Description

The "Set Up Task-Specific Sub-Agents" page (docs/inference/set-up-sub-agent.mdx) uses commands that exec into the openshell-cluster-nemoclaw container for a sandbox named my-assistant:

export SANDBOX=my-assistant
export DOCKER_CTR=openshell-cluster-nemoclaw
docker exec "$DOCKER_CTR" kubectl exec -n openshell "$SANDBOX" -c agent -- \
  cat /sandbox/.openclaw/openclaw.json > /tmp/openclaw.json

and a follow-on sequence that uploads openclaw.json and refreshes .config-hash.

The doc does not state that a NemoClaw sandbox must already be installed and running before these commands will work. On a host without NemoClaw, the commands fail with:

Error response from daemon: No such container: openshell-cluster-nemoclaw
(Block 1, FAIL_PREREQ)

Expected Result

The doc should clearly tell users that:

1. NemoClaw must be installed on the host, and
2. At least one sandbox (e.g., my-assistant) must already be onboarded and running — producing an openshell-cluster-nemoclaw container —

before they run the docker exec … kubectl exec commands.

Actual Result

The page jumps straight into docker exec … kubectl exec against openshell-cluster-nemoclaw without an explicit prerequisite section. Users who follow this page as their first touchpoint may run the commands on a machine without NemoClaw and see "No such container" errors, with no explanation that they needed to complete the quickstart first.

Suggested Fix

Add an explicit prerequisite sentence or short section near "Update the Sandbox Config", for example:

Prerequisite: NemoClaw is installed and at least one sandbox (for example my-assistant) is already onboarded and running. The openshell-cluster-nemoclaw container must exist. If you have not created a sandbox yet, follow the NemoClaw Quickstart and onboard my-assistant before running these commands.

Optionally, link "NemoClaw Quickstart" to the appropriate quickstart doc.

Severity / Priority

• Severity: Minor (content is mostly correct, but missing prereq causes confusing No such container errors)
• Priority: Medium (affects anyone using this page as their first sub-agent guide)

Additional Doc-Validation Findings (Execution Flow Gaps)

These are smaller flow gaps compared to the missing prerequisite, but fixing them would make the page runnable as a true how-to:

• Missing openclaw-patch.py setup: The "Update the Sandbox Config" section tells users to create /tmp/openclaw.updated.json but doesn't show how to obtain or run vlm-demo/vlm-subagent/openclaw-patch.py (no git clone or sample command to produce the file).

• Missing upload step for auth-profiles.json: "Add Sub-Agent Credentials" names /sandbox/.openclaw/agents/vision-operator/agent/auth-profiles.json but doesn't show how to upload auth-profiles.json into the sandbox — only the chown command is provided.

• No example log output for hot reload: After "Check /tmp/gateway.log…", there is no example log line to look for, so users don't know what a successful hot reload looks like.