fix: follow native agent gateway lifecycle (#49)

Co-authored-by: mmckeen-nv <mmckeen-nv@users.noreply.github.com>
This commit is contained in:
mmckeen-nv
2026-09-17 23:10:53 -07:00
committed by GitHub
co-authored by mmckeen-nv
parent 85c6777c55
commit 4e2d51e84a
5 changed files with 31 additions and 78 deletions
+2
View File
@@ -225,6 +225,8 @@ The dashboard shells out to the OpenShell CLI for several operations:
The **Quick Deploy New NemoClaw Sandbox** path clones a registered image through OpenShell, but it deliberately does not write the clone into NemoClaw's private `sandboxes.json`. Current NemoClaw entries include lifecycle identity, workload provenance, and agent-specific state that must be issued by NemoClaw itself. The clone remains visible and operable through OpenShell Control; use **Fresh NemoClaw Image** when the new sandbox must participate in NemoClaw-managed lifecycle operations.
Sandbox gateway restarts use NemoClaw's native-agent lifecycle command. The controller does not kill or launch OpenClaw or Hermes gateway processes directly; if NemoClaw cannot verify the restart, the request fails without bypassing the agent-owned lifecycle boundary.
OpenClaw dashboard access is loopback-only inside the host/sandbox context, so the UI uses local proxy routes:
- `/api/openshell/dashboard/proxy`