dsh-connect-workbuddy 3.0.0 → 3.0.2

package/CHANGELOG.md

@@ -1,5 +1,66 @@
 # Changelog
 
+## 3.0.2 (2026-09-30)
+
+### Bug Fixes
+
+- **修复 issue #24：宿主 pi-ai 0.87 下所有 WorkBuddy 模型立即失败**（`Cannot read properties of undefined (reading 'length')`）。
+  - **根因**：pi-ai 0.87 的 `normalizeContext` 把 system 提示词搬进 `messages`，成了 `{ role: 'system', content: '<字符串>' }`。而 pi-ai 0.85 的 `Message` 联合类型里**根本没有 `system` 变体**（只有 `UserMessage | AssistantMessage | ToolResultMessage`），于是它的 `estimateMessageTokens` 没有对应分支：`for (const block of message.content)` 会把**字符串逐字符**迭代，`block` 变成单个字符，`block.name.length` 抛错。崩溃发生在 `buildBaseOptions → clampMaxTokensToContext → estimateContextTokens`，即**构造请求体之前**——所以每个模型都在十几毫秒内失败，且**没有任何上游请求发出**。
+  - **修法**：新增 `adaptLegacyPiAiContext()`，在把 context 交给本插件的 pi-ai 之前，把 `messages` 里的 system 消息折回 `Context.systemPrompt`（0.85 期望的位置），并用 `withLegacyContext()` 包住 `openAICompletionsApi()` 的 `stream`/`streamSimple` 两个入口。
+  - **按形状适配，不按版本判断**：判断依据是「context 里有没有 system 消息」，不是「加载的是哪个 pi-ai 版本」。版本判断会在任一侧升级时立刻失效；本插件必须在**两代宿主**下都能活。对原生 0.85 形状（已有 `systemPrompt`）返回**同一个对象**，原路径零改动。
+  - **实测**（用插件真正加载的那份 pi-ai 0.85.1）：修复前 0.87 形状 → `Cannot read properties of undefined (reading 'length')`；修复后 → `Connection error.`（走到传输层，即请求已构造完成）。旧形状恒等，`system` 用 block 内容时也能正确合并。
+  - **守卫**：`tests/adapter.spec.ts` 新增 8 例，其中**接线用例**（`vi.mock` 掉 stream API，断言 provider 收到的 context 已被折平）是必要的——只测纯函数时，把修复从 provider 上摘掉的变异**不会被打红**（已实测），等于留了一个「修复被摘掉而测试全绿」的缺口。变异验证：去掉 `withLegacyContext` → 该用例变红。
+  - **顺带核实**：宿主 asar 声明 `@earendil-works/pi-ai ^0.87.1` 且含 `normalizeContext`；磁盘上四份 pi-ai（profiles / 插件嵌套 / 仓库 / 旧安装树）**全是 0.85.1**。issue 建议的「插件包内放一份 0.87.1」绕行在本机并未生效——嵌套副本是 0.85.1，漂移依旧，只是被当前加载的旧 `dsh-llm-pi-ai`（0.1.5-rc.2，不含 `normalizeContext`）掩盖了。
+  - **peer 范围放宽**：`@earendil-works/pi-ai` 由 `>=0.85.0 <0.86.0` 改为 `>=0.85.0 <0.88.0`（peer 与 dev 两处）。原范围**不承认 0.87**，与实际已验证的兼容性不符：适配按形状生效，代码在 0.85 与 0.87 下都已验证可用。放宽后 `pnpm install` 报 "Lockfile is up to date"，未产生 lockfile 变更。
+
+## 3.0.1 (2026-09-30)
+
+### Bug Fixes
+
+#### 保存账号池会把该区域的模型目录整段删掉，于是「保存后目标模型就下架了」
+
+**用户报告**：在账号池里点保存后，目标模型立刻显示「已下架」，并且「一键测试所有账号」被停用；**macOS 上不出现**。
+
+**根因（真机取证）**：`__save` 的路由此前做的是 `{ ...current, ...incoming }` —— **区域级整体替换**。而客户端只能按自己的 settings 镜像拼出槽位，这个镜像**本仓库自己注释过对本命名空间不可靠**（经本端点写入不会刷新它，在 0.1.7 线上可长期陈旧）。于是：
+
+1. 保存账号池时客户端 POST 的是 `{ cn: { pool } }`；
+2. Host 用这一个键**替换掉整个 cn 槽** → `lastCatalog`、`enabledModelIds`、`imageModelIds`、`contextBudgets`、`enabled` **全被删除**；
+3. `displayModels` 见 `lastCatalog` 为空即回落到**静态 CN 目录**（`src/catalog.ts`），而它只有 `deepseek-v4-flash`（第 39 行），**没有 `deepseek-v4.1-flash`**（那在第 59 行，属国际版）；
+4. 于是判「已下架」，`resolveTargetModel` 返回 `stale`，测试随之停用。
+
+**证据**：同一 profile 下留有 1:58 的配置备份（20711 B，`cn.lastCatalog` 含 `auto`/`hy4-preview`/… 完整清单），而当前配置（14873 B）的 `cn` 槽**只剩 `pool`**——差出的约 6 KB 正是被删掉的目录。本项目审计 `docs/audit/D-verification.md` L-4 记录过同一机制（当时是 `toggleRegion` 的过期整槽写入），但只修了写入交错，没有修「整体替换」这一步。
+
+**修法**：`__save` 的合并**下沉一层**——字段的键做一层（保兄弟区域），被触及区域槽内的键再做一层。**只保留客户端未提及的键**；客户端显式发来的字段（含空数组）依旧覆盖，因此「主动清空」不受影响。刻意不再更深：`contextBudgets` 这类嵌套 map 由其所有者整体发送，继续递归会让单个条目无法删除。
+
+**为什么不动 macOS**：该合并只在「客户端漏发了某个键」时才有差别；macOS 上的槽位是完整的，因此对它是恒等操作。
+
+守卫：`tests/web-status.spec.ts` 新增 4 例——① 只发 `{cn:{pool}}` 时 `contextBudgets`/`enabled` 必须存活（**这条直接复现本次数据丢失**）；② 国际版同理（`{global:{pool}}` 不得删掉 global 的目录与预算）；③ 显式发空数组仍能清空；④ `accounts` 的账号 id 是字符串，不得被对象化。**变异验证**：把合并改回一层展开 → ① 与 ② 同时变红（已实测）。
+
+**即时补救**：卡片上的「切回自动」会把目标换成该区域的免费模型（`hy3`，静态目录里也有），立即可用；被删掉的模型勾选与上下文预算需在「模型管理」里重新保存一次以重建 `lastCatalog`。
+
+**国际版（global）同样已覆盖，而且它更需要这个修复**：
+
+- **为什么覆盖**：合并遍历的是**字段自己的键**，不认区域名——`{cn:{…}}` 与 `{global:{…}}` 走的是同一段代码，没有区域分支。有专门用例钉住（`preserves omitted fields in the INTERNATIONAL region too`）。
+- **为什么它更需要**：同一个删除在国际版**不表现出任何症状**。`displayModels` 在 `lastCatalog` 为空时回落到静态目录，而**国际版静态目录里有 `deepseek-v4.1-flash`（`src/catalog.ts:59`，倍率 0）**，国内版静态目录没有（只有 `deepseek-v4-flash`，L39）。所以国际版用户的目标模型照旧解析成功——目录、勾选、上下文预算**被默默删掉却毫无提示**。国内版至少还会喊「已下架」。
+- 这条区域差异已用 `tests/catalog.spec.ts` 的断言钉住（`cn.has('deepseek-v4.1-flash') === false` 且 `cn.has('deepseek-v4-flash') === true`），避免日后改动任一份目录时悄悄改变「哪些故障可见」。
+
+**另说明**：本次两处修复都不区分区域。`isFileContentionWriteError` 是**平台**维度的（只在 Windows 上会命中 `EPERM/EBUSY/EACCES`），两条区域共用同一段客户端代码，因此国际版的保存失败提示也一并改善。
+
+- **Windows 上文件被占用时，两条保存路径只甩原始 EPERM、不给做法**（模型列表与账号池；账号路径本来就有）。真机实测（3.0.0 宿主 + 锁住 profile 的 `cordis.patch.yml`）：写入返回 HTTP 500，报文是
+  `EPERM: operation not permitted, rename 'C:\…\cordis.patch.yml.79498ff8fb27.tmp' -> '…\cordis.patch.yml'`
+  ——只报了一个临时文件路径，**没有说明原因、也没告诉用户怎么办**。而 `row.accountsWriteFailed` 早就把「杀毒软件/同步盘占用 → 关闭占用者后重试」写进了文案。
+  - **修法**：新增 `isFileContentionWriteError()`（匹配 `EPERM`/`EBUSY`/`EACCES` 三个**词边界**标记——正是 `dsh-atomic-write` 在 win32 上重试的那三个），命中时在原因后追加 `row.saveContentionHint`；模型列表与账号池两条路径接上。
+  - **只对文件占用生效，不做无条件提示**：校验类失败走的是同一个 catch，对那种失败说「去关闭占用文件的程序」会把用户引向不存在的原因。有专门用例钉住「无关失败不得出现该提示」。
+  - **macOS 行为不变**：POSIX 的 `rename` 无论目标是否被打开都直接替换，所以这三个错误码在 macOS/Linux 上基本不会出现——检测器天然只在 Windows 上生效。
+- **顺带记录的核验结论：Windows 保存逻辑本身没有「静默成功」**。锁住文件时 `regions` 与 `accounts` 两条写入都返回 500 并带 `EPERM`，客户端随即抛 `WorkBuddySettingsWriteError`，卡片明确报错并**保留草稿/原选择**（三条保存路径都不会丢弃用户改动）。回退路径的回读校验读的是「被拒后重新加载的 Host 状态」，不会把乐观写入误判为已落盘。因此本次修的是**报错文案的可操作性**，不是落盘判定。
+
+### Docs
+
+- **改正 3.0.0 里一处过期的免费模型举例**（README 两份 + 上方 3.0.0 条目）：原文写「该区域倍率为 0 的免费模型（**国内版即 `deepseek-v4.1-flash`**）」。2026-09-30 在真机上核对 live catalog（16 个模型）后，`deepseek-v4.1-flash` 的倍率是 **0.11**，国内版**唯一**倍率为 0 的是 **`hy3`**（ctx 192000）。
+  - **机制本来是对的，错的只是举例**：`pickFreeModel` 严格按 `creditMultiplier === 0` 现挑、多个候选取上下文窗口最大者，实测输入真实目录时返回 `hy3`，规则成立；`docs/ACCOUNT_POOL_PLAN.md` §4.4 早就写明「不写死 `deepseek-v4.1-flash`」并解释了原因。所以这条只是文档里的名字腐坏了，不是行为缺陷。
+  - **为什么不改成另一个写死的名字**：上游会改倍率（本次观察到的就是一次改动），任何名字都会再次过期。现改为陈述「按 `creditMultiplier === 0` 现挑」并附本次实测值，同时说明为何不写死。
+  - **顺带印证的一条正确设计**：`auto` 在上游目录里**没有** `creditMultiplier` 字段，`pickFreeModel` 不把它当免费——这正是该模块注释警告过的陷阱（缺失 ≠ 0），实测确认未踩。
+
 ## 3.0.0 (2026-09-30)
 
 > **主版本号升级的理由**：账号池是第一个会改变「哪个账号被计费」的功能。在此之前插件只有一条硬规则——**绝不为了找积分而重排账号**（`src/auth.ts` 的原注释：*This is NOT credit-seeking*）。账号池把这条规则变成**用户显式授权的例外**：打开「按积分自动轮换」后，被计费的账号可能不是你手动选的那个。这是语义级的变化，因此走主版本号。
@@ -9,7 +70,7 @@
 #### 账号池：多账号批量操作与自动轮换
 
 - **一键签到所有账号**：逐个账号签到，**签到前先查当天状态**，已签到的直接标记、**不发写入请求**（幂等，不重复领取）；单个账号失败只标记该行，**不中断整批**。
-- **一键测试所有账号**：对每个成员发一次**真实体积**的探测（约 25k 输入 token，复用模型行的 `probeModel`），得出该账号此刻能否用这个模型。默认目标是**该区域倍率为 0 的免费模型**（国内版即 `deepseek-v4.1-flash`），也可指定具体模型；指定的模型若离开目录，手动动作**明确拒绝**（`target-model-stale`）而不是偷换一个。
+- **一键测试所有账号**：对每个成员发一次**真实体积**的探测（约 25k 输入 token，复用模型行的 `probeModel`），得出该账号此刻能否用这个模型。默认目标是**该区域倍率为 0 的免费模型**（**不写死名字**：实测国内版当前只有 `hy3` 一个免费模型，而 `deepseek-v4.1-flash` 已从 0 变为 0.11——上游会改倍率，所以按目录里的 `creditMultiplier === 0` 现挑），也可指定具体模型；指定的模型若离开目录，手动动作**明确拒绝**（`target-model-stale`）而不是偷换一个。
 - **按积分自动轮换**：在可用成员中按 **① 积分余额高者优先 → ② 快过期的积分优先用完 → ③ 平手时凭据更新鲜者优先** 选出计费账号。**与手动选择互斥**：开启后手动下拉立即置灰并就地给出原因与关闭按钮（否则会出现「下拉写着 A、实际计费 B」）。**轮换只在运行时决定用谁，绝不写回 `config.accounts[region]`**，关掉开关即原样恢复你上次的手动选择；每次换号都在卡片上留一条记录。
 - **成员是显式勾选、默认空**：「本机有登录」不等于「要拿它去签到和花积分」。**空池不会退化成「全部账号」**——未勾选时两个批量按钮直接拒绝。
 - **定时自动测试**：间隔 **5 ~ 1440 分钟（默认 30）**，定时器每 60 秒走一次。下限 5 分钟是刻意的：更密只会把整池打成限流。

package/README.en.md

@@ -71,7 +71,7 @@ A machine usually carries more than one WorkBuddy sign-in. The **account pool**
 - **Check in all accounts** — signs in account by account. It **reads today's status first**, so an account that is already checked in is marked as such and receives **no write request at all** (idempotent — no duplicate claims); a single account failing marks only that row and **never aborts the batch**.
 - **Test all accounts** — sends one **real-volume** probe per member (about 25k input tokens; the same `probeModel` the model rows use, so it inherits both measured facts: throttling is triggered by request **size**, and the reset time is written into the response **body** rather than a header), reporting whether that account can serve that model right now.
 
-**Which model is tested:** by default the **zero-multiplier free model in that region's catalog** (`deepseek-v4.1-flash` on the domestic side), or a specific model you pick in the dropdown. If the model you named has left the catalog, the **manual action refuses outright** (`target-model-stale`) instead of quietly substituting another one — you asked to test *that* model, and swapping it answers a question you did not ask.
+**Which model is tested:** by default the **zero-multiplier free model in that region's catalog** (measured: the domestic catalog currently has exactly one, `hy3`; the name is deliberately **not** hard-coded, because the upstream changes multipliers — `deepseek-v4.1-flash` has already moved from 0 to 0.11), or a specific model you pick in the dropdown. If the model you named has left the catalog, the **manual action refuses outright** (`target-model-stale`) instead of quietly substituting another one — you asked to test *that* model, and swapping it answers a question you did not ask.
 
 **Cost and boundaries:** testing is a real request. Pointed at the free model (multiplier 0) it costs virtually nothing, but it is **not a free no-op**; check-in claims that day's real reward. Both actions run **only when you press the button** — the plugin **does not offer automatic check-in**, and the card does not show a switch that can never be flipped.
 

package/README.md

@@ -78,7 +78,7 @@ DSH PiAiAdapter（每个 provider 一套）
 - **一键签到所有账号** —— 逐个账号签到。**签到前先查当天状态**，已签到的直接标记为「已签到」而**不发写入请求**（幂等，不重复领取）；任何单个账号失败只标记该行，**不中断整批**。
 - **一键测试所有账号** —— 对每个成员发一次**真实体积**的探测请求（约 25k 输入 token；与模型行的「测试」是同一套 `probeModel`，因而同样继承两个实测结论：限流**按请求体积**触发、恢复时间写在**响应体**里而不是响应头），得出该账号此刻能否用这个模型。
 
-**测试目标是哪个模型**：默认是**该区域目录里倍率为 0 的免费模型**（国内版即 `deepseek-v4.1-flash`），也可以在下拉里指定一个具体模型。指定的模型若已离开目录，**手动动作会明确拒绝**（`target-model-stale`），而不是偷偷换一个——你点的是「测这个模型」，换掉就是答非所问。
+**测试目标是哪个模型**：默认是**该区域目录里倍率为 0 的免费模型**（实测国内版当前只有 `hy3` 一个；**不写死名字**，因为上游会改倍率——`deepseek-v4.1-flash` 就已从 0 变成 0.11），也可以在下拉里指定一个具体模型。指定的模型若已离开目录，**手动动作会明确拒绝**（`target-model-stale`），而不是偷偷换一个——你点的是「测这个模型」，换掉就是答非所问。
 
 **成本与边界**：测试是真实请求，指向免费模型时（倍率 0）几乎不花钱，但它**不是零成本的空转**；签到则会真实领取当日奖励。两个动作都**只在你点按钮时执行**——插件**不提供自动签到**，卡片上也不放一个永远点不动的开关。
 

package/lib/bin.js

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { O as workbuddyOwnAuthPath, T as legacyWorkbuddyOwnAuthPath, W as findWorkbuddyAppExecutable, _ as WorkBuddyCredentialStore, a as readHostHeartbeat, b as defaultDesktopAuthCandidates, c as WORKBUDDY_CONNECT_VERSION, g as WORKBUDDY_AUTH_FILE_ENV, j as WorkBuddyUpstreamClient, l as FALLBACK_WORKBUDDY_MODELS, o as workbuddyHostHeartbeatPath, r as isHeartbeatProcessAlive, u as FALLBACK_WORKBUDDY_MODELS_GLOBAL, x as defaultDesktopAuthDirs, z as WORKBUDDY_APP_EXECUTABLE_ENV } from "./host-heartbeat-JODdwFRx.js";
+import { O as workbuddyOwnAuthPath, T as legacyWorkbuddyOwnAuthPath, W as findWorkbuddyAppExecutable, _ as WorkBuddyCredentialStore, a as readHostHeartbeat, b as defaultDesktopAuthCandidates, c as WORKBUDDY_CONNECT_VERSION, g as WORKBUDDY_AUTH_FILE_ENV, j as WorkBuddyUpstreamClient, l as FALLBACK_WORKBUDDY_MODELS, o as workbuddyHostHeartbeatPath, r as isHeartbeatProcessAlive, u as FALLBACK_WORKBUDDY_MODELS_GLOBAL, x as defaultDesktopAuthDirs, z as WORKBUDDY_APP_EXECUTABLE_ENV } from "./host-heartbeat-Dwmfxuyw.js";
 import { realpathSync } from "node:fs";
 import { fileURLToPath } from "node:url";
 //#region src/bin.ts

package/lib/client.js

@@ -264,6 +264,31 @@ window.__ModuleLoader__.load({
 			}
 		};
 		/**
+		* Whether a save failure is a FILE-CONTENTION refusal.
+		*
+		* On Windows, replacing a file that another process holds open fails with
+		* `EPERM` / `EBUSY` / `EACCES` — an antivirus scanner, a sync client (OneDrive),
+		* or an editor with the profile's patch file open. POSIX `rename` replaces
+		* outright regardless of open handles, so this branch effectively never fires
+		* on macOS/Linux: the diagnosis is Windows-shaped because the failure is.
+		*
+		* It exists so a save failure can say WHAT TO DO. Without it the card pasted the
+		* raw error at the user — `EPERM: operation not permitted, rename
+		* 'C:\…\cordis.patch.yml.79498ff8fb27.tmp' -> '…\cordis.patch.yml'` — which
+		* names a temp path and no remedy. Measured live against the 3.0.0 host with the
+		* file held open: both the `regions` and `accounts` writes came back HTTP 500
+		* with exactly that text.
+		*
+		* Deliberately matched on the message rather than on `WorkBuddySettingsWriteError`
+		* alone: a validation refusal is also that class but has nothing to do with file
+		* contention, and telling such a user to close their editor would send them
+		* chasing a cause that is not there.
+		*/
+		function isFileContentionWriteError(error) {
+			const message = error instanceof Error ? error.message : String(error);
+			return /\b(?:EPERM|EBUSY|EACCES)\b/u.test(message);
+		}
+		/**
 		* Read one settings field's current object value from the scope snapshot.
 		*
 		* The snapshot holds the RESOLVED section, where a volatile field
@@ -1113,7 +1138,10 @@ window.__ModuleLoader__.load({
 					}
 					committed = true;
 				} catch (error) {
-					if (mounted.current) setSaveError(error instanceof Error ? error.message : t("row.requestFailed"));
+					if (mounted.current) {
+						const reason = error instanceof Error ? error.message : t("row.requestFailed");
+						setSaveError(isFileContentionWriteError(error) ? `${reason}${t("row.saveContentionHint")}` : reason);
+					}
 				} finally {
 					if (committed) discard();
 					if (mounted.current) setSaving(false);
@@ -2542,7 +2570,10 @@ window.__ModuleLoader__.load({
 					discardModels();
 					await refreshUsage(activeRegion);
 				} catch (error) {
-					if (mounted.current) setSaveError(error instanceof Error ? error.message : t("row.requestFailed"));
+					if (mounted.current) {
+						const reason = error instanceof Error ? error.message : t("row.requestFailed");
+						setSaveError(isFileContentionWriteError(error) ? `${reason}${t("row.saveContentionHint")}` : reason);
+					}
 				} finally {
 					if (mounted.current) setSaving(false);
 				}
@@ -3132,6 +3163,7 @@ window.__ModuleLoader__.load({
 			"row.save": "Save",
 			"row.saving": "Saving…",
 			"row.saveError": "Save failed: {message}",
+			"row.saveContentionHint": " (The file was locked, so the write was refused — on Windows this is usually an antivirus scanner, a sync client such as OneDrive, or an editor with that file open. Close whatever holds it and save again; your edits are still here.)",
 			"row.modelContext": "Maximum context {context}",
 			"row.contextBudget": "DSH context budget",
 			"row.modelOutput": "Output {output}",
@@ -3324,6 +3356,7 @@ window.__ModuleLoader__.load({
 			"row.save": "保存",
 			"row.saving": "保存中…",
 			"row.saveError": "保存失败：{message}",
+			"row.saveContentionHint": "（该文件被占用，写入被拒——Windows 上通常是杀毒软件、OneDrive 这类同步盘，或正打开该文件的编辑器。请先关闭占用者再保存一次；你的改动仍然还在。）",
 			"row.modelContext": "最大上下文 {context}",
 			"row.contextBudget": "DSH 上下文预算",
 			"row.modelOutput": "最大输出 {output}",

package/lib/{host-heartbeat-JODdwFRx.js → host-heartbeat-Dwmfxuyw.js}

@@ -2286,7 +2286,7 @@ var WorkBuddyCatalog = class {
 //#endregion
 //#region src/version.ts
 /** The npm package version this build was produced from. */
-const WORKBUDDY_CONNECT_VERSION = "3.0.0";
+const WORKBUDDY_CONNECT_VERSION = "3.0.2";
 //#endregion
 //#region src/host-heartbeat.ts
 /**

package/lib/index.d.ts

@@ -1,4 +1,5 @@
 import z from "@deepseek-ai/schemastery";
+import "@earendil-works/pi-ai";
 import { PiAiAdapter } from "@deepseek-ai/dsh-llm-pi-ai";
 import { Context } from "@deepseek-ai/cordis";
 import { SettingsNamespace } from "@deepseek-ai/dsh-settings";

package/lib/index.js

@@ -1,4 +1,4 @@
-import { A as WorkBuddyCredentialRejectedError, B as clearAtRestKeyCache, C as hasEncryptedCredentialFields, D as workbuddyAccountId, E as parseWorkBuddyAuth, F as parseReasoning, G as isEncryptedFieldWrapper, H as deriveAtRestKeyId, I as parseUpstreamModel, J as macosNestedAppBundles, K as isWorkbuddyBundle, L as prepareChatBody, M as classifyUpstreamError, N as isCredentialRejectedError, O as workbuddyOwnAuthPath, P as parseCreditMultiplier, R as regionOf, S as defaultDesktopAuthPath, T as legacyWorkbuddyOwnAuthPath, U as fetchAtRestKeyPayload, V as deriveAtRestKey, W as findWorkbuddyAppExecutable, X as readAtRestKey, Y as openEncryptedField, Z as workbuddyAppExecutableCandidates, _ as WorkBuddyCredentialStore, a as readHostHeartbeat, b as defaultDesktopAuthCandidates, c as WORKBUDDY_CONNECT_VERSION, d as WorkBuddyCatalog, f as deriveCatalog, g as WORKBUDDY_AUTH_FILE_ENV, h as WORKBUDDY_AUTH_FILENAME, i as processStartTimeMs, j as WorkBuddyUpstreamClient, k as CREDENTIAL_REJECTED_CODE, l as FALLBACK_WORKBUDDY_MODELS, m as ENCRYPTED_CREDENTIAL_CODE, n as clearHostHeartbeat, o as workbuddyHostHeartbeatPath, p as fallbackModelsFor, q as macosBundleExecutable, r as isHeartbeatProcessAlive, s as writeHostHeartbeat, t as WORKBUDDY_HOST_HEARTBEAT_FILENAME, u as FALLBACK_WORKBUDDY_MODELS_GLOBAL, v as WorkBuddyEncryptedCredentialError, w as isEncryptedCredentialError, x as defaultDesktopAuthDirs, y as authFileName, z as WORKBUDDY_APP_EXECUTABLE_ENV } from "./host-heartbeat-JODdwFRx.js";
+import { A as WorkBuddyCredentialRejectedError, B as clearAtRestKeyCache, C as hasEncryptedCredentialFields, D as workbuddyAccountId, E as parseWorkBuddyAuth, F as parseReasoning, G as isEncryptedFieldWrapper, H as deriveAtRestKeyId, I as parseUpstreamModel, J as macosNestedAppBundles, K as isWorkbuddyBundle, L as prepareChatBody, M as classifyUpstreamError, N as isCredentialRejectedError, O as workbuddyOwnAuthPath, P as parseCreditMultiplier, R as regionOf, S as defaultDesktopAuthPath, T as legacyWorkbuddyOwnAuthPath, U as fetchAtRestKeyPayload, V as deriveAtRestKey, W as findWorkbuddyAppExecutable, X as readAtRestKey, Y as openEncryptedField, Z as workbuddyAppExecutableCandidates, _ as WorkBuddyCredentialStore, a as readHostHeartbeat, b as defaultDesktopAuthCandidates, c as WORKBUDDY_CONNECT_VERSION, d as WorkBuddyCatalog, f as deriveCatalog, g as WORKBUDDY_AUTH_FILE_ENV, h as WORKBUDDY_AUTH_FILENAME, i as processStartTimeMs, j as WorkBuddyUpstreamClient, k as CREDENTIAL_REJECTED_CODE, l as FALLBACK_WORKBUDDY_MODELS, m as ENCRYPTED_CREDENTIAL_CODE, n as clearHostHeartbeat, o as workbuddyHostHeartbeatPath, p as fallbackModelsFor, q as macosBundleExecutable, r as isHeartbeatProcessAlive, s as writeHostHeartbeat, t as WORKBUDDY_HOST_HEARTBEAT_FILENAME, u as FALLBACK_WORKBUDDY_MODELS_GLOBAL, v as WorkBuddyEncryptedCredentialError, w as isEncryptedCredentialError, x as defaultDesktopAuthDirs, y as authFileName, z as WORKBUDDY_APP_EXECUTABLE_ENV } from "./host-heartbeat-Dwmfxuyw.js";
 import z from "@deepseek-ai/schemastery";
 import { readFile } from "node:fs/promises";
 import { randomBytes, timingSafeEqual } from "node:crypto";
@@ -213,6 +213,86 @@ function toPiModel(info, baseUrl, providerId) {
 	};
 }
 /**
+* Rewrite a pi-ai 0.87-shaped context into the 0.85 shape this plugin's pi-ai
+* understands.
+*
+* WHY THIS EXISTS (issue #24). pi-ai 0.87's `normalizeContext` moves the system
+* prompt INTO `messages` as `{ role: 'system', content: '<string>' }`. pi-ai
+* 0.85's own `Message` union has no `system` variant
+* (`UserMessage | AssistantMessage | ToolResultMessage`), so its
+* `estimateMessageTokens` has no branch for one: `for (const block of
+* message.content)` iterates the CONTENT STRING character by character, `block`
+* is a single character, and `block.name.length` throws
+* `Cannot read properties of undefined (reading 'length')`.
+*
+* That crash happens inside the library, in `buildBaseOptions ->
+* clampMaxTokensToContext -> estimateContextTokens`, i.e. BEFORE any request is
+* built — so every model fails instantly and no upstream traffic is sent. A
+* host running 0.87 hands us the normalized transcript while our own provider
+* is 0.85, and the plugin cannot patch the library. What it CAN do is not hand
+* a 0.87 transcript to a 0.85 API object: fold the system text back into
+* `Context.systemPrompt`, which is where 0.85 expects it.
+*
+* Deliberately SHAPE-based, not version-based: it asks "does this context carry
+* a system message inside `messages`?" rather than "which pi-ai version is
+* loaded?". A version check would be wrong the moment either side moves, and
+* this plugin has to survive both host generations.
+*
+* Returns the SAME object when there is nothing to adapt — in particular when
+* `systemPrompt` is already set (the native 0.85 shape), so the ordinary path
+* is byte-for-byte untouched.
+*/
+function adaptLegacyPiAiContext(context) {
+	if (context === null || typeof context !== "object") return context;
+	if (context.systemPrompt !== void 0) return context;
+	const messages = Array.isArray(context.messages) ? context.messages : void 0;
+	if (messages === void 0) return context;
+	const systemTexts = [];
+	const rest = [];
+	let sawSystemMessage = false;
+	for (const message of messages) {
+		if (message?.role !== "system") {
+			rest.push(message);
+			continue;
+		}
+		sawSystemMessage = true;
+		const text = systemTextOf(message);
+		if (text !== "") systemTexts.push(text);
+	}
+	if (!sawSystemMessage) return context;
+	return {
+		...context,
+		...systemTexts.length === 0 ? {} : { systemPrompt: systemTexts.join("\n\n") },
+		messages: rest
+	};
+}
+/** The text of one system message, whether it is a string or text blocks. */
+function systemTextOf(message) {
+	const content = message.content;
+	if (typeof content === "string") return content;
+	if (!Array.isArray(content)) return "";
+	const texts = [];
+	for (const block of content) {
+		if (typeof block !== "object" || block === null) continue;
+		const candidate = block;
+		if (candidate.type === "text" && typeof candidate.text === "string") texts.push(candidate.text);
+	}
+	return texts.join("\n\n");
+}
+/**
+* Wrap one pi-ai API module so both stream entry points receive a context this
+* build can actually consume (see {@link adaptLegacyPiAiContext}).
+*
+* The deferred-fetch entry points take no context and are passed through as-is.
+*/
+function withLegacyContext(api) {
+	return {
+		...api,
+		stream: (model, context, options) => api.stream(model, adaptLegacyPiAiContext(context), options),
+		streamSimple: (model, context, options) => api.streamSimple(model, adaptLegacyPiAiContext(context), options)
+	};
+}
+/**
 * Assemble the adapter. The provider's `getModels` reads the live catalog,
 * and every model's `baseUrl` is re-resolved per read so the shim's
 * ephemeral port applies from the first snapshot after startup.
@@ -240,7 +320,7 @@ function createWorkBuddyAdapter(options) {
 				}
 			} },
 			models: buildModels(),
-			api: openAICompletionsApi()
+			api: withLegacyContext(openAICompletionsApi())
 		}),
 		getModels: () => buildModels()
 	};
@@ -1540,6 +1620,10 @@ const WORKBUDDY_SETTINGS_NS = "workbuddy";
 function safeMessage(error) {
 	return (error instanceof Error ? error.message : String(error)).replace(/\beyJ[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\.[A-Za-z0-9_-]+\b/gu, "[redacted token]").replace(/(\b(?:code|token|refresh_token|access_token)=)[^&\s]+/giu, "$1[redacted]").slice(0, 500);
 }
+/** A non-null, non-array object: the only shape the slot merge applies to. */
+function isPlainRecord(value) {
+	return typeof value === "object" && value !== null && !Array.isArray(value);
+}
 function json(res, status, body) {
 	const payload = JSON.stringify(body);
 	res.writeHead(status, {
@@ -2044,10 +2128,14 @@ function registerWorkBuddyStatusRoute(ctx, deps) {
 					if (row === void 0) return json(res, 503, { error: "workbuddy namespace missing from describe()" });
 					const current = unwrapVolatileDeep(row.value?.[field] ?? {});
 					const incoming = body.value ?? {};
-					const merged = {
-						...current,
-						...incoming
-					};
+					const merged = { ...current };
+					for (const [key, slot] of Object.entries(incoming)) {
+						const prior = current[key];
+						merged[key] = isPlainRecord(prior) && isPlainRecord(slot) ? {
+							...prior,
+							...slot
+						} : slot;
+					}
 					await settings.mutate(row.ns, [{
 						op: "set",
 						path: [field],

package/package.json

@@ -2,7 +2,7 @@
   "name": "dsh-connect-workbuddy",
   "displayName": "DSH Connect WorkBuddy",
   "description": "把本机登录的 WorkBuddy 模型接入 DeepSeek Harness，并提供只读的积分概览与模型管理。",
-  "version": "3.0.0",
+  "version": "3.0.2",
   "author": "dingminhua",
   "copyright": "Copyright (c) 2026 LaoDing",
   "license": "MIT",
@@ -91,7 +91,7 @@
     "@deepseek-ai/dsh-llm-pi-ai": ">=0.1.7-rc.1 <0.3.0-0",
     "@deepseek-ai/dsh-settings": ">=0.1.7-rc.1 <0.3.0-0",
     "@deepseek-ai/schemastery": ">=3.18.1-rc.1 <4.0.0 || >=3.18.2 <4.0.0",
-    "@earendil-works/pi-ai": ">=0.85.0 <0.86.0",
+    "@earendil-works/pi-ai": ">=0.85.0 <0.88.0",
     "react": "^18.2.0"
   },
   "devDependencies": {
@@ -110,7 +110,7 @@
     "@deepseek-ai/dsh-llm-pi-ai": ">=0.1.7-rc.1 <0.3.0-0",
     "@deepseek-ai/dsh-settings": ">=0.1.7-rc.1 <0.3.0-0",
     "@deepseek-ai/schemastery": ">=3.18.1 <4.0.0 || >=3.18.2 <4.0.0",
-    "@earendil-works/pi-ai": ">=0.85.0 <0.86.0",
+    "@earendil-works/pi-ai": ">=0.85.0 <0.88.0",
     "@types/node": "^22.20.0",
     "@types/react": "~18.3.31",
     "@types/react-dom": "18.3.7",