Hermes "backend exited before it became ready (SIGTERM)"

Likely cause

The backend process starts and dies before it can report ready, and the cause is almost always a broken virtual environment left behind by a partial update rather than anything in your agent config. Reported variants: a `ModuleNotFoundError` for a dependency that never installed (yaml, annotated_doc) after a git code update succeeded but the dependency install failed, leaving stale packages in the venv; Windows auto-update failing every time because .pyd files in the venv are locked by still-running Hermes processes, so recreation cannot complete (issues #43268, #38122); and a dirty bundled repo checkout leaving the update in a loop. There is also a config-shaped variant: pinning HERMES_DASHBOARD_SESSION_TOKEN in .env puts Desktop LOCAL mode into a SIGTERM boot loop (issue #38575).

The fix

1. 1 Run the backend in the foreground from a terminal and read the actual traceback. The SIGTERM message is the supervisor's summary; the real error is the Python exception underneath it.
2. 2 If you pinned HERMES_DASHBOARD_SESSION_TOKEN in ~/.hermes/.env, remove that line. It puts Desktop LOCAL mode into a boot loop.
3. 3 On Windows, fully quit Hermes and Hermes Desktop and confirm no hermes processes remain before updating — locked .pyd files are what makes venv recreation fail silently.
4. 4 Delete and recreate the virtual environment, then reinstall. A partial update leaves stale packages that a repeat update will not repair.
5. 5 If the traceback names a missing module, install that dependency explicitly before retrying the update.