dsh-connect-trae 2.3.1 → 2.5.0

package/CHANGELOG.md

@@ -1,5 +1,103 @@
 # Changelog
 
+## 2.5.0 (2026-10-01)
+
+> **国际版（AI 区域）三条用户可见故障的修复**，全部来自 [issue #19](https://github.com/dingminhua/dsh-connect-trae/issues/19) 的报告与
+> 在真实 SG 凭据上的逐条实测（取证过程见 `docs/ISSUE19_DIAGNOSIS.md`，档位接口探索见 `docs/INTL_PAID_TIER_PROBE.md`）。
+
+### Features
+
+- **国际版模型目录改为读取全部组并集——此前只读 `solo_agent_remote` 一个组，另外 9 个模型从未出现**（issue #19 症状一）：
+  - Trae 的 remote `/models` 答复**按 function 分组**，且组间不是互替关系：实测 `solo_agent` 组 19 个模型、`solo_agent_remote` 组 10 个。旧代码请求 `solo_agent_remote,solo_work_remote` 后**只遍历 preferred 一个组**，报告人 Trae IDE 里能看到、插件里却没有的 `GPT-6-Astra`、`GPT-5.6-Sol/Terra/Luna`、`GPT-5.5`、`GLM-5.2` 等**恰好全部在没读的 `solo_agent` 组里**。
+  - 现在 `TRAE_REMOTE_DIRECTORY_FUNCTIONS` 按区域声明要请求的组（AI 区域三个全取），**每个返回的组都并入目录**，同一模型先出现者胜、只发一次。请求一个网关不认识的组是无害的（未知组名被忽略而非拒绝）。
+  - **发现面变宽，调用面反而收窄**：目录 union 让模型「可见」，但模型**只能通过列出它的那个 function 调用**。实测 `get_detail_param` 对 `solo_agent` 有求必应地返回 43 条，而 `llm_utils_chat` 对**所有**以 `function: "solo_agent"` 发出的模型一律回 `code 4011`——所以 wire 函数表 `TRAE_DIRECTORY_FUNCTIONS.ai` 从 `['solo_agent','solo_work_remote','solo_work_lite']` 改为 `['solo_work_remote','solo_work_lite','solo_agent_remote']`：让目录可见的新模型仍通过真正可调用的 function 去调，`solo_agent_remote` 放最后，使两个 function 都列出的模型保持无 Remote 门禁的 IDE function。CN 区域未重新测量，维持原列表不变（union 会保留网关返回的任何组）。
+
+### Bug Fixes
+
+- **国际版订阅状态显示错误——Pro 会员被显示为「暂无有效套餐」**（issue #19 症状二）：
+  - 卡片的判据只有 `has_package`，但那个字段回答的是「是否存在套餐权益对象」，不是「账号处于什么档位」——付费会员可以持有档位而没有套餐行，报告人的账号正是如此。
+  - Trae 同时返回的**权威档位字段** `user_pay_identity` / `user_pay_identity_str`（实测免费账号为 `0` / `"Free"`）此前在插件源码里**零引用**，被直接丢弃。
+  - 现在 `payStatus()` 完整保留两个档位字段（缺失时省略，不伪造），卡片判定改为 `has_package ∪ 档位非 Free`，并把档位名渲染进套餐行：「已有有效套餐 · Pro」/「暂无有效套餐 · Free」——免费用户看到**为什么**没有套餐，而不是只看到一个布尔。
+  - **只有精确的 `Free`（大小写不敏感）算免费**：未知或未来的档位名按付费处理。把付费会员错标成免费，比把免费账号显示成未知档位，是伤害更大的那个错误。
+- **订阅门禁错误被吞成「Stream ended without finish_reason」**（issue #19 症状三）：
+  - **现象**：免费/低档位账号选中需订阅的模型（如 `GPT-6-Sol`），得到的是一句像网络故障的报错，且会被重试多次。Trae 实际返回的是 **HTTP 200** + `event:error {"code":1005,"extra":"{\"plan\":N}"}`（`extra` 是 wire 上的 JSON **字符串**，订阅门禁只体现在 `extra.plan`）。
+  - **根因**：旧实现把上游答复原样以 200 转发，中途 `controller.error()` 截断流——HTTP 头已按 200 发出，`code 1005` 与 `plan` 在 HTTP 边界处丢失，pi-ai 只看到「流没给 finish_reason 就断了」，按传输故障处理。
+  - **修法**：`bridgeTraeSoloStreamOrFail` 在转发前**窥探流头**——任何输出开始之前的拒绝被翻译成真实的错误结果：`1005` → `402 hard_credit`（携带 plan 与「需要付费档」的文案，`hard_credit` 是让客户端停止而非重试的语义，402 是如实的状态码），`4011` / `4001` / 未知码 → `400 client`。code、plan、以及 Trae 那句与 code 相矛盾的原文（如 4011 自称"rate limit"）**全部保留在 message 里**。
+  - 输出已开始后到来的拒绝（拒绝打断正在进行的回答）：流以翻译后的可读错误终止，而不是裸截断。
+  - **窥探不伤健康流**：被窥探消费的头部字节逐字节重放；队列/进度/输出等任何 Trae 自有事件即触发直通，缓冲不超过首个事件。排队事件、分片工具调用等既有行为不变。
+  - **顺带登记两个此前不认识的错误码**（`grep` 证实源码与测试均零命中）：`1005` = 订阅门禁（档位不够，`extra.plan` 指明所需档位）；`4011` = 该 function 不提供此模型（其自报文案声称 rate limit，具有误导性）。已知的 `4001` = `config_name` 不被该 function 接受，语义不变。
+
+### Tests
+
+- **全仓 373 → 392（+19），含 5 次变异验证全部被抓**：
+  - `tests/solo-bridge.spec.ts` **+11**：`describeTraeFailure` 文案 5 例（1005 带 plan、4011 用 hint 而把误导原文降级为 detail、未知码回退、无 code 事件不虚构后缀、已解包的 extra 对象也能读 plan）；`bridgeTraeSoloStreamOrFail` 策略 5 例（1005→402 `hard_credit`、4011→400 `client`、无名事件经错误码带识别、健康流含排队事件直通且头部重放逐字节不丢、输出先行后拒绝的流中场景以可读 message 终止）；`TraeSoloBridge` 端到端 1 例（完整链路上订阅门禁成为错误结果而非被截断的 200 流）。其中「流中拒绝」用分块流构造：单块 body 会被窥探一起解码，无法测到该路径。
+  - `tests/card-subscription.spec.tsx` **新文件 5 例**：渲染真实卡片断言档位行——Pro 从档位单独判定、无档位时套餐行仍工作、Free 命名而非只报布尔、试用窗口优先于档位文案，及 `traePlanIsPaid` 判据表（只有精确 Free 算免费，未知档位按付费）。一例说明：无档位字符串时 badge 与 value 渲染同一裸标签，`getByText` 会命中多个元素，须用 `getAllByText`——这例测试在上一会话中断前正以多元素匹配失败，本次接续修复。
+  - `tests/usage.spec.ts` **+2**：档位字段与 `has_package` 分开保留；上游不带档位时省略而非伪造。
+  - `tests/solo-remote.spec.ts` **+1**、`tests/solo.spec.ts` 改写既有断言：目录 union 行为、AI 区域 function 顺序（`solo_agent` 必须绝迹）。
+  - **变异验证**（确认测试有约束力，5 次全部被抓）：① 分类忽略 1005 → 2 例失败；② plan 提取禁用 → 4 例失败；③ 窥探失效（打回旧行为，即本次要修的 bug）→ 4 例失败；④ 档位恒判免费 → 2 例失败；⑤ union 退回单组 → 3 例失败。恢复后 45 文件 392 例全绿。
+
+### Docs
+
+- `docs/ISSUE19_DIAGNOSIS.md`（随 2.4.0 后的 `009d9a7` 提交）：三个症状的完整取证——remote 目录两组的差集与报告人 IDE 的模型逐一吻合、`user_pay_identity_str` 零引用的 grep 证据、以真实上游字节喂给 bridge 复现「截断的 200」。
+- `docs/INTL_PAID_TIER_PROBE.md`（`d76800d`）：付费账号是否有专属接口的探索结论——**没有**，免费/付费同端点同目录，差别在调用时授权；含 `1005` / `4011` 的对照实验记录。
+
+## 2.4.0 (2026-09-29)
+
+> **包含仍未发布的 2.3.2。** 2.3.2 的 peer 范围修复（`>=0.1.7-rc.1 <0.2.0-0` → `<0.3.0-0`，见下一节）
+> 只提交到 git、未推送也未发 npm，因此**没有用户拿到过 2.3.2**；2.4.0 在同一棵树上，同时带上该修复。
+> 若你希望「先单独发一个只含关键修复的补丁版」，就先推 2.3.2 的 tag 并 `npm publish`，再把本节改号为 2.4.0。
+
+### Features
+
+- **识图能力改为跟随 Trae 自身的 `multimodal` 标注，不再要求逐一手勾**（issue #16）：
+  - **原来的行为**：插件唯一决定是否给某个模型开图片输入的入口是 `applyImageSelection()`，判据只有用户手动勾选的 `imageModelIds`；上游 `multimodal` 字段虽然在 `parseTraeRemoteModel` 里被解析出来，但**从不进入 `TraeModelInfo`**，等于被丢弃。于是 Trae 自己标为多模态的模型，在这里也必须手动勾一次才认图。
+  - **改后的判据**：`你手动勾选的 ∪ Trae 标注 multimodal=true 的`（新增 `traeModelAcceptsImage()` 作为唯一判据函数，`applyImageSelection()` 走它）。
+  - **为什么是并集而不是"只信上游"**：上游标注是**建议**而非契约——`llm_utils_chat` 是轻量 SOLO 通道，本插件历史上多次被 `4001 param is invalid` 坑过。所以上游标注**只决定图片输入，从不决定模型可用性**：模型能否出现在目录里仍只由 wire（`get_detail_param` 的 `config_name`）join 决定，标注不会把一个不可调用的模型放进来。
+  - **为什么也不"只信手动"**：Trae IDE 自己的附图按钮就是纯粹由 `capabilities.vision` 驱动的，没有任何用户开关。厂商已声明的能力还要用户再勾一次，只有本插件这一处。
+  - **缺失或为 false 一律按纯文本**：`multimodal` 缺失（旧版本保存的目录、静态兜底目录）或明确 `false` 都不自动开。宁可不发，也不对能力不确定的模型谎报"支持"——一旦声明 `image`，DSH 会把图片原样发出而不是降级成文字。
+  - **自动开启 ≠ 自动发图**：它只是让 DSH 允许给该模型附图；图片始终由你在对话里自己贴，插件不会替你发送任何东西。
+  - **卡片同步**：被自动开启的行显示为**已勾选且只读**并带说明（新增 `row.modelImageAuto`）。并集语义下取消勾选会被立刻重新加上，给它一个可点的框是在撒谎。目前**不提供**单个模型"强制关闭"的出口（那需要额外的 opt-out 状态）；如需请在该 issue 下提出。
+  - **实测结果**（本机 CN 账号，实时目录 15 个模型，用户零勾选）：8 个自动获得图片输入（`Doubao-Seed-Evolving`、`Doubao-Seed-2.1-Pro`、`Doubao-Seed-2.1-Turbo`、`deepseek-v4.1-flash`、`kimi-k3`、`minimax-m3`、`qwen3.8-max`、`qwen-3.7-plus`）；Trae 明确标 `false` 的 4 个（`glm-5.3`、`glm-5.2`、`DeepSeek-V4-Flash-Official`、`DeepSeek-V4-Pro-Official`）**正确地保持纯文本**；目录未标注的 3 个同样保持纯文本。
+  - **文档**：`docs/MODEL_MANAGEMENT_DESIGN.md` 里「仅展示，不作为图片授权依据」的描述随此改动失效（该字段现在确实参与授权，但**不**决定可用性）；README 新增「图片输入（识图）」一节说明完整语义（中英同步）。
+
+### Bug Fixes
+
+- **`multimodal` 必须随 `lastCatalog` 一起持久化，否则"自动识图"会在保存/重启后失效**：运行时目录每次保存与启动都从 `lastCatalog` 重新推导（`derive()` → `applyImageSelection()`）。若 schema 不声明该字段，它被剥掉后自动开图会立刻退回手动。已在 `modelConfig`（`src/index.ts`）显式声明 `multimodal: z.boolean()`，并在卡片保存处显式带上——卡片原本会强制把 `input` 写回 `['text']`（这是对的，`input` 由宿主推导），但 `multimodal` 是上游事实，不能跟着一起丢。
+- **README「暂不支持」表格过期**：`deepseek-v4.1-flash` 已从该表移除。`docs/DS41_CALLABILITY.md`（2026-09-15）取证时 8 个 SOLO function 全部返回 `4001`，但**上游后来把它开放到了 SOLO 通道**——当前实时目录里它带 `wireFunction: solo_work_remote`，已实测刷新确认可调用。表内其余 3 个（`glm-5.3-flash`、`kimi-k2.8-preview`、`qwen3.8-flash`）实测仍不在目录里，保留在表中。
+
+### Tests
+
+- **新增 6 例**（全仓 367 → 373），覆盖并集的四个方向：标注 true 自动开、标注 false 不开、缺失按不开、手动勾选永不撤销（含上游后来不再标注的情况），以及"别的模型的勾选不得串到本行"。
+- **新增 2 例端到端**（走真实插件 + `ctx.llm.listModels()`）：零勾选时标注为 true 的模型 `inputModalities === ['text','image']`；标注 false 的模型仍为 `['text']`。后者同时证明**手动勾选路径没被破坏**，且旧形状（不带 `multimodal` 的 `lastCatalog` 行）仍能通过 schema 校验。
+- **变异验证**（确认测试有约束力，3 次全部被抓）：① 把并集改回"仅手动勾选" → 3 例失败（含端到端那例）；② 让 schema 不声明 `multimodal` → 无法编译（弱信号，已改写为更真实的形式）；③ 在 `derive` 里于 `applyImageSelection` 之前剥掉该字段 → 端到端那例精确失败。恢复后 44 文件 373 例全绿。
+
+## 2.3.2 (2026-09-29)
+
+### Bug Fixes
+
+- **插件在 DSH 0.2.0-rc.1 上整个不加载——peer 上界 `<0.2.0-0` 把 0.2.0 的预发布线一起挡掉了**（用户升级宿主后发现，2.3.1 的实际故障）：
+  - **现象**：升级到 0.2.0-rc.1 后插件「凭空消失」——模型选择器里没有 `trae` / `trae-global`，插件设置卡片不见，用量路由 `/plugins/dsh-connect-trae/usage` 返回 **404**。不是某个功能坏掉，是 bundle 从未挂载。
+  - **根因**：DSH 自 0.1.7-rc.1 起有一道 **bundle 级 peer 门禁**（`@deepseek-ai/dsh-app-boot` 的 `evaluatePluginCompatibility`）：`loadProfileDirectory()` 逐条比对 bundle 自己声明的 DSH peer，任一条不满足就**抛错并整体跳过该 bundle**，只往 stderr 写一行 `skipping profile bundle ...`。本插件 8 条范围写的是 `>=0.1.7-rc.1 <0.2.0-0`，而 **SemVer 里 `0.2.0-rc.1 < 0.2.0-0`**——`-0` 这个上界本意是「排除 0.2.0 正式版」，实际却把**整个 0.2.0 预发布线**也排除了。宿主的判定是 `semver.satisfies(runtime, range, {includePrerelease: true})`，在这条范围下对 `0.2.0-rc.1` 返回 `false`。
+  - **不是 0.2.0 破坏了契约**：逐符号核对确认本插件消费的 API 在 0.2.0-rc.1 上**全部未变**（`resolveRetryPolicy`、`PiAiAdapter`、`ResolvedPiAiProviderProfile`、`withFileLock`、`writeFileAtomic`、`resolveDshHome`、`AttachmentStore` 的声明逐字一致；插件触及的 `dsh-atomic-write` / `dsh-home-paths` / `dsh-attachment` / `dsh-settings` / `dsh-host-webserver` / `dsh-client-ui-slots` 源码**逐字节相同**）。同级 `dsh-connect-workbuddy` 2.1.1 写的是 `<0.2.0`——**放行 0.2.0 预发布、只挡正式版**，因此照常加载；本插件用的 `-0` 是同一意图下多写了一个 `-0`，恰好把预发布线也挡在门外。差别只在这一个字符。
+  - **也不是本插件独有的现象**：本机 profile 的 19 个 bundle 里有 **5 个**同时被这道门禁拒绝（本插件、`@changfenhuang/dsh-genui`、`dsh-better-reasoning-effort`、`dsh-free-search`、`dsh-rewind-plugin`）。用宿主自带的门禁逐条判定，这 5 个全部不在挂载行里、其余全部 `active`——判定与观测逐一吻合，排除了「判定模型与宿主不一致」。
+  - **修法**：8 条 peer 上界改为 `>=0.1.7-rc.1 <0.3.0-0`——放行 0.1.7 与 0.2 两条线（含各自的 alpha/rc），仍挡住未审查的 0.3 线。用宿主自己发行的门禁代码复验：0.1.7-rc.1 / 0.1.7-rc.2 / 0.2.0-alpha.1 / 0.2.0-rc.1 / 0.2.0 / 0.2.9 全部放行，0.3.0-rc.1 / 0.3.0 / 1.0.0 仍被拒绝。
+  - **`devDependencies` 一并升到 0.2.0-rc.1**：否则本地 typecheck / 测试验的是 0.1.7 的契约，而用户跑的是 0.2.0——「开发树与用户解析不一致」正是上一轮 `pnpm-workspace.yaml` 里为 schemastery 写过的那类隐患。
+  - **为什么 CI 全绿却漏掉了**：范围写错不会让任何断言变红——`tsc` 通过（范围只是字符串）、345 条单测通过（插件自身代码没动）、`pnpm install` 通过（devDependency 钉的版本满足那段写错的范围）、CI 通过。**只有真实宿主会拒绝加载**。这个缺口已由下面的回归测试补上。
+
+### Tests
+
+- **新增 `tests/dsh-peer-range.spec.ts`（+22 例，全仓 345 → 367）**：把「peer 范围必须覆盖哪些运行时」变成可执行的断言，而不只是一句文档。
+  - **逐条范围断言四个边界**：**必须放行** 0.1.7-rc.1 / 0.1.7-rc.2 / 0.1.9 / **0.2.0-alpha.1** / 0.2.0-rc.1 / 0.2.0 / 0.2.9；**必须拒绝** 0.3.0-rc.1 / 0.3.0 / 1.0.0。只断言「0.2.0-rc.1 能过」是不够的——那样把范围放宽成完全没有上界的 `>=0.1.7-rc.1` 也会通过。
+  - **镜像宿主而不是自创判定**：宿主用 `semver.satisfies(runtime, range, {includePrerelease: true})`，本测试用同一选项，并**单独断言该选项确实是必需的**（同一条范围在默认选项下会拒绝 `0.2.0-rc.1`）——否则将来 semver 语义一变，断言会静默地不再测宿主真正做的事。
+  - **点名禁止 `<0.2.0-0` 回归**：把这次故障的写法单独钉死，因为它在「整理版本范围」时很容易被误改回去。
+  - **守护开发树一致性**：断言每个声明了 peer 范围的包都在 `devDependencies` 里有钉住的版本、且该版本落在自己声明的范围内；另断言 `dsh.client.inject` 里的每个客户端包都有 devDependency 钉版本。（客户端 roster 那 5 个包没有 peer 范围——宿主从自己的图解析它们，因此只断言「有钉版本」，不虚构范围。）
+  - **变异验证**（确认测试真的有约束力，不是空转）：把范围改回 `<0.2.0-0` → **10 例失败**；改成无上界的 `>=0.1.7-rc.1` → **8 例失败**；把 peer 下界抬到 `>=0.2.0-rc.1` 但留下 0.1.7 的 devDependency 钉 → **8 例失败**。三次变异全部被抓住，恢复后 22 例全绿。
+
+### Docs
+
+- **新增 `docs/DSH_0.2.0_RC1_IMPACT_CHECK.md`**：0.1.7-rc.2 → 0.2.0-rc.1 的完整影响面核对（261 提交 / 1109 文件 / 0 处破坏性提交；插件触及的包逐字节比对；类型级与运行时值级符号核对各 0 处移除；端到端类型验证）。含方法论备注：宿主 `app.asar` 内的 payload 基准在本机实测为 `8 + headerSize`（不是常见的 `16 + headerSize`），需用条目 `integrity.hash` 做 SHA256 校验后定位，否则会读出错位数据。
+- **README（中英）新增「peer 上界为什么不写 `<0.2.0`」**：给出三种写法的对照表与故障成因，说明该类错误为什么没有 CI 能拦、只能靠专门的测试。
+
 ## 2.3.1 (2026-09-26)
 
 ### Features

package/README.en.md

@@ -39,6 +39,7 @@ A [DeepSeek Harness (DSH)](https://github.com/deepseek-ai/deepseek-harness) bund
 - **DSH local tool loop** — gets pending structured `tool_calls` from Trae `llm_utils_chat`, lets DSH execute its own local tools, then returns tool results to the model.
 - **Dual-region detection (CN / international)** — detects local sign-ins across Trae CN, TRAE SOLO CN, Trae, and TRAE SOLO installs. The region is derived from the credential's own `userRegion` claim (host suffix and edition label as fallbacks), with no manual switch.
 - **Region-isolated directories and selections** — CN and international each keep their own model directory, enabled picks, image opt-ins, and context budgets, and each provider reads only its own slot.
+- **Vision follows Trae's own declaration** — when Trae marks a model `multimodal`, the plugin advertises it as such with no manual tick per model (issue #16). The rule is "**models you tick ∪ models Trae marks multimodal**": models Trae marks `false`, or does not mark at all, stay text-only and can still be turned on by hand. A row enabled automatically is shown ticked and read-only, with the reason. See "Image input" below.
 - **Account switching** — refreshes the token list and lets users select an account without storing tokens in DSH settings.
 - **Read-only usage and model management** — Work/general credits on CN accounts, subscription/trial status on international ones; enable Trae models freely. Read-only queries consume nothing.
 - **Daily check-in claiming** — a one-click claim on the CN card (the button shows the daily reward, and reads "Claimed today" and disables itself once that account has been paid). This is the plugin's **only** operation that changes account state, and it runs only when you click. Trae allows **one check-in per device per day**: after switching accounts, if this machine's slot is already spent (by any account), the selected account cannot claim here that day — the card explains that rather than pretending the account was paid. The international side has no check-in activity, so no button is shown there.
@@ -63,24 +64,43 @@ Usage overview hits the read-only `https://api.trae.cn/trae/api/v2/pay/*` and `/
 
 > See `docs/IMPLEMENTATION_PLAN.md`, `docs/SOLO_ROUTE_DECISION.md`, `docs/USAGE_API_RESEARCH.md`.
 
+## Image input
+
+The plugin declares whether a model accepts images, and DSH uses that to decide whether an attachment is **sent as-is** or **degraded into a text description**. That makes the declaration a hard switch, not a cosmetic label.
+
+The rule is the **union of two sources** (issue #16):
+
+| Source | Meaning |
+|---|---|
+| Trae's own `multimodal` field | Trae's own answer to "does this model read images". The Trae IDE's attach-image button is driven purely by this field, with no user switch |
+| Your manual `imageModelIds` ticks | Explicit choice. Even if Trae later stops reporting the flag for that model, your tick is never revoked |
+
+Notes:
+
+- **Only `multimodal === true` enables it automatically.** A missing field (a directory saved by an older release, a fallback row) or an explicit `false` stays text-only — better to not send than to advertise a capability we cannot confirm.
+- **Automatic does not mean automatic sending.** It only lets DSH attach an image to that model; you still choose to attach one per message. The plugin never sends an image on your behalf.
+- **An automatically enabled row is shown ticked and read-only.** Under the union, unticking it would be re-applied immediately, so offering a clickable box would be a lie. There is currently no per-model "force off" escape hatch; if you want one, say so on issue #16.
+- Manual ticks still work for models Trae marks `false` (or does not mark), which is the escape hatch for "upstream does not say so, but it actually works".
+
 ## Model coverage
 
 This plugin serves models from Trae's **SOLO channel** (`DeepSeek-V4-Flash-Official`, `DeepSeek-V4-Pro-Official`, `GLM-5.3`, `GLM-5.2`, `Kimi-K3`, `MiniMax-M3`, `Qwen3.8-Max`, `Doubao-Seed-*`, …), for both the domestic and international sides.
 
-The following four models currently come from the **TraeCode (Trae IDE) channel** and have **not** been opened up to the SOLO channel, so this plugin does **not** support them yet:
+The following three models currently come from the **TraeCode (Trae IDE) channel** and have **not** been opened up to the SOLO channel, so this plugin does **not** support them yet:
 
 | Model |
 | --- |
-| `deepseek-v4.1-flash` |
 | `glm-5.3-flash` |
 | `kimi-k2.8-preview` |
 | `qwen3.8-flash` |
 
-These four are reachable only inside the Trae IDE client; there is no callable channel for them on the plugin side. **They can only be supported once Trae officially opens them up to the SOLO channel.**
+These three are reachable only inside the Trae IDE client; there is no callable channel for them on the plugin side. **They can only be supported once Trae officially opens them up to the SOLO channel.**
+
+> `deepseek-v4.1-flash` used to be listed here — `docs/DS41_CALLABILITY.md` (evidence gathered 2026-09-15) measured `4001 param is invalid` from all eight SOLO functions. **Trae has since opened it up to the SOLO channel**: it now ships in the live directory with `wireFunction: solo_work_remote` and calls succeed (verified against a live refresh). This page had not caught up, so it is removed from the table.
 
 > Note the distinction: **`GLM-5.3` is supported** (it goes through the SOLO channel), but **`glm-5.3-flash` is not** — they are different models.
 
-To use these four models today, use the **Trae IDE itself**.
+To use these three models today, use the **Trae IDE itself**.
 
 ## Install
 
@@ -99,6 +119,23 @@ Restart the DSH process after install/update/uninstall.
 ## DSH version compatibility
 
 **Requires DSH 0.1.7-rc.1 and up** (since 2.3.0 the plugin no longer targets hosts older than 0.1.7; the peer dependency ranges are narrowed accordingly).
+**The peer upper bound is `<0.3.0-0`, so both the 0.1.7 and the 0.2 line are supported.**
+
+The 0.2.0 line breaks **no** contract this plugin consumes — a symbol-by-symbol check confirmed every API it imports is unchanged ([full 0.2.0-rc.1 impact check](https://github.com/dingminhua/dsh-connect-trae/blob/main/docs/DSH_0.2.0_RC1_IMPACT_CHECK.md)). But since 0.1.7 DSH enforces a **bundle-level peer gate**: a bundle whose peers are unsatisfied is skipped **whole**, with one line on stderr and nothing in the UI — the plugin simply disappears.
+
+### Why the upper bound is not `<0.2.0`
+
+2.3.1 wrote `<0.2.0-0`, meaning "not the 0.2.0 release yet". That spelling is **wrong**: in SemVer `0.2.0-rc.1 < 0.2.0-0`, so it excluded the entire 0.2.0 **prerelease** line, and the plugin went dead the moment a user installed the first 0.2.0 prerelease.
+
+| Spelling | 0.1.7 line | 0.2.0 alpha/rc | 0.2.0 release | 0.3.0 |
+| --- | --- | --- | --- | --- |
+| `>=0.1.7-rc.1 <0.2.0-0` ❌ | ✅ | ❌ | ❌ | ❌ |
+| `>=0.1.7-rc.1 <0.2.0` | ✅ | ✅ | ❌ | ❌ |
+| `>=0.1.7-rc.1 <0.3.0-0` ✅ | ✅ | ✅ | ✅ | ❌ |
+
+`tests/dsh-peer-range.spec.ts` pins this invariant: every peer range must admit the 0.1.7 and 0.2 lines, must reject the 0.3 line, and `<0.2.0-0` may not come back.
+
+> **Why this needs a dedicated test**: a wrong range passes `tsc`, every unit test, `pnpm install`, and CI — only a real host refuses to load it. That is exactly how 2.3.1 shipped green.
 
 0.1.7 rebuilt the settings machinery, and the plugin follows the 0.1.7 contract:
 

package/README.md

@@ -39,6 +39,7 @@
 - **DSH 本地工具循环** —— 通过 Trae `llm_utils_chat` 获取待执行的结构化 `tool_calls`，交由 DSH 自带的本地工具执行，再将工具结果回传模型。
 - **国内国际双区域自动识别** —— 自动发现 Trae CN / TRAE SOLO CN / Trae / TRAE SOLO 四个本地安装的登录账号；区域由凭证自带的 `userRegion` 声明自动判定（host 后缀与 edition 标签兜底），无需手动指定。
 - **目录与勾选按区域隔离** —— 国内版与国际版各一套模型目录、勾选、图片开关与上下文预算，两个供应商各自读各自的槽位。
+- **识图能力跟随 Trae 自身的标注** —— Trae 把某个模型标为 `multimodal` 时，插件直接按多模态对外声明，无需你在卡片上逐一手勾（issue #16）。判据是「**你手动勾选的 ∪ Trae 标注多模态的**」：Trae 明确标 `false` 或未标注的模型保持纯文本，你可以照旧手动为它们打开图片输入。卡片上被自动开启的那一行显示为已勾选且只读，并说明原因。详见下方「图片输入（识图）」。
 - **多账号切换** —— 支持重新读取 Token 列表并选择账号；Token 不写入 DSH 设置。
 - **只读用量与模型管理** —— 国内版查看 Work 积分与通用积分，国际版查看订阅/试用状态；刷新/启用 Trae 模型；只读查询不消耗额度。
 - **每日签到领取** —— 国内版卡片上一键领取当日签到积分（按钮显示每日奖励，该账号已领取则显示「今日已领取」并禁用）；这是插件里**唯一会改变账号状态**的操作，由你主动点击触发。Trae 的签到按「**每台设备每天一次**」限制：切换账号后如果本机今天的名额已经用掉（无论是哪个账号用的），当前账号在本机当天无法再签，卡片会说明原因而不是谎报「已领取」。国际版没有签到活动，因此不显示该按钮。
@@ -63,24 +64,45 @@ DSH PiAiAdapter（每个 provider 一套）
 
 > 详见 `docs/IMPLEMENTATION_PLAN.md`、`docs/SOLO_ROUTE_DECISION.md`、`docs/USAGE_API_RESEARCH.md`。
 
+## 图片输入（识图）
+
+插件对外声明「这个模型是否接受图片」，DSH 据此决定附图是**原样发出**还是**降级成文字描述**。所以这条声明是硬开关，不是一个无关紧要的标记。
+
+判据是**两个来源的并集**（issue #16）：
+
+| 来源 | 含义 |
+|---|---|
+| Trae 目录自带的 `multimodal` 字段 | Trae 自己对「这模型读不读图」的回答。Trae IDE 本体的附图按钮就是纯粹由这个字段驱动的，没有用户开关 |
+| 你在卡片上手动勾选的 `imageModelIds` | 显式选择。即便 Trae 后来不再标注该模型，你的勾选也不会被撤掉 |
+
+要点：
+
+- **只有 `multimodal === true` 才自动开启。** 字段缺失（旧版本保存的目录、兜底目录）或明确 `false`，都按纯文本处理——宁可不发，也不对一个能力不确定的模型谎报「支持图片」。
+- **自动开启并不等于自动发图。** 它只是让 DSH 允许你给该模型附图；图片始终由你在对话里自己贴，插件不会替你发送任何东西。
+- **被自动开启的行在卡片上显示为已勾选且只读。** 因为并集语义下取消勾选会被立刻重新加上，给它一个可点的框是在撒谎。目前不为单个模型提供「强制关闭」的出口；如果你需要这个开关，可以在 issue #16 下提出。
+- 手动勾选对 Trae 标注 `false`（或未标注）的模型依然有效，这是为那些「上游没标但实际可用」的情况留的口子。
+
 ## 模型覆盖范围
 
 插件提供的是 Trae **SOLO 通道**的模型（`DeepSeek-V4-Flash-Official`、`DeepSeek-V4-Pro-Official`、`GLM-5.3`、`GLM-5.2`、`Kimi-K3`、`MiniMax-M3`、`Qwen3.8-Max`、`Doubao-Seed-*` 等），国内版与国际版均走这条通道。
 
-以下 4 个模型目前来自 **TraeCode（Trae IDE）通道**，尚未开放到 SOLO 通道，因此本插件**暂不支持**：
+以下 3 个模型目前来自 **TraeCode（Trae IDE）通道**，尚未开放到 SOLO 通道，因此本插件**暂不支持**：
 
 | 模型 |
 | --- |
-| `deepseek-v4.1-flash` |
 | `glm-5.3-flash` |
 | `kimi-k2.8-preview` |
 | `qwen3.8-flash` |
 
-这 4 个模型只在 Trae IDE 客户端内可用，插件侧没有可调用的通道。**需要等官方把它们开放到 SOLO 通道后，插件才能支持。**
+这 3 个模型只在 Trae IDE 客户端内可用，插件侧没有可调用的通道。**需要等官方把它们开放到 SOLO 通道后，插件才能支持。**
+
+> `deepseek-v4.1-flash` 曾在此列——`docs/DS41_CALLABILITY.md`（2026-09-15 取证）实测 8 个 SOLO function
+> 全部返回 `4001 param is invalid`。**上游后来把它开放到了 SOLO 通道**：当前实时目录里它带
+> `wireFunction: solo_work_remote`，可正常调用（已实测刷新确认）。文档此前未同步，现已从表中移除。
 
 > 注意区分：**`GLM-5.3` 是支持的**（走 SOLO 通道），但 **`glm-5.3-flash` 不支持**——两者是不同的模型。
 
-如果你现在就要用这 4 个模型，请直接在 **Trae IDE 本体**中使用。
+如果你现在就要用这 3 个模型，请直接在 **Trae IDE 本体**中使用。
 
 ## 安装
 
@@ -101,6 +123,28 @@ npm install dsh-connect-trae
 ## DSH 版本兼容
 
 **要求 DSH 0.1.7-rc.1 及以上**（2.3.0 起不再支持 0.1.7 之前的宿主；peer 依赖范围已按此收窄）。
+**peer 上界为 `<0.3.0-0`：0.1.7 与 0.2 两条线均受支持。**
+
+0.2.0 线对本插件**没有契约破坏**——逐符号核对确认插件消费的 API 全部未变（[完整的 0.2.0-rc.1 影响面核对](https://github.com/dingminhua/dsh-connect-trae/blob/main/docs/DSH_0.2.0_RC1_IMPACT_CHECK.md)）。但 0.1.7 起 DSH 有一道**bundle 级 peer 门禁**：
+peer 范围不满足的 bundle 会被**整体跳过**，只往 stderr 写一行，界面上表现为「插件凭空消失」。
+
+### peer 上界为什么不写 `<0.2.0`
+
+2.3.1 曾写 `<0.2.0-0`，意图是「先不支持 0.2.0 正式版」。这个写法**是错的**：
+SemVer 里 `0.2.0-rc.1 < 0.2.0-0`，于是它把**整个 0.2.0 预发布线**也一并排除，
+用户装上第一个 0.2.0 预发布版后插件立即整体失效（2.3.1 的实际故障）。
+
+| 写法 | 0.1.7 线 | 0.2.0 alpha/rc | 0.2.0 正式 | 0.3.0 |
+| --- | --- | --- | --- | --- |
+| `>=0.1.7-rc.1 <0.2.0-0` ❌ | ✅ | ❌ | ❌ | ❌ |
+| `>=0.1.7-rc.1 <0.2.0` | ✅ | ✅ | ❌ | ❌ |
+| `>=0.1.7-rc.1 <0.3.0-0` ✅ | ✅ | ✅ | ✅ | ❌ |
+
+`tests/dsh-peer-range.spec.ts` 钉住这条不变式：逐条断言每个 peer 范围**必须**放行
+0.1.7 与 0.2 两条线、**必须**挡住 0.3 线，并禁止 `<0.2.0-0` 这种写法回归。
+
+> **为什么需要专门的测试**：范围写错时 `tsc` 通过、全部单测通过、`pnpm install` 通过、
+> CI 全绿——只有真实宿主会拒绝加载。这正是 2.3.1 一路绿着上线的原因。
 
 0.1.7 把设置体系整体重建，插件按 0.1.7 的契约工作：
 

package/lib/client.js

@@ -140,6 +140,21 @@ window.__ModuleLoader__.load({
 				enabled
 			});
 		}
+		/**
+		* Whether Trae's plan identity says this account is on a paying plan.
+		*
+		* `has_package` alone is NOT that question — it reports whether a package
+		* entitlement row exists, and a member can be on a plan without one (issue
+		* #19: a Pro account rendered as "No active package"). Trae spells the plan in
+		* `user_pay_identity_str`, so this reads that field, and only the exact word
+		* `Free` counts as free: an unknown or future tier name is treated as paid,
+		* because claiming "Free" for a tier we simply do not recognise is the more
+		* damaging of the two errors.
+		*/
+		function traePlanIsPaid(payIdentityStr) {
+			const tier = payIdentityStr?.trim();
+			return tier !== void 0 && tier !== "" && !/^free$/i.test(tier);
+		}
 		//#endregion
 		//#region src/client/icon.ts
 		/**
@@ -355,6 +370,24 @@ window.__ModuleLoader__.load({
 			if (value >= 1e3 && value % 1e3 === 0) return `${value / 1e3}K`;
 			return formatNumber(value);
 		}
+		/**
+		* The subscription row's verdict.
+		*
+		* Two independent things can entitle an international account: a package row
+		* (`has_package`) and a plan identity (`user_pay_identity_str`). Reading only
+		* the first is what rendered a Pro member as "No active package" in issue #19,
+		* so either one counts. The tier string is carried separately so the row can
+		* name the plan instead of just asserting a boolean.
+		*/
+		function payVerdict(payStatus) {
+			const tier = payStatus.payIdentityStr?.trim();
+			const paid = payStatus.hasPackage || traePlanIsPaid(payStatus.payIdentityStr);
+			return {
+				paid,
+				...tier === void 0 || tier === "" ? {} : { tier },
+				label: paid ? "row.subscribed" : "row.noPackage"
+			};
+		}
 		function dotStyle(status) {
 			return { background: status === "signed-in" ? "var(--dsw-alias-state-success-primary, #22a06b)" : status === "error" ? "var(--dsw-alias-state-error-primary, #d92d20)" : "var(--dsw-alias-label-dimmed, #9aa0a6)" };
 		}
@@ -663,6 +696,25 @@ window.__ModuleLoader__.load({
 			const savedEnabledIds = status.status === "signed-in" ? new Set(status.enabledModelIds) : /* @__PURE__ */ new Set();
 			const activeEnabledIds = draft?.enabledIds ?? savedEnabledIds;
 			const activeImageIds = draft?.imageIds ?? savedImageIds;
+			/**
+			* Effective image capability of one row, mirroring what the Host reports to
+			* DSH: the user's opt-in unioned with Trae's own `multimodal` flag (issue
+			* #16). Rendering the union is what keeps this checkbox truthful — showing the
+			* raw `imageModelIds` would draw an unticked box next to a model the plugin
+			* is in fact advertising as image-capable.
+			*/
+			const rowAcceptsImage = (model) => activeImageIds.has(model.id) || model.multimodal === true;
+			/**
+			* A row Trae itself declares multimodal needs no manual opt-in, so its box is
+			* shown ticked and read-only rather than offering a toggle that the union
+			* would immediately undo. Turning image input off for such a model is
+			* deliberately not supported: the flag is the vendor's own capability answer,
+			* and enabling it only lets DSH attach an image the user still has to pick per
+			* message — it never sends one by itself.
+			*/
+			const rowImageIsAutomatic = (model) => model.multimodal === true;
+			/** The subscription row's verdict for the current account, when it has one. */
+			const payPlan = status.status === "signed-in" && status.payStatus !== void 0 ? payVerdict(status.payStatus) : void 0;
 			const activeContextBudgets = draft?.contextBudgets ?? savedContextBudgets;
 			const dirty = draft !== void 0;
 			/** Apply an edit to the ACTIVE tab's draft, seeding it from the current view. */
@@ -725,7 +777,8 @@ window.__ModuleLoader__.load({
 						enabled: regionOn(activeRegion),
 						lastCatalog: visibleModels.map((model) => ({
 							...model,
-							input: ["text"]
+							input: ["text"],
+							...model.multimodal === void 0 ? {} : { multimodal: model.multimodal }
 						})),
 						enabledModelIds: [...activeEnabledIds],
 						imageModelIds: [...activeImageIds].filter((id) => activeEnabledIds.has(id)),
@@ -887,7 +940,7 @@ window.__ModuleLoader__.load({
 								})
 							}) : null,
 							status.status === "signed-in" ? /* @__PURE__ */ (0, react_jsx_runtime.jsxs)(react_jsx_runtime.Fragment, { children: [
-								status.payStatus === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("div", {
+								status.payStatus === void 0 || payPlan === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("div", {
 									className: "dsm-trae-usage-list",
 									children: /* @__PURE__ */ (0, react_jsx_runtime.jsx)("div", {
 										className: "dsm-trae-usage-stats dsm-trae-usage-stats-two",
@@ -900,13 +953,13 @@ window.__ModuleLoader__.load({
 														className: "dsm-trae-usage-stat-label",
 														children: t("row.subscriptionLabel")
 													}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-														className: `dsm-trae-usage-stat-badge ${status.payStatus.hasPackage ? "dsm-trae-usage-stat-badge-on" : "dsm-trae-usage-stat-badge-off"}`,
-														children: t(status.payStatus.hasPackage ? "row.subscribed" : "row.noPackage")
+														className: `dsm-trae-usage-stat-badge ${payPlan.paid ? "dsm-trae-usage-stat-badge-on" : "dsm-trae-usage-stat-badge-off"}`,
+														children: t(payPlan.label)
 													})]
 												}),
 												/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 													className: "dsm-trae-usage-stat-value dsm-trae-usage-stat-value-general",
-													children: status.payStatus.inTrial && status.payStatus.trialEndTimeMs > 0 ? t("row.trialUntil", { date: formatDateTime(status.payStatus.trialEndTimeMs) }) : t(status.payStatus.hasPackage ? "row.subscribed" : "row.noPackage")
+													children: status.payStatus.inTrial && status.payStatus.trialEndTimeMs > 0 ? t("row.trialUntil", { date: formatDateTime(status.payStatus.trialEndTimeMs) }) : [t(payPlan.label), payPlan.tier].filter(Boolean).join(" · ")
 												}),
 												/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 													className: "dsm-trae-usage-stat-hint",
@@ -1077,8 +1130,9 @@ window.__ModuleLoader__.load({
 															className: "dsm-trae-model-image",
 															children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("input", {
 																type: "checkbox",
-																checked: activeImageIds.has(model.id),
-																disabled: settingsScope?.getSnapshot().writable !== true || saving,
+																checked: rowAcceptsImage(model),
+																disabled: rowImageIsAutomatic(model) || settingsScope?.getSnapshot().writable !== true || saving,
+																title: rowImageIsAutomatic(model) ? t("row.modelImageAuto") : void 0,
 																onChange: () => {
 																	toggleImage(model.id);
 																}
@@ -1270,11 +1324,12 @@ window.__ModuleLoader__.load({
 			"row.contextBudget": "DSH context budget",
 			"row.modelOutput": "Output {output}",
 			"row.modelImage": "Image",
+			"row.modelImageAuto": "Enabled automatically: Trae marks this model multimodal.",
 			"row.modelRate": "{rate}x credits",
 			"row.modelReasoning": "Reasoning: {efforts}",
 			"row.modelReasoningSupported": "Reasoning: supported",
 			"row.modelUnknown": "Unknown",
-			"row.modelCapabilityPending": "Image input is enabled only for models you select.",
+			"row.modelCapabilityPending": "Image input follows Trae's own multimodal flag, plus anything you tick yourself.",
 			"row.cheer": "Star on GitHub"
 		};
 		const zh = {
@@ -1349,11 +1404,12 @@ window.__ModuleLoader__.load({
 			"row.contextBudget": "DSH 上下文预算",
 			"row.modelOutput": "最大输出 {output}",
 			"row.modelImage": "图片",
+			"row.modelImageAuto": "已自动开启：Trae 把该模型标为多模态。",
 			"row.modelRate": "积分 {rate}x",
 			"row.modelReasoning": "推理强度：{efforts}",
 			"row.modelReasoningSupported": "推理：支持",
 			"row.modelUnknown": "未知",
-			"row.modelCapabilityPending": "仅对你手动勾选的模型启用图片输入。",
+			"row.modelCapabilityPending": "图片输入跟随 Trae 自身的多模态标注，并叠加你手动勾选的模型。",
 			"row.cheer": "鼓励一下"
 		};
 		//#endregion

package/lib/index.d.ts

@@ -120,6 +120,23 @@ interface TraeModelInfo {
   wireConfigName?: string;
   /** The directory function this model must be called through (see {@link TraeWireModel.function}). */
   wireFunction?: string;
+  /**
+   * Trae's own `multimodal` flag for this model, carried through from the
+   * directory that advertised it. It is the upstream's answer to "does this
+   * model read images", and {@link applyImageSelection} unions it with the
+   * user's explicit opt-ins.
+   *
+   * Deliberately NOT the same thing as `input`: `input` is what DSH is told
+   * this route accepts, while this field records what Trae said. Keeping both
+   * lets a future refresh re-derive `input` without losing the upstream fact,
+   * and lets the card explain WHY a model is image-capable.
+   *
+   * Absent means the directory did not say (an older `lastCatalog` saved before
+   * this field existed, a fallback row, or a directory entry that omitted the
+   * key). Absent is treated as false, never as "assume capable": sending an
+   * image to a model that cannot read one burns credits for a silent failure.
+   */
+  multimodal?: boolean;
 }
 /**
  * Bootstrap catalog: identity only where current Trae metadata has not been
@@ -175,7 +192,7 @@ declare function traeInputModalities(model: Pick<TraeModelInfo, 'input'>): TraeI
  * model discovery) use this decorated name.
  */
 declare function traeModelDisplayName(model: Pick<TraeModelInfo, 'name' | 'creditMultiplier'>): string;
-/** Apply the user's explicit image opt-ins; upstream and saved row hints are ignored. */
+/** Apply the user's explicit image opt-ins plus Trae's own multimodal flag. */
 declare function applyImageSelection(models: readonly TraeModelInfo[], selected: ReadonlySet<string>): TraeModelInfo[];
 /** One row from `get_detail_param`: the authoritative llm_utils_chat wire id + display name. */
 interface TraeWireModel {
@@ -239,7 +256,7 @@ type TraeContextBudget = number;
  * never to a fabricated number. Everything else keeps the dev window.
  */
 declare function applyContextBudgets(catalog: readonly TraeModelInfo[], budgets?: Readonly<Record<string, TraeContextBudget | undefined>>): TraeModelInfo[];
-/** Convert Trae metadata into text-only model rows; image support is user-owned configuration. */
+/** Convert Trae metadata into catalog rows; image input is decided by {@link applyImageSelection}. */
 declare function discoveredCatalog(models: readonly TraeDiscoveredModel[]): TraeModelInfo[];
 /**
  * Drop rows saved by older releases that generated `@1m` variant models, so a
@@ -1309,6 +1326,19 @@ interface TraePayStatus {
   enableSoloBuilder: boolean;
   enableSoloCoder: boolean;
   enableSoloWeb: boolean;
+  /**
+   * The account's plan identity (`user_pay_identity`), Trae's authoritative
+   * tier number: `0` is Free. Measured 2026-10-01 on a live SG community
+   * account, which reported `0` / `"Free"`.
+   *
+   * This is a DIFFERENT question from {@link hasPackage}, which asks whether a
+   * package/entitlement object exists. A paying member can hold a plan
+   * identity without any package row, which is why issue #19 saw "No active
+   * package" on a Pro account — the card only ever read `has_package`.
+   */
+  payIdentity?: number;
+  /** The plan identity as Trae spells it (`user_pay_identity_str`), e.g. `Free`. */
+  payIdentityStr?: string;
   /** The fission (referral) program window and cap, when exposed. */
   fission?: {
     startTimeMs: number;
@@ -1524,6 +1554,13 @@ interface TraeWebModel {
     defaultEffort?: string;
   };
   maxContextWindow?: number;
+  /**
+   * Trae's own image-capability flag, as the runtime catalog carries it. The
+   * card renders these rows with their image checkbox already ticked so the
+   * control matches what DSH is actually told (issue #16); absent means the
+   * directory did not declare it, which is not a capability claim either way.
+   */
+  multimodal?: boolean;
 }
 /**
  * Daily check-in state rendered below the credit stats.
@@ -1671,6 +1708,10 @@ interface TraeWebPayStatus {
   enableSoloBuilder: boolean;
   enableSoloCoder: boolean;
   enableSoloWeb: boolean;
+  /** Trae's authoritative tier number (`user_pay_identity`); `0` means Free. */
+  payIdentity?: number;
+  /** Trae's authoritative tier label (`user_pay_identity_str`), e.g. `Free`. */
+  payIdentityStr?: string;
   fission?: {
     startTimeMs: number;
     expireTimeMs: number;

package/lib/index.js

@@ -132,11 +132,45 @@ function traeInputModalities(model) {
 function traeModelDisplayName(model) {
 	return model.creditMultiplier === void 0 ? model.name : `${model.name} · x${model.creditMultiplier.toFixed(2)}`;
 }
-/** Apply the user's explicit image opt-ins; upstream and saved row hints are ignored. */
+/**
+* Whether one model should be offered image input.
+*
+* Two independent sources, unioned (issue #16):
+*
+*  - the user's explicit `imageModelIds` opt-in, which stays authoritative and
+*    can still switch an individual model on;
+*  - Trae's own `multimodal` flag, which is what makes "auto" work at all.
+*
+* Rationale: Trae's IDE enables its attach-image button purely from this flag
+* (`metadata.capabilities.vision`), with no user switch, so a model Trae itself
+* calls multimodal is one the vendor expects to receive images. Requiring a
+* manual tick for those made this plugin the only place where a vendor-declared
+* capability had to be re-enabled by hand.
+*
+* Why the flag alone is not enough, and the union is:
+*
+*  - the flag is advisory, not a contract. `llm_utils_chat` is a lightweight
+*    SOLO channel and this plugin has been burned by `4001 param is invalid`
+*    for ids that LOOKED valid, so the flag is not treated as proof of
+*    callability — a model still has to have survived the wire join to appear
+*    at all. The flag only decides image input, never model availability.
+*  - Users who already ticked a model keep that tick even after a refresh stops
+*    reporting the flag, so the union never silently revokes a working setup.
+*  - `input` remains the ONLY thing told to DSH: a model that ends up without
+*    `image` here is one DSH will degrade images to text for, which is the
+*    safe direction for a model whose capability we cannot confirm.
+*
+* `multimodal === true` is a strict test: absent, `false`, or any non-boolean
+* leaves the model text-only unless the user opted in.
+*/
+function traeModelAcceptsImage(model, selected) {
+	return selected.has(model.id) || model.multimodal === true;
+}
+/** Apply the user's explicit image opt-ins plus Trae's own multimodal flag. */
 function applyImageSelection(models, selected) {
 	return models.map((model) => ({
 		...model,
-		input: selected.has(model.id) ? ["text", "image"] : ["text"]
+		input: traeModelAcceptsImage(model, selected) ? ["text", "image"] : ["text"]
 	}));
 }
 /** Normalise a display name for cross-source joining. */
@@ -192,6 +226,7 @@ function mergeTraeModelSources(remote, wire) {
 			...model.maxContextWindow === void 0 ? {} : { maxContextWindow: model.maxContextWindow },
 			...creditMultiplier === void 0 ? {} : { creditMultiplier },
 			input: ["text"],
+			...model.multimodal === void 0 ? {} : { multimodal: model.multimodal },
 			reasoningSupported: model.reasoningSupported,
 			...model.reasoning === void 0 ? {} : {
 				reasoning: model.reasoning,
@@ -214,7 +249,7 @@ function applyContextBudgets(catalog, budgets = {}) {
 		...model.maxContextWindow !== void 0 && budgets[model.id] === model.maxContextWindow ? { contextWindow: model.maxContextWindow } : {}
 	}));
 }
-/** Convert Trae metadata into text-only model rows; image support is user-owned configuration. */
+/** Convert Trae metadata into catalog rows; image input is decided by {@link applyImageSelection}. */
 function discoveredCatalog(models) {
 	const result = [];
 	for (const model of models) result.push({
@@ -224,6 +259,7 @@ function discoveredCatalog(models) {
 		...model.maxContextWindow === void 0 ? {} : { maxContextWindow: model.maxContextWindow },
 		input: ["text"],
 		...model.creditMultiplier === void 0 ? {} : { creditMultiplier: model.creditMultiplier },
+		...model.multimodal === void 0 ? {} : { multimodal: model.multimodal },
 		reasoningSupported: model.reasoningSupported,
 		...model.reasoning === void 0 ? {} : {
 			reasoning: model.reasoning,
@@ -1906,7 +1942,7 @@ function applyReasoningEffort(body, effort, capability) {
 //#region src/solo.ts
 const TRAE_SOLO_FUNCTION = "solo_work_lite";
 /**
-* Directory functions to union per region, in priority order.
+* `llm_utils_chat` functions to union per region, in priority order.
 *
 * Trae spreads its callable roster across several SOLO-mode functions, and a
 * model is only usable through the one that lists it: `glm-5.3` is absent from
@@ -1914,14 +1950,32 @@ const TRAE_SOLO_FUNCTION = "solo_work_lite";
 * calling it through the former answers `4001 param is invalid`, through the
 * latter streams normally). Rather than betting on a single function, the
 * directory unions them; the first function to provide a config wins, so the
-* order below decides which wire name a model is called with.
+* order below decides which wire name AND which function a model is called with.
+*
+* THIS LIST IS NOT "directory functions" — it is the set of functions
+* `llm_utils_chat` actually accepts, because ownership here is what the chat
+* call replays. `solo_agent` is deliberately ABSENT (measured 2026-10-01 on a
+* live SG credential, issue #19): it is a Remote-agent roster function, not a
+* chat function. Every model sent with `function: "solo_agent"` — `gpt-5.4`,
+* `gpt-5.6-sol`, `gpt-6-astra`, `glm-5.2`, `kimi-k2.7-code` — answered
+* `event:error {"code":4011}` (message says "rate limit", it is really "not
+* served here"), while the same model with the same body answered normally
+* under `solo_work_remote` / `solo_work_lite` / `solo_agent_remote`.
+*
+* Note `get_detail_param` DOES answer `solo_agent` with a 43-entry list, so
+* asking it looks helpful while making 8 of the 10 exposed ai-region models
+* uncallable: it won precedence and stamped its own name onto configs it
+* cannot serve. `solo_agent_remote` is the chat-callable half of that roster
+* (it answers `minimax-m3`, `gemini-3.1-pro`, … that no `solo_work_*` lists)
+* and therefore takes the last slot — after the two IDE functions that serve
+* their shared configs without the Remote plan gate.
 */
 const TRAE_DIRECTORY_FUNCTIONS = {
 	cn: ["solo_work_remote", TRAE_SOLO_FUNCTION],
 	ai: [
-		"solo_agent",
 		"solo_work_remote",
-		TRAE_SOLO_FUNCTION
+		TRAE_SOLO_FUNCTION,
+		"solo_agent_remote"
 	]
 };
 const TRAE_SOLO_CHAT_PATH = "/api/agent/v3/llm_utils_chat";
@@ -2268,6 +2322,75 @@ function decodeTraeEvent(event) {
 }
 //#endregion
 //#region src/solo-bridge.ts
+/**
+* What Trae's in-stream error codes mean. Measured 2026-10-01 on a live SG
+* credential (issue #19), by sending the same body to several functions:
+*
+*  - `1005` — the model needs a paid plan. Trae answers HTTP 200 with
+*    `event:error {"code":1005,"message":"","extra":"{\"plan\":1}"}`, so the
+*    gate is only visible in `extra.plan`; `gpt-6-sol` returned it under both
+*    `solo_work_remote` and `solo_agent_remote` on a free account.
+*  - `4011` — the function the request named does not serve that model. Its
+*    message claims "exceeded the rate limit", which is misleading: the same
+*    model answers normally when sent under `solo_work_remote`.
+*  - `4001` — `config_name` is not one this function serves.
+*/
+const TRAE_ERROR_HINTS = {
+	1005: "Trae requires a paid plan for this model; the current account's plan does not cover it",
+	4001: "Trae does not serve this model under the SOLO function the request used (config_name rejected)",
+	4011: "Trae refused this model under the SOLO function the request used"
+};
+/** `extra` is a JSON *string* on the wire; unwrap it to read `plan`. */
+function traeErrorPlan(extra) {
+	let value = extra;
+	if (typeof value === "string") try {
+		value = JSON.parse(value);
+	} catch {
+		return;
+	}
+	if (typeof value !== "object" || value === null) return void 0;
+	const plan = value["plan"];
+	return typeof plan === "number" ? plan : void 0;
+}
+/** A readable message for an in-stream Trae failure, with its code preserved. */
+function describeTraeFailure(failure) {
+	const plan = traeErrorPlan(failure.extra);
+	const parts = [
+		failure.code === void 0 ? "Trae refused the request" : TRAE_ERROR_HINTS[failure.code] ?? "Trae refused the request",
+		...plan === void 0 ? [] : [`plan ${plan}`],
+		...failure.code === void 0 ? [] : [`Trae code ${failure.code}`]
+	];
+	const detail = failure.message.trim();
+	return detail === "" ? parts.join(" · ") : `${parts.join(" · ")} · upstream: ${detail}`;
+}
+/**
+* HTTP status + error kind for an in-stream failure, so the loopback shim can
+* answer with a real error instead of a truncated 200 stream.
+*
+* `1005` is a subscription gate: 402 is what makes clients stop rather than
+* retry, and it is truthful. Everything else is a client-side rejection.
+*/
+function classifyTraeFailure(failure) {
+	return failure.code === 1005 ? {
+		status: 402,
+		kind: "hard_credit"
+	} : {
+		status: 400,
+		kind: "client"
+	};
+}
+/** Recognise an in-stream failure event, or return undefined for anything else. */
+function traeFailureOf(decoded) {
+	if (decoded.type !== "unknown") return void 0;
+	const payload = decoded.data;
+	const code = typeof payload?.["code"] === "number" ? payload["code"] : void 0;
+	if (decoded.event !== "error" && !(code !== void 0 && code >= 4e3)) return void 0;
+	return {
+		...code === void 0 ? {} : { code },
+		message: typeof payload?.["message"] === "string" ? payload["message"] : "",
+		...payload?.["extra"] === void 0 ? {} : { extra: payload["extra"] }
+	};
+}
 function normalizeToolCalls(value) {
 	if (!Array.isArray(value)) return [];
 	const calls = [];
@@ -2292,6 +2415,10 @@ function normalizeToolCalls(value) {
 function bridgeTraeSoloStream(response, model) {
 	const source = response.body;
 	if (source === null) return new Response(null, { status: 502 });
+	return bridgeTraeSource(source, model);
+}
+/** {@link bridgeTraeSoloStream} over an already-opened byte stream. */
+function bridgeTraeSource(source, model) {
 	const id = `chatcmpl-${randomUUID().replaceAll("-", "").slice(0, 24)}`;
 	const created = Math.floor(Date.now() / 1e3);
 	const decoder = new TextDecoder();
@@ -2320,9 +2447,8 @@ function bridgeTraeSoloStream(response, model) {
 			const consume = (event) => {
 				const decoded = decodeTraeEvent(event);
 				if (decoded.type === "unknown") {
-					const payload = decoded.data;
-					const code = typeof payload?.["code"] === "number" ? payload["code"] : void 0;
-					if (decoded.event === "error" || code !== void 0 && code >= 4e3) upstreamError = new Error(typeof payload?.["message"] === "string" && payload["message"] !== "" ? payload["message"] : `Trae upstream error (code ${code ?? "?"})`);
+					const failure = traeFailureOf(decoded);
+					if (failure !== void 0) upstreamError = new Error(describeTraeFailure(failure));
 					return;
 				}
 				if (decoded.type === "delta") {
@@ -2392,6 +2518,123 @@ function bridgeTraeSoloStream(response, model) {
 		headers: { "Content-Type": "text/event-stream" }
 	});
 }
+/**
+* How far {@link openTraeSource} reads before giving up on finding a failure.
+* Trae opens a real answer with small `metadata` / `timing_cost` events, so a
+* healthy stream is recognised well inside these bounds; they exist only so a
+* pathological stream cannot stall the response headers.
+*/
+const PEEK_MAX_CHUNKS = 64;
+const PEEK_MAX_BYTES = 262144;
+/** Replay the bytes the peek consumed, then drain the rest of the upstream. */
+function replayTraeSource(buffered, reader, ended) {
+	let index = 0;
+	return new ReadableStream({
+		async pull(controller) {
+			const chunk = buffered[index];
+			if (chunk !== void 0) {
+				index += 1;
+				controller.enqueue(chunk);
+				return;
+			}
+			if (ended) {
+				controller.close();
+				return;
+			}
+			const next = await reader.read();
+			if (next.done) {
+				controller.close();
+				return;
+			}
+			controller.enqueue(next.value);
+		},
+		async cancel(reason) {
+			await reader.cancel(reason).catch(() => {});
+		}
+	});
+}
+/**
+* Open the upstream answer, failing fast when Trae refuses BEFORE producing any
+* output.
+*
+* This exists because Trae reports authorisation and subscription failures
+* inside a **200** SSE body. Relaying that as a 200 and aborting mid-stream
+* (what this bridge used to do) loses the code at the HTTP boundary: pi-ai only
+* sees a truncated stream and reports `Stream ended without finish_reason`,
+* which reads as a network fault and is retried five times — the exact symptom
+* in issue #19. Peeking until the first real output lets a refusal become a
+* normal error response with its code and plan intact.
+*
+* The peek is limited to the head of the stream and replayed verbatim, so a
+* healthy answer keeps streaming with no buffering beyond the first event.
+*/
+async function openTraeSource(response) {
+	const body = response.body;
+	if (body === null) return {
+		ok: false,
+		status: 502,
+		kind: "server",
+		message: "Trae upstream returned no response body"
+	};
+	const reader = body.getReader();
+	const decoder = new TextDecoder();
+	const sse = new SseDecoder();
+	const buffered = [];
+	let bytes = 0;
+	let ended = false;
+	try {
+		while (buffered.length < PEEK_MAX_CHUNKS && bytes < PEEK_MAX_BYTES) {
+			const next = await reader.read();
+			if (next.done) {
+				ended = true;
+				break;
+			}
+			buffered.push(next.value);
+			bytes += next.value.byteLength;
+			let answered = false;
+			for (const event of sse.push(decoder.decode(next.value, { stream: true }))) {
+				const decoded = decodeTraeEvent(event);
+				const failure = traeFailureOf(decoded);
+				if (failure !== void 0) {
+					await reader.cancel().catch(() => {});
+					return {
+						ok: false,
+						...classifyTraeFailure(failure),
+						message: describeTraeFailure(failure)
+					};
+				}
+				if (decoded.type !== "unknown") answered = true;
+			}
+			if (answered) break;
+		}
+	} catch (error) {
+		reader.releaseLock();
+		return {
+			ok: false,
+			status: 0,
+			kind: "server",
+			message: `transport error: ${String(error)}`
+		};
+	}
+	return {
+		ok: true,
+		source: replayTraeSource(buffered, reader, ended)
+	};
+}
+/**
+* Bridge an upstream answer, turning an early refusal into a real error result.
+*
+* The sibling {@link bridgeTraeSoloStream} stays synchronous and always answers
+* 200; it is the primitive, and this is the policy the bridge class uses.
+*/
+async function bridgeTraeSoloStreamOrFail(response, model) {
+	const opened = await openTraeSource(response);
+	if (!opened.ok) return opened;
+	return {
+		ok: true,
+		response: bridgeTraeSource(opened.source, model)
+	};
+}
 /** Native SOLO client wrapper used by the loopback OpenAI adapter. */
 var TraeSoloBridge = class {
 	upstream;
@@ -2439,10 +2682,7 @@ var TraeSoloBridge = class {
 		}
 		const result = await this.upstream.chatStream(prepared, signal);
 		if (!result.ok) return result;
-		return {
-			ok: true,
-			response: bridgeTraeSoloStream(result.response, model)
-		};
+		return await bridgeTraeSoloStreamOrFail(result.response, model);
 	}
 };
 //#endregion
@@ -2505,6 +2745,38 @@ function parseTraeRemoteModel(value) {
 //#region src/solo-remote.ts
 const TRAE_SOLO_REMOTE_BASE = "https://solo.trae.cn/api/remote/v1";
 /**
+* Remote-directory functions to ask for, per region.
+*
+* The remote `/models` answer is **grouped by function** and the groups are not
+* interchangeable (measured 2026-10-01, issue #19):
+*
+* ```
+* solo_agent        : 19 models   ← superset; also the ONLY source of
+*                                   gpt-6-astra / gpt-5.6-sol|terra|luna /
+*                                   glm-5.2 / gpt-5.5 / Seed-2.1-Turbo
+* solo_agent_remote : 10 models   ← what this client used to read, and all it
+*                                   read: the preferred group only
+* ```
+*
+* `solo_agent` ⊇ `solo_agent_remote`, so reading just the latter hid nine
+* models that the Trae IDE does show — exactly what the reporter saw. Both
+* groups are therefore requested, and every returned group is unioned.
+*
+* Asking for a function the gateway does not serve is harmless: unknown names
+* are ignored rather than rejected (`solo_work_lite` returns no group, and the
+* request still answers HTTP 200). The CN list is left as it was — the CN
+* rosters were not re-measured here, and the union keeps whichever groups come
+* back.
+*/
+const TRAE_REMOTE_DIRECTORY_FUNCTIONS = {
+	cn: ["solo_agent_remote", "solo_work_remote"],
+	ai: [
+		"solo_agent",
+		"solo_agent_remote",
+		"solo_work_remote"
+	]
+};
+/**
 * Region-scoped request dressing. The CN portal is `solo.trae.cn` with the
 * CN locale headers; the international directory lives on the shared
 * `coresg-normal.trae.ai` gateway and was verified (2026-09-15) with the
@@ -2554,16 +2826,16 @@ var TraeSoloRemoteCatalogClient = class {
 		const region = regionOfCredential(await this.options.credential());
 		const base = this.baseUrl ?? REGION_GATEWAYS[region].remote;
 		const headers = await this.headers(region);
-		const response = await this.fetchImpl(`${base}/models?functions=solo_agent_remote,solo_work_remote`, {
+		const functions = TRAE_REMOTE_DIRECTORY_FUNCTIONS[region].join(",");
+		const response = await this.fetchImpl(`${base}/models?functions=${functions}`, {
 			headers,
 			signal: signal ?? AbortSignal.timeout(3e4)
 		});
 		if (!response.ok) throw new Error(`SOLO remote models returned HTTP ${response.status}`);
 		const groups = (await response.json()).data?.list ?? [];
-		const preferred = groups.find((group) => group.function === "solo_agent_remote") ?? groups[0];
 		const seen = /* @__PURE__ */ new Set();
 		const models = [];
-		for (const raw of preferred?.models ?? []) {
+		for (const group of groups) for (const raw of group.models ?? []) {
 			const model = parseTraeRemoteModel(raw);
 			if (model === void 0 || seen.has(model.id)) continue;
 			seen.add(model.id);
@@ -3249,6 +3521,8 @@ var TraeUsageClient = class {
 		const fissionStart = asNumber(payload["solo_fission_start_time"]);
 		const fissionExpire = asNumber(payload["solo_fission_expire_time"]);
 		const fissionMax = asNumber(payload["solo_fission_max_usage"]);
+		const payIdentity = asNumber(payload["user_pay_identity"]);
+		const payIdentityStr = typeof payload["user_pay_identity_str"] === "string" && payload["user_pay_identity_str"].trim() !== "" ? payload["user_pay_identity_str"].trim() : void 0;
 		return {
 			isDollarUsageBilling: flag("is_dollar_usage_billing"),
 			hasPackage: flag("has_package"),
@@ -3259,6 +3533,8 @@ var TraeUsageClient = class {
 			enableSoloBuilder: flag("enable_solo_builder"),
 			enableSoloCoder: flag("enable_solo_coder"),
 			enableSoloWeb: flag("enable_solo_web"),
+			...payIdentity === void 0 ? {} : { payIdentity },
+			...payIdentityStr === void 0 ? {} : { payIdentityStr },
 			...fissionStart === void 0 || fissionExpire === void 0 || fissionMax === void 0 ? {} : { fission: {
 				startTimeMs: fissionStart,
 				expireTimeMs: fissionExpire,
@@ -4063,7 +4339,8 @@ const modelConfig = z.object({
 	maxTokens: z.number().step(1).min(1),
 	input: z.array(z.union(["text", "image"])),
 	creditMultiplier: z.number(),
-	wireFunction: z.string()
+	wireFunction: z.string(),
+	multimodal: z.boolean()
 });
 const regionStateConfig = z.object({
 	enabled: z.boolean().default(true).description("Whether this region's provider is offered to DSH (opt-out; false withdraws it entirely)"),

package/package.json

@@ -2,7 +2,7 @@
   "name": "dsh-connect-trae",
   "displayName": "DSH Connect Trae",
   "description": "把本机登录的 Trae 模型接入 DeepSeek Harness：国内版与国际版双供应商并行，提供用量/积分概览与每日签到领取。",
-  "version": "2.3.1",
+  "version": "2.5.0",
   "author": "dingminhua",
   "copyright": "Copyright (c) 2026 LaoDing",
   "license": "MIT",
@@ -76,42 +76,44 @@
   },
   "peerDependencies": {
     "@deepseek-ai/cordis": ">=4.0.4 <5.0.0",
-    "@deepseek-ai/dsh-atomic-write": ">=0.1.7-rc.1 <0.2.0-0",
-    "@deepseek-ai/dsh-attachment": ">=0.1.7-rc.1 <0.2.0-0",
-    "@deepseek-ai/dsh-client-ui-primitives": ">=0.1.7-rc.1 <0.2.0-0",
-    "@deepseek-ai/dsh-home-paths": ">=0.1.7-rc.1 <0.2.0-0",
-    "@deepseek-ai/dsh-host-webserver": ">=0.1.7-rc.1 <0.2.0-0",
-    "@deepseek-ai/dsh-llm": ">=0.1.7-rc.1 <0.2.0-0",
-    "@deepseek-ai/dsh-llm-pi-ai": ">=0.1.7-rc.1 <0.2.0-0",
-    "@deepseek-ai/dsh-settings": ">=0.1.7-rc.1 <0.2.0-0",
+    "@deepseek-ai/dsh-atomic-write": ">=0.1.7-rc.1 <0.3.0-0",
+    "@deepseek-ai/dsh-attachment": ">=0.1.7-rc.1 <0.3.0-0",
+    "@deepseek-ai/dsh-client-ui-primitives": ">=0.1.7-rc.1 <0.3.0-0",
+    "@deepseek-ai/dsh-home-paths": ">=0.1.7-rc.1 <0.3.0-0",
+    "@deepseek-ai/dsh-host-webserver": ">=0.1.7-rc.1 <0.3.0-0",
+    "@deepseek-ai/dsh-llm": ">=0.1.7-rc.1 <0.3.0-0",
+    "@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": ">=3.18.4 <4.0.0",
     "@earendil-works/pi-ai": ">=0.85.1",
     "react": "^18.2.0"
   },
   "devDependencies": {
     "@deepseek-ai/cordis": "^4.0.4",
-    "@deepseek-ai/dsh-atomic-write": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-attachment": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-client-locale": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-client-ui-primitives": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-client-ui-renderer": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-client-ui-settings": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-client-ui-settings-plugins": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-client-ui-slots": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-home-paths": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-host-webserver": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-llm": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-llm-pi-ai": "0.1.7-rc.1",
-    "@deepseek-ai/dsh-settings": "0.1.7-rc.1",
+    "@deepseek-ai/dsh-atomic-write": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-attachment": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-client-locale": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-client-ui-primitives": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-client-ui-renderer": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-client-ui-settings": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-client-ui-settings-plugins": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-client-ui-slots": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-home-paths": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-host-webserver": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-llm": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-llm-pi-ai": "0.2.0-rc.1",
+    "@deepseek-ai/dsh-settings": "0.2.0-rc.1",
     "@deepseek-ai/schemastery": "3.18.4",
     "@earendil-works/pi-ai": "^0.85.1",
     "@testing-library/dom": "^10.4.2",
     "@testing-library/react": "^16.3.3",
     "@types/node": "^22.20.0",
     "@types/react": "~18.3.31",
+    "@types/semver": "^7.7.0",
     "jsdom": "^25.0.1",
     "react": "^18.3.1",
     "react-dom": "18.3.1",
+    "semver": "^7.7.2",
     "tsdown": "^0.22.2",
     "typescript": "^5.9.3",
     "vitest": "^4.1.8"