Description
Description
Template 6047909 validates that Hermes startup recovers when /sandbox/.hermes/.tirith-install-failed contains download_failed. On vm2 with NemoClaw v0.0.53, the Hermes sandbox returns to Ready after active-runtime restart, but the download_failed marker remains and no tirith/download_failed cleanup retry message is emitted.Environment
Device: vm2, NVIDIA A100-SXM4-40GB
OS: Ubuntu 24.04.4 LTS
Architecture: x86_64
Docker: Docker version 29.5.2
OpenShell CLI: openshell 0.0.44
NemoClaw: nemoclaw v0.0.53
NemoHermes: nemohermes v0.0.53
OpenClaw: OpenClaw v2026.5.22 / Hermes Agent v2026.5.16Steps to Reproduce
- Start with a Ready Hermes sandbox named hermes-rerun.
- Plant the marker:
nemoclaw hermes-rerun exec -- sh -lc 'mkdir -p /sandbox/.hermes && printf download_failed > /sandbox/.hermes/.tirith-install-failed'
- Restart the active Docker-driver container:
docker restart openshell-hermes-rerun-...
- Wait 30 seconds and run:
nemohermes hermes-rerun status
- Check marker and startup log:
nemoclaw hermes-rerun exec -- sh -lc 'ls /sandbox/.hermes/.tirith-install-failed 2>&1 || echo marker absent'
nemoclaw hermes-rerun exec -- sh -lc 'grep -i "tirith|download_failed" /tmp/nemoclaw-start.log | tail -20 || true'Expected Result
Hermes startup should remove the download_failed marker or otherwise complete the documented recovery path. The marker check should print marker absent, and startup logs should include the expected retry/cleanup message.
Actual Result
The sandbox returned to Phase=Ready and Inference healthy, but the marker still existed:
/sandbox/.hermes/.tirith-install-failed
The startup log grep produced no tirith/download_failed cleanup evidence.
Bug Details
| Field |
Value |
| Priority |
Unprioritized |
| Action |
Dev - Open - To fix |
| Disposition |
Open issue |
| Module |
Machine Learning - NemoClaw |
| Keyword |
NemoClaw, NemoClaw_Agent&Skills, NEMOCLAW_GH_SYNC_APPROVAL, NemoClaw_Sandbox |
[NVB#6239962]
Description
Description
Template 6047909 validates that Hermes startup recovers when /sandbox/.hermes/.tirith-install-failed contains download_failed. On vm2 with NemoClaw v0.0.53, the Hermes sandbox returns to Ready after active-runtime restart, but the download_failed marker remains and no tirith/download_failed cleanup retry message is emitted.Environment
Device: vm2, NVIDIA A100-SXM4-40GB
OS: Ubuntu 24.04.4 LTS
Architecture: x86_64
Docker: Docker version 29.5.2
OpenShell CLI: openshell 0.0.44
NemoClaw: nemoclaw v0.0.53
NemoHermes: nemohermes v0.0.53
OpenClaw: OpenClaw v2026.5.22 / Hermes Agent v2026.5.16Steps to Reproduce
nemoclaw hermes-rerun exec -- sh -lc 'mkdir -p /sandbox/.hermes && printf download_failed > /sandbox/.hermes/.tirith-install-failed'
docker restart openshell-hermes-rerun-...
nemohermes hermes-rerun status
nemoclaw hermes-rerun exec -- sh -lc 'ls /sandbox/.hermes/.tirith-install-failed 2>&1 || echo marker absent'
nemoclaw hermes-rerun exec -- sh -lc 'grep -i "tirith|download_failed" /tmp/nemoclaw-start.log | tail -20 || true'Expected Result
Hermes startup should remove the download_failed marker or otherwise complete the documented recovery path. The marker check should print marker absent, and startup logs should include the expected retry/cleanup message.
Actual Result
The sandbox returned to Phase=Ready and Inference healthy, but the marker still existed:
/sandbox/.hermes/.tirith-install-failed
The startup log grep produced no tirith/download_failed cleanup evidence.
Bug Details
[NVB#6239962]