plugin_manager writes succeed, but profile reload always fails (dsh: profile reload requires the root Include entry)

[Mal-Suen]

Environment

• dsh 0.2.1-rc.2, npm global install, Windows 11, web profile (dsh web behind a reverse proxy, if that matters)
• Reproduced in two different sessions (one healthy, one degraded) on the same host

Steps to reproduce

1. Call the plugin_manager tool with any write action, e.g. install_bundle with a local tarball, or set_plugin/set_bundle enable/disable.
2. The pnpm step succeeds (package.json and node_modules update correctly).
3. The operation result is application: "failed" with:

dsh: profile reload requires the root Include entry

4. The change only takes effect after a full DSH restart.

Expected: live profiles apply plugin changes immediately (per the plugin_manager contract).

Actual: every write operation reports the same diagnostic; nothing applies until restart.

Root-cause pointer

The diagnostic comes from recompose() in @deepseek-ai/dsh-app-boot (lib/index.js, the bootstrapIncludes.get(ctx) check returning undefined). It looks like the root Include entry is not registered in bootstrapIncludes for this launch mode, so the reload path always throws. I have not determined why the entry is missing — possibly specific to the web-profile launch path.

Two related observations from the same host

1. A failed operation leaves package.json.lock behind. In our case a stale lock (containing a long-dead PID) blocked subsequent installs for ~14 hours until removed manually. The lock file was not cleaned up when the operation failed.
2. In the affected sessions, the runtime context reported the implementation checkout as the profile directory itself rather than the npm installation directory — possibly the same launch-mode quirk, reported here in case it is related.

Happy to provide logs or test a fix. Thanks!

[ltxhhz]

Run where.exe dsh and check if the drive letter is lowercase. If it is, update it to uppercase in the PATH environment variable.
This is how I resolved it.