Agent Diagnostic
antigravity X claude
Description
Environment
| Element |
Detail |
| Machine |
Apple M3, 16GB RAM |
| OS |
macOS (arm64) |
| Docker |
Docker Desktop (also tested with Colima) |
| OpenShell CLI |
v0.0.12 |
| NemoClaw |
v0.1.0 |
| Network |
Overseas connection (Guadeloupe, French Caribbean) |
Problem
nemoclaw onboard and openshell gateway start fail systematically
at step [2/7] with:
K8s namespace not ready: timed out waiting for namespace
'openshell' to exist
Observed startup times (from container logs)
| Pod |
Startup time |
| local-path-provisioner |
~1min08s |
| coredns |
~1min09s |
| metrics-server |
~1min10-31s |
| agent-sandbox-controller |
~1min48s |
The gateway actually finishes initializing ~10-18 seconds
AFTER OpenShell's hardcoded timeout triggers and destroys
the container.
Root cause
The timeout in the openshell binary is too short for:
- Apple Silicon ARM Macs (k3s runs slower in Docker on ARM)
- Overseas/slow connections (images not cached between attempts
— container is destroyed on each crash, forcing a full re-pull)
Requested fix
Please add a --timeout <seconds> flag to openshell gateway start
to allow users to extend the namespace readiness timeout.
Alternatively, preserve the Docker image cache between failed
attempts to eliminate re-pull time on retry.
---
### Reproduction Steps
## Steps to Reproduce
1. Install NemoClaw on Apple Silicon Mac (M3, arm64)
```bash
curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash
-
Start Docker Desktop (or Colima with --cpu 4 --memory 8)
-
Run onboarding
-
At step [2/7] "Starting OpenShell gateway", observe:
- Docker pulls ~500MB of images from scratch on every attempt
(images are destroyed on each crash, no caching between retries)
- k3s initializes: metrics-server takes ~1min10-31s,
agent-sandbox-controller takes ~1min48s
- OpenShell timeout triggers at ~1min30-40s, before k3s is ready
- Container is destroyed, error displayed:
timed out waiting for namespace 'openshell' to exist
-
Retry any number of times — result is identical every time
Expected behavior
Gateway starts successfully once all k3s pods are ready.
Actual behavior
OpenShell abandons and destroys the container before k3s
finishes initializing, making installation impossible on
Apple Silicon + overseas network.
Environment
MacOS M3 16GO RAM
Logs
Agent-First Checklist
Agent Diagnostic
antigravity X claude
Description
Environment
Problem
nemoclaw onboardandopenshell gateway startfail systematicallyat step [2/7] with:
Observed startup times (from container logs)
The gateway actually finishes initializing ~10-18 seconds
AFTER OpenShell's hardcoded timeout triggers and destroys
the container.
Root cause
The timeout in the
openshellbinary is too short for:— container is destroyed on each crash, forcing a full re-pull)
Requested fix
Please add a
--timeout <seconds>flag toopenshell gateway startto allow users to extend the namespace readiness timeout.
Alternatively, preserve the Docker image cache between failed
attempts to eliminate re-pull time on retry.
Start Docker Desktop (or Colima with
--cpu 4 --memory 8)Run onboarding
At step [2/7] "Starting OpenShell gateway", observe:
(images are destroyed on each crash, no caching between retries)
agent-sandbox-controller takes ~1min48s
timed out waiting for namespace 'openshell' to existRetry any number of times — result is identical every time
Expected behavior
Gateway starts successfully once all k3s pods are ready.
Actual behavior
OpenShell abandons and destroys the container before k3s
finishes initializing, making installation impossible on
Apple Silicon + overseas network.
Environment
MacOS M3 16GO RAM
Logs
Agent-First Checklist
debug-openshell-cluster,debug-inference,openshell-cli)