境变量（48）集合相同，`/alpha/generate` 调用点在屏蔽混淆标识符后相似度 1.0，`promptCache`（5）／`reasoning_effort`（7）／`max_tokens`（12）／`cache_control`（1）计数相同，审计 `cliDiff`（模型、档位、订阅、ZDR 例外集）为空。服务端目录未动（`?view=client` 的 `catalogVersion` 仍 `7f405e24…`，默认 94、client 106 行），所有能力快照零变化，`COMMAND_CODE_CLI_VERSION` 只随请求头更新，运行时行为不变。
+- **同步定价页两条模型额度（`planAllowanceUsd`）。** 官网把 DeepSeek V4.1 Flash 与 GLM-5.3 Flash 的加成月度额度从 GOAT `$80`／Pro `$90` 上调到 GOAT `$100`／Pro `$110`（Go `$10` 不变），per-token 价格、Peak 时段与套餐可用性均未变；设置页「模型额度」随之显示新值。价格表由 `node scripts/sync-model-prices.mjs` 重新生成，全表只有这两行变化。GLM-5.3 Flash 仍标注 `while capacity lasts`；V4.1 Flash 的文案从「with no end date」改为「comes with bigger monthly credits」，官方未给结束时间，若之后回调按同一脚本再同步。**验证**：`typecheck` 退出码 0，`npm test` 1142／1142，`build` 两段 `Build complete`；未做真实账号登录与付费生成验证。
+
+### Fixed
+
+- **探活丢路径前缀、断连与保存失败提示不可诊断（#111、#110）。** 探活原先用 `new URL('/', apiBase)` 探 API base 根，会丢掉带路径前缀的反向代理前缀（如 `http://127.0.0.1:18797/bili/https://api.commandcode.ai`），状态行必然误报「本机无法连接」；现在改探真实端点 `${apiBase}/alpha/whoami` 并用与真实请求相同的字符串拼接保留前缀，未认证的 401/403 计为「可达」（探活刻意不带凭据），404/429/5xx 仍判异常，超时与传输失败继续分开。用量报告的每条端点失败在原文后追加中文原因与目标 `scheme://host:port`（映射 fetch `error.cause` 的 `ECONNREFUSED`／`ENOTFOUND`／连接超时等，原始错误码入括号），反代换端口、代理未启动与 API 故障因此可区分。设置保存的「配置提交未获确认」不再固定说「密钥未写入」：只有本次提交带密钥时才提密钥，否则只报配置未写入；宿主拒绝的原文（如 profile 写锁超时）以「宿主返回：…」附在提示后，折叠为一行并限长 240 字符。**验证**：`typecheck` 退出码 0，`npm test` 1153／1153，`build` 两段 `Build complete`；另在隔离 `DSH_HOME` + 真实宿主（`dsh 0.2.2-alpha.2`）+ 带路径前缀反代 + 无头 Chrome 下端到端复现：反代在线时状态行「本机 连通正常 · 781 ms」（旧探测目标反代根 `/` 返回 404），停反代后「本机无法连接」且失败详情带目标与 `ECONNREFUSED`，写锁超时保存提示带宿主原文；再用真实密钥复验：只读接口全部 200、状态行「连通正常」、真实生成 `POST /alpha/generate` 返回 200（会话费用 `≈$0.0015` 与上游扣减 `0.001471284` 一致）；最后复验浏览器登录链路：隔离宿主用凭据库中不存在的引用起步（界面「未配置」），点「网页登录」后按 Studio 回调契约提交真实密钥，回调 `200 {"success":true}`、校验请求经反代 `/alpha/whoami` 200、凭据落库且界面转为「已登录为 mars-sea」。Studio 页面本身的账号登录与 Cloudflare 挑战需真人完成，不在自动验证范围内。
+
 ## [0.16.2] - 2026-10-30
 
 ### Changed

package/lib/client.js

@@ -76,6 +76,13 @@ window.__ModuleLoader__.load({
 		}
 		//#endregion
 		//#region src/client/settings-scope.ts
+		/** 组装报告；没有原文时不带 `message` 键（`exactOptionalPropertyTypes`）。 */
+		function mutationReport(result, message) {
+			return message === void 0 ? { result } : {
+				result,
+				message
+			};
+		}
 		/**
 		* Backoff before re-reading a directory whose FIRST read failed, in millis.
 		*
@@ -328,19 +335,20 @@ window.__ModuleLoader__.load({
 					const response = await this.mutateRemote(ownedOps, revision);
 					if (!response.ok) {
 						await this.recover(generation);
-						return response.error?.message?.includes("SETTINGS_CONFLICT") ? "conflict" : "failed";
+						const message = response.error?.message;
+						return message?.includes("SETTINGS_CONFLICT") ? mutationReport("conflict", message) : mutationReport("failed", message);
 					}
 					const row = response.value;
 					if (row?.ns !== this.namespace || decodeRow(row.value) === void 0 || typeof row.revision !== "number") {
 						await this.recover(generation);
-						return "failed";
+						return mutationReport("failed");
 					}
-					if (this.disposed) return "accepted";
+					if (this.disposed) return mutationReport("accepted");
 					if (generation === this.writeGeneration) {
 						this.pendingRevision = void 0;
 						this.mirror.acceptView(row);
 					} else this.pendingRevision = typeof row.revision === "number" ? row.revision : void 0;
-					return "accepted";
+					return mutationReport("accepted");
 				});
 			}
 			async mutateRemote(ops, revision) {
@@ -406,9 +414,9 @@ window.__ModuleLoader__.load({
 			}
 			/** Queue operations one at a time; disposal makes the queue inert. */
 			enqueue(operation) {
-				if (this.disposed) return Promise.resolve("cancelled");
+				if (this.disposed) return Promise.resolve(mutationReport("cancelled"));
 				const task = this.tail.then(async () => {
-					if (this.disposed) return "cancelled";
+					if (this.disposed) return mutationReport("cancelled");
 					return operation();
 				});
 				this.tail = task.then(() => {}, () => {});
@@ -780,22 +788,26 @@ window.__ModuleLoader__.load({
 				return success;
 			}
 			async commit(ops, baseline) {
-				if (this.closed) return "cancelled";
+				if (this.closed) return { result: "cancelled" };
 				const snapshot = this.scope.getSnapshot();
-				if (!snapshot.writable || this.scope.mutate === void 0) return "failed";
-				if (baseline && [...baseline].some(([field, before]) => !sameValue(before, fieldBaseline(snapshot, field)))) return "conflict";
+				if (!snapshot.writable || this.scope.mutate === void 0) return { result: "failed" };
+				if (baseline && [...baseline].some(([field, before]) => !sameValue(before, fieldBaseline(snapshot, field)))) return { result: "conflict" };
 				try {
 					return await this.scope.mutate(ops, snapshot.revision);
-				} catch {
-					return "failed";
+				} catch (error) {
+					return {
+						result: "failed",
+						message: readableErrorText(error)
+					};
 				}
 			}
 			async configure(outcome, ops, baseline) {
 				if (ops.length === 0) return true;
 				const confirmation = await this.commit(ops, baseline);
-				outcome.config = confirmation === "accepted" ? "confirmed" : confirmation === "conflict" ? "conflict" : "unconfirmed";
-				if (confirmation === "cancelled" || this.closed) outcome.issue = "closed";
-				return confirmation === "accepted" && !this.closed;
+				outcome.config = confirmation.result === "accepted" ? "confirmed" : confirmation.result === "conflict" ? "conflict" : "unconfirmed";
+				if (confirmation.result === "failed" && confirmation.message !== void 0) outcome.hostError = confirmation.message;
+				if (confirmation.result === "cancelled" || this.closed) outcome.issue = "closed";
+				return confirmation.result === "accepted" && !this.closed;
 			}
 			async changeCredential(outcome, ref, value, defaultRef) {
 				outcome.credential = "not-issued";
@@ -963,9 +975,15 @@ window.__ModuleLoader__.load({
 					path: ["credentialCleanupRefs"],
 					value: this.cleanupRefs().filter((entry) => entry !== ref)
 				}])) {
-					if (outcome.config === "none") outcome.config = cleanup.config;
+					if (outcome.config === "none") {
+						outcome.config = cleanup.config;
+						if (cleanup.hostError !== void 0) outcome.hostError = cleanup.hostError;
+					}
 					outcome.issue = cleanup.issue ?? "cleanup-pending";
-				} else if (outcome.config === "none") outcome.config = cleanup.config;
+				} else if (outcome.config === "none") {
+					outcome.config = cleanup.config;
+					if (cleanup.hostError !== void 0) outcome.hostError = cleanup.hostError;
+				}
 			}
 		};
 		//#endregion
@@ -3243,7 +3261,9 @@ window.__ModuleLoader__.load({
 			saved: "设置已保存",
 			saveFailed: "保存失败，请重试。",
 			writeConflict: "待提交字段已被其他操作修改，配置未写入；请核对后再保存。",
-			writeConfigUnconfirmed: "配置提交未获确认；密钥未写入，请核对宿主状态后重试。",
+			writeConfigUnconfirmed: "配置提交未获确认，未写入宿主；请核对后重试。",
+			writeConfigUnconfirmedKey: "配置与密钥均未写入宿主；请核对后重试。",
+			writeHostError: "宿主返回：{message}",
 			writeConfigSaved: "配置已保存。",
 			writeKeyNotIssued: "密钥操作尚未发出；请核对后重试。",
 			writeKeyUnconfirmed: "密钥操作未获确认；请核对后手动重试。",
@@ -3514,7 +3534,9 @@ window.__ModuleLoader__.load({
 			saved: "Settings saved",
 			saveFailed: "Save failed, please retry.",
 			writeConflict: "Pending fields were modified by another operation; the configuration was not written. Please review and save again.",
-			writeConfigUnconfirmed: "The configuration submission was not confirmed; the key was not written. Please check the host state and retry.",
+			writeConfigUnconfirmed: "The configuration submission was not confirmed and was not written. Please review and retry.",
+			writeConfigUnconfirmedKey: "Neither the configuration nor the key was written. Please review and retry.",
+			writeHostError: "Host returned: {message}",
 			writeConfigSaved: "Configuration saved.",
 			writeKeyNotIssued: "The key operation has not been issued; please review and retry.",
 			writeKeyUnconfirmed: "The key operation was not confirmed; please review and retry manually.",
@@ -3699,7 +3721,7 @@ window.__ModuleLoader__.load({
 			else if (result.issue === "target-changed") key = result.credential === "not-issued" ? "writeTargetBefore" : "writeTargetAfter";
 			else if (result.issue === "cleanup-pending") key = "writeCleanupPending";
 			else if (result.config === "conflict") key = "writeConflict";
-			else if (result.config === "unconfirmed") key = "writeConfigUnconfirmed";
+			else if (result.config === "unconfirmed") key = result.credential === "not-issued" ? "writeConfigUnconfirmedKey" : "writeConfigUnconfirmed";
 			else if (result.credential === "unconfirmed") key = "writeKeyUnconfirmed";
 			else if (result.credential === "not-issued") key = "writeKeyNotIssued";
 			const messages = key ? [t(key)] : [];
@@ -3709,8 +3731,18 @@ window.__ModuleLoader__.load({
 			}
 			if (key && result.config === "confirmed" && result.issue !== "cleanup-pending") messages.unshift(t("writeConfigSaved"));
 			if (result.refreshFailed) messages.push(t("writeRefreshFailed"));
+			const hostError = result.hostError;
+			if (hostError !== void 0 && hostError.trim() !== "") messages.push(t("writeHostError").replace("{message}", () => hostErrorText(hostError)));
 			return messages.join(" ");
 		}
+		/**
+		* 宿主原文压成一行并限长：锁超时、写权限这类诊断足够；换行会撑高提示区，
+		* 超长正文（整页 HTML、大对象）也不该被原样铺到设置页上。
+		*/
+		function hostErrorText(message) {
+			const oneLine = message.replace(/\s+/g, " ").trim();
+			return oneLine.length > 240 ? `${oneLine.slice(0, 240)}…` : oneLine;
+		}
 		//#endregion
 		//#region src/plan-tiers.ts
 		/** Shared plan-tier presentation facts; safe to inline into the browser bundle. */
@@ -3864,7 +3896,7 @@ window.__ModuleLoader__.load({
 		}
 		//#endregion
 		//#region package.json
-		var version = "0.16.2";
+		var version = "0.16.4";
 		var repository = {
 			"type": "git",
 			"url": "git+https://github.com/Mars-Sea/dsh-commandcode-provider.git"

package/lib/index.d.ts

@@ -61,7 +61,7 @@ type CommandCodeProtocol = 'cli' | 'openai' | 'messages';
 declare const DEFAULT_MAX_OUTPUT_TOKENS = 131072;
 //#endregion
 //#region src/adapter.d.ts
-declare const COMMAND_CODE_CLI_VERSION = "1.88.1";
+declare const COMMAND_CODE_CLI_VERSION = "1.88.2";
 declare const DEFAULT_API_BASE = "https://api.commandcode.ai";
 /**
  * The output budget this bundle asks for, and the ceiling it will never exceed.
@@ -1232,7 +1232,7 @@ declare const KNOWN_THINKING_MODELS: ReadonlySet<string>;
  * `/docs/plans/max` and `/docs/resources/pricing-limits`). Each plan's model
  * list is a superset of the one below it: Go ⊂ GOAT ⊂ Pro ⊂ Provider/Max.
  * Models absent from every plan list (Claude Opus/Fable, Fugu Ultra) are
- * Provider-tier. Re-verified at command-code@1.88.1 (2026-10-28) against the
+ * Provider-tier. Re-verified at command-code@1.88.2 (2026-10-31) against the
  * live `?view=client` catalog: 106 ids, still a strict superset chain. Mistral
  * Large 4 (1.75.0) was the only addition at the 1.77.0 check; the 1.78.0/1.79.0 pair
  * added Claude Haiku 5.5 at Pro and the free Glyph Cluster preview at Go.

package/lib/index.js

@@ -123,7 +123,7 @@ const PLAN_ORDER = Object.fromEntries(PLAN_TIER_ORDER.map((tier, index) => [tier
 * subscription-plan labels, deals, and hourly (peak/off-peak) pricing.
 *
 * Everything here is synced from official sources — the command-code CLI
-* bundle's model table (`dist/cli.mjs`, re-verified at command-code@1.88.1), the
+* bundle's model table (`dist/cli.mjs`, re-verified at command-code@1.88.2), the
 * live model catalog (`GET /provider/v1/models?view=client`, the authority for
 * capability fields since 1.82.4), and the official plan/pricing/model docs; see
 * the dsh-commandcode-upstream skill for the extraction procedures. The
@@ -145,7 +145,11 @@ const PLAN_ORDER = Object.fromEntries(PLAN_TIER_ORDER.map((tier, index) => [tier
 * step and `$2/$7.5` beyond. The 2026-10-28 check moves the constant to 1.88.1,
 * whose registry now names both catalog-only rows (Kimi K3.1 Pro and GPT-6.1
 * Terra) with efforts and context identical to this snapshot, so no table
-* changes. Keeping the snapshot in its own module
+* changes. The 2026-10-31 check moves it to 1.88.2 — the npm package's own
+* CHANGELOG names its sole change as a launch-crash fix after a fresh install
+* or update — and again leaves every table here identical; that round's only
+* data change is two per-model allowances in ./model-prices.ts. Keeping the
+* snapshot in its own module
 * confines those frequent sync diffs here: src/adapter.ts holds only the stable
 * wire/runtime logic. The read helpers live here too, for the same reason.
 *
@@ -517,6 +521,11 @@ const KNOWN_EFFORTS = {
 		"medium",
 		"high"
 	],
+	"inclusionai/ling-3.1-flash-1019:free": [
+		"low",
+		"medium",
+		"high"
+	],
 	"mistral/mistral-large-4": [
 		"low",
 		"medium",
@@ -948,7 +957,7 @@ const MODEL_OUTPUT_TOKEN_LIMITS = /* @__PURE__ */ new Map([
 * `/docs/plans/max` and `/docs/resources/pricing-limits`). Each plan's model
 * list is a superset of the one below it: Go ⊂ GOAT ⊂ Pro ⊂ Provider/Max.
 * Models absent from every plan list (Claude Opus/Fable, Fugu Ultra) are
-* Provider-tier. Re-verified at command-code@1.88.1 (2026-10-28) against the
+* Provider-tier. Re-verified at command-code@1.88.2 (2026-10-31) against the
 * live `?view=client` catalog: 106 ids, still a strict superset chain. Mistral
 * Large 4 (1.75.0) was the only addition at the 1.77.0 check; the 1.78.0/1.79.0 pair
 * added Claude Haiku 5.5 at Pro and the free Glyph Cluster preview at Go.
@@ -999,6 +1008,7 @@ const KNOWN_PLANS = {
 	"gpt-6-luna": "go",
 	"meituan/LongCat-2.0": "go",
 	"inclusionai/ling-3.1-flash": "go",
+	"inclusionai/ling-3.1-flash-1019:free": "go",
 	"meta/muse-spark-1.2-contributor": "go",
 	"meta/muse-spark-1.3-contributor": "go",
 	"meta/muse-spark-open": "go",
@@ -1203,6 +1213,11 @@ function modelVisibleForAnyAccount(modelId, accounts) {
 }
 const KNOWN_DEALS = {
 	"MiniMaxAI/MiniMax-M3": { label: "50% off" },
+	"inclusionai/ling-3.1-flash-1019:free": {
+		label: "FREE",
+		free: true,
+		expiresAt: "2026-11-03T16:00:00Z"
+	},
 	"poolside/laguna-s-2.1-free": {
 		label: "FREE",
 		free: true
@@ -1651,8 +1666,8 @@ const MODEL_PRICE_ROWS = [
 		],
 		allowance: {
 			go: 10,
-			goat: 80,
-			pro: 90
+			goat: 100,
+			pro: 110
 		}
 	},
 	{
@@ -1852,8 +1867,8 @@ const MODEL_PRICE_ROWS = [
 		],
 		allowance: {
 			go: 10,
-			goat: 80,
-			pro: 90
+			goat: 100,
+			pro: 110
 		}
 	},
 	{
@@ -3934,6 +3949,76 @@ function readableErrorText(error) {
 	}
 	return String(error);
 }
+//#endregion
+//#region src/transport-failure.ts
+/**
+* Transport failure reasons for the usage and settings diagnostics.
+*
+* Node's fetch rejects with `TypeError('fetch failed')` and keeps the real
+* cause one level down (`error.cause`): `ECONNREFUSED` when nothing listens on
+* the port, `ENOTFOUND` on DNS, a connect timeout behind a dead proxy. That
+* message alone cannot tell "the reverse proxy changed port" from "the proxy is
+* not running" from "the API is down", so per-endpoint failure lines and the
+* settings notice append a short Chinese reason plus the raw code (issue #111).
+*
+* Only the code is quoted back, never the system message: those may embed
+* internal paths, and the original message stays available through
+* `readableErrorText` for the detail line. An error with no code yields no
+* reason — the caller still renders what it has instead of an invented guess.
+*
+* @module dsh-commandcode-provider/transport-failure
+*/
+/** 原始错误码 → 界面原因；未列出的码只回显错误码，不做猜测。 */
+const REASONS$1 = {
+	ECONNREFUSED: "连接被拒绝",
+	ECONNRESET: "连接被重置",
+	ENOTFOUND: "域名解析失败",
+	EAI_AGAIN: "域名解析临时失败",
+	EHOSTUNREACH: "主机不可达",
+	ENETUNREACH: "网络不可达",
+	ETIMEDOUT: "连接超时",
+	UND_ERR_CONNECT_TIMEOUT: "连接超时",
+	UND_ERR_SOCKET: "连接中断",
+	UND_ERR_CLOSED: "连接已关闭",
+	CERT_HAS_EXPIRED: "证书已过期",
+	DEPTH_ZERO_SELF_SIGNED_CERT: "自签名证书不受信任",
+	SELF_SIGNED_CERT_IN_CHAIN: "证书链中有自签名证书",
+	UNABLE_TO_VERIFY_LEAF_SIGNATURE: "无法验证证书"
+};
+/** 取错误链上的第一个错误码：先看自身，再看 AggregateError 子项，最后看 cause。 */
+function codeOf(value, depth = 0) {
+	if (depth > 4 || value === null || typeof value !== "object") return void 0;
+	const entry = value;
+	if (typeof entry.code === "string" && entry.code !== "") return entry.code;
+	if (Array.isArray(entry.errors)) for (const nested of entry.errors) {
+		const code = codeOf(nested, depth + 1);
+		if (code !== void 0) return code;
+	}
+	return codeOf(entry.cause, depth + 1);
+}
+/**
+* One diagnostic clause for a transport failure, or undefined when the error
+* carries no recognizable code.
+*/
+function transportFailureReason(error) {
+	const code = codeOf(error);
+	if (code === void 0) return void 0;
+	const reason = REASONS$1[code];
+	return reason === void 0 ? `网络错误（${code}）` : `${reason}（${code}）`;
+}
+/**
+* `scheme://host:port` of a base, with path and query dropped: enough to name
+* the endpoint a failed request aimed at without echoing proxy prefixes, and
+* without any chance of a token riding along in a path.
+*/
+function httpOrigin(base) {
+	try {
+		const url = new URL(base);
+		return `${url.protocol}//${url.host}`;
+	} catch {
+		return;
+	}
+}
 const THROTTLED_CODE = "THROTTLED";
 const boundedCodes = /* @__PURE__ */ new Set([
 	"EMPTY_RESPONSE",
@@ -4758,6 +4843,8 @@ function generateHttpError(status, errText, retryAfterMs, parsed = parseProvider
 	if (status === 422 && isZdrNoProviders(providerCode, providerDetail, errText)) return bilingual("PROVIDER_HTTP_ERROR", `Command Code API error 422 (${detail}): zero data retention is enforced but no ZDR-capable upstream is available for this model — turn the ZDR setting off (or use another model)`, "Command Code API 返回 422：已开启零数据保留（ZDR），但该模型当前没有可用的 ZDR 上游——请关闭 ZDR 开关，或改用其他模型", { status: 422 });
 	if (status === 401) return bilingual("INVALID_CREDENTIAL", `Command Code API error 401 (${detail}): the API key is missing or invalid — check the key stored for COMMANDCODE_API_KEY (Models page) or the auth file`, "Command Code API 返回 401：API 密钥缺失或无效——请在设置页检查 COMMANDCODE_API_KEY 存储的密钥，或检查 auth 文件", { status: 401 });
 	if (status === 413) return bilingual("PROVIDER_HTTP_ERROR", "Command Code API error 413: the request body exceeds the provider's size limit (the cap is undocumented, measured at about 50 MB) — the session history is too large to send, usually because of accumulated image attachments. Start a new session, or drop the image-heavy part of this one, and retry", "Command Code API 返回 413：请求体超过服务端的体积上限（官方未公开，实测约 50 MB）——会话历史过大，通常是历史图片累积所致。请新建会话，或移除本会话中图片较多的部分后重试", { status: 413 });
+	const creditsCode = providerCode?.toUpperCase();
+	if (creditsCode !== void 0 && INSUFFICIENT_CREDITS_CODES.has(creditsCode) || isInsufficientCreditsText(overflowDetail) || isInsufficientCreditsText(errText)) return bilingual(httpErrorCode(status), `Command Code API error ${status}: this account has insufficient credits for the request — top up credits or switch to another account`, `Command Code API 返回 ${status}：账户额度不足，无法完成本次请求——请充值额度，或切换到其他账号`, { status });
 	const safeIdentifier = (value) => value !== void 0 && /^[A-Za-z0-9_.:-]{1,80}$/.test(value) ? value : void 0;
 	const diagnosticCode = safeIdentifier(parsed.code) ?? /^\s*(MODEL_NOT_IN_PLAN|INSUFFICIENT_CREDITS|USAGE_EXCEEDED|PREMIUM_CREDITS_EXHAUSTED)\b/i.exec(parsed.message ?? "")?.[1];
 	const diagnosticType = safeIdentifier(parsed.type);
@@ -4850,6 +4937,30 @@ const ACCOUNT_UNAVAILABLE_CODES = /* @__PURE__ */ new Set([
 	"PREMIUM_CREDITS_EXHAUSTED",
 	"MODEL_NOT_IN_PLAN"
 ]);
+/**
+* 余额用尽的两种结构化代码（`USAGE_EXCEEDED`／`MODEL_NOT_IN_PLAN` 不属于这一类：
+* 前者是窗口额度，后者是套餐不含该模型，用户可行动的方向不同）。
+*/
+const INSUFFICIENT_CREDITS_CODES = /* @__PURE__ */ new Set(["INSUFFICIENT_CREDITS", "PREMIUM_CREDITS_EXHAUSTED"]);
+/**
+* 正文里「额度用尽」的特征。
+*
+* 账号轮换判定与用户可见文案共用这一份 pattern：两处各写一份必然漂移，出现
+* 「轮换认得出、文案认不出」的错位——2026-10-31 的实测就是这样，账号池已经按
+* 这个原因换号，而单账号场景下用户看到的仍是通用 400 提示。
+*
+* 下划线与空格是同一个词：JSON 正文常写 `insufficient_credits`，CLI 自己的
+* marker 列表写 `insufficient credits`，先归一分隔符再匹配。
+*/
+function isInsufficientCreditsText(text) {
+	const lower = text.toLowerCase().replaceAll("_", " ");
+	return lower.includes("insufficient credits") || lower.includes("premium credits exhausted") || /insufficient (credit|balance)/.test(lower) || /out of credit/.test(lower);
+}
+/** 正文里「模型不在本账号套餐内」的特征，与上者同一归一规则。 */
+function isModelOutsidePlanText(text) {
+	const lower = text.toLowerCase().replaceAll("_", " ");
+	return lower.includes("model not in plan") || /not (available|included)[^.]{0,40}plan/.test(lower) || lower.includes("upgrade your plan");
+}
 function classifyAccountRejection(status, errText, parsed = parseProviderError(errText)) {
 	const code = (parsed.code ?? parsed.type)?.toUpperCase();
 	if (status === 429 || code === "RATE_LIMITED") {
@@ -4863,8 +4974,7 @@ function classifyAccountRejection(status, errText, parsed = parseProviderError(e
 	if (status === 401) return { reason: "invalid-credential" };
 	if (code !== void 0 && ACCOUNT_UNAVAILABLE_CODES.has(code)) return { reason: "unavailable" };
 	if (status < 400 || status >= 500) return void 0;
-	const lower = errText.toLowerCase().replaceAll("_", " ");
-	if (lower.includes("insufficient credits") || lower.includes("premium credits exhausted") || lower.includes("model not in plan") || /insufficient (credit|balance)/.test(lower) || /out of credit/.test(lower) || /not (available|included)[^.]{0,40}plan/.test(lower) || lower.includes("upgrade your plan")) return { reason: "unavailable" };
+	if (isInsufficientCreditsText(errText) || isModelOutsidePlanText(errText)) return { reason: "unavailable" };
 }
 /**
 * The usage-window evidence carried by a rejected request, mirroring the CLI's
@@ -6535,7 +6645,7 @@ var GatewayFacts = class {
 * per-request options thunk and an API-key resolver from the plugin entry
 * (src/index.ts), so a settings change reaches the very next request.
 */
-const COMMAND_CODE_CLI_VERSION = "1.88.1";
+const COMMAND_CODE_CLI_VERSION = "1.88.2";
 const DEFAULT_API_BASE = "https://api.commandcode.ai";
 /**
 * The output budget this bundle asks for, and the ceiling it will never exceed.
@@ -8719,7 +8829,9 @@ var CommandCodeAdapter = class CommandCodeAdapter extends LlmAdapter {
 				}
 				return record;
 			} catch (error) {
-				failures.push(`${path}: ${readableErrorText(error)}`);
+				const reason = transportFailureReason(error);
+				const origin = httpOrigin(base);
+				failures.push(`${path}: ${readableErrorText(error)}${reason === void 0 ? "" : `；${reason}`}${origin === void 0 ? "" : `；目标 ${origin}`}`);
 				failed.push({ kind: "transport" });
 				return;
 			}
@@ -9923,13 +10035,18 @@ function applyUsageRemote(ctx, deps) {
 		remoteCtx.effect(() => () => void unregister(), "dsh-commandcode-provider: usage remote");
 	});
 }
+/** The endpoint whose answer proves the API's auth layer is reachable. */
+const PROBE_PATH = "/alpha/whoami";
 /**
-* HTTP statuses Upptime's default monitor accepts as `up` (from
-* `uptime-monitor`'s `expectedStatusCodes` fallback). Kept in sync with the
-* page's monitor rather than invented here: a 401 or 404 means the endpoint
-* answered but is not the surface the official check calls healthy.
+* HTTP statuses that prove the probed endpoint answered.
+*
+* 2xx/3xx are the positive answers. 401/403 belong here because the probe is
+* deliberately credential-less: the expected answer to an unauthenticated GET
+* of `/alpha/whoami` IS 401, and treating it as a failure would report every
+* healthy host as down. Everything else (404 from a wrong base path, 429, 5xx)
+* means the endpoint the plugin actually uses did not answer as a live API.
 */
-const UP_STATUSES = /* @__PURE__ */ new Set([
+const REACHABLE_STATUSES = /* @__PURE__ */ new Set([
 	200,
 	201,
 	202,
@@ -9948,25 +10065,27 @@ const UP_STATUSES = /* @__PURE__ */ new Set([
 	305,
 	306,
 	307,
-	308
+	308,
+	401,
+	403
 ]);
 /** Turn the response status into the outcome the row renders. */
 function probeOutcomeForStatus(status) {
-	return UP_STATUSES.has(status) ? "ok" : "http-error";
+	return REACHABLE_STATUSES.has(status) ? "ok" : "http-error";
 }
 /**
 * Probe the configured API base once. Never rejects: an invalid base, a
 * timeout and a transport failure all resolve to an outcome.
 *
-* `redirect: 'manual'` keeps the classification on the FIRST response, exactly
-* like the monitor's HTTP client; following a redirect would report the final
-* destination's status as if it were the API's own.
+* `redirect: 'manual'` keeps the classification on the FIRST response; the
+* probe reports what this Host reached, not where a redirect chain ends up.
 */
 async function probeCommandCode(options) {
 	const now = options.now ?? Date.now;
 	let url;
 	try {
-		url = new URL("/", options.apiBase).toString();
+		if (!URL.canParse(options.apiBase)) throw new TypeError("invalid api base");
+		url = `${options.apiBase}${PROBE_PATH}`;
 	} catch {
 		return {
 			outcome: "unreachable",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mars-sea/dsh-commandcode-provider",
-  "version": "0.16.2",
+  "version": "0.16.4",
   "description": "Unofficial DeepSeek Harness LLM provider plugin for Command Code, ported from pi-commandcode-provider (MIT). Registers the 'commandcode' provider route with a Models-page card and live model catalog.",
   "icon": "./icon.svg",
   "type": "module",