inject 声明的 workflows 服务在标准 web profile 无提供方：条目永久 pending，整个 profile 约 40 秒后静默退出

[xiaoyuyu6420]

环境

dsh web profile + @dsh-external/dsh-deep-research（当前 main，c0b329e）

症状

启动 dsh --profile web 后约 40 秒进程静默退出（exit 0），无任何错误输出。

复现与对照

1. 安装并启用本插件，启动 profile → 必现静默退出
2. 仅禁用本插件（其余全部不变）→ profile 完全正常存活

根因

src/index.ts（及已提交产物 lib/types/index.js）声明 export const inject = ["tools", "workflows"]。标准 web 组合不提供 workflows 服务，inject 永不满足，条目永久 pending；cordis Loader 的 EntryGroup 用 Promise.allSettled 等所有条目，一棵树被挂死，进程最终静默退出。最新 commit c0b329e（"docs: declare workflows-provider requirement"）已意识到该依赖，但只改了文档未改代码——缺 provider 的 profile 仍会被拖死。

期望

无 workflows provider 的 profile 照常加载；deep_research 工具被调用时返回明确的"引擎不可用"错误（指向 tui/headless 等可用组合）。

[ydh0411]

补充一个复现环境：**DSH Desktop 2.0.0（macOS, Electron 桌面版）**同样必现。桌面 profile（@deepseek-ai/dsh-base + @deepseek-ai/dsh-web-app bundles）同样不提供 workflows 服务，与根因完全吻合：

• 条目永久 pending → 加载器报 1 entry did not activate → 应用直接打不开（不只是静默退出，桌面版会直接启动失败）
• 临时处理：从 profile 移除 @dsh-external/dsh-deep-research 后应用恢复正常

支持本 issue 的期望：缺 provider 时 profile 应照常加载、deep_research 工具被调用时返回明确的"引擎不可用"，而不是把整个 profile 拖死。

[kenchioua]

Same problem here — also reproduced on Windows + DSH Desktop (Links2008 build) + web profile.

Environment: DSH Desktop 2.2.1 (Windows), @deepseek-ai/dsh-base + @deepseek-ai/dsh-web-app bundles, plugin installed via dsh plugin --profile web add git+https://github.com/dsh-external/dsh-deep-research.git.

Symptoms:

• After installing the plugin and restarting, the app failed to start / backend exited (the web profile bundle silently exits ~40s after launch, matching this issue's root cause: no workflows provider in the standard web composition, entry stays pending forever).
• Removing @dsh-external/dsh-deep-research from the profile and clearing the leftover copy under ~/.dsh restored the app, and the profile loads normally again.

+1 to the fix direction in PR #4 (make workflows an optional runtime capability instead of a hard inject dependency): profiles without the provider should load normally, and calling deep_research should return an explicit "engine unavailable" error pointing to tui/headless, rather than dragging the whole profile down. Looking forward to seeing it merged — happy to test once it lands.

Thanks for the work on this plugin!

[xiaoyuyu6420]

Thanks for the detailed repro — the DSH Desktop 2.2.1 (Windows) + dsh-base/dsh-web-app bundle environment is a useful new data point beyond my original report, and both symptoms match exactly: the ~40s silent exit after launch, and full recovery once the plugin is removed from the profile and the leftover copy under ~/.dsh is cleared.

Status update for anyone else hitting this family of bug: the sibling fix for dsh-inspect (omdsh-dev/dsh-inspect#5 — load without a workflows provider, report clearly at tool-call time) has been merged. For dsh-deep-research, PR #4 implements the direction you +1'd: workflows becomes an optional runtime capability instead of a hard inject dependency, so the profile starts normally and the tool reports an explicit error at call time when no provider is present — which covers your scenario.