Description
[Description]
On the “Customize the Sandbox Network Policy” page, the “Static Changes” section mixes instructions for editing the baseline openclaw-sandbox.yaml file with usage of nemoclaw policy-add. The wording implies that policy-add is an alternative way to perform static changes (i.e., edits that persist across sandbox recreation), while in reality policy-add operates on the live policy of a running sandbox. This makes it easy for readers to assume that running policy-add has permanently updated the sandbox’s baseline network policy, when it only affects the current sandbox session unless they also update the YAML or ship a preset file.
[Environment]
-
Component: NemoClaw documentation
-
Page: “Customize the Sandbox Network Policy”
-
Affected sections:
-
“Static Changes → Edit the Policy File”
-
Related references: Network Policy / Presets docs and CLI commands for
policy-add, policy-remove, and policy-list.
[Steps to Reproduce]
-
Open the “Customize the Sandbox Network Policy” page in the NemoClaw docs.
-
Scroll to the Static Changes section, under Edit the Policy File.
-
Read the instructions:
-
First, the doc tells you to open and edit
nemoclaw-blueprint/policies/openclaw-sandbox.yaml. -
Immediately after that, it says:
“If you only need one of the built-in presets, use nemoclaw policy-add instead of editing YAML by hand:
nemoclaw my-assistant policy-add”
Compare this wording to the CLI/Network Policy documentation, which describes policy-add as:
Applying a preset to a running sandbox, by fetching the live policy via openshell policy get --full, merging in the preset’s network_policies, and writing the merged result back with openshell policy set.
Not described as an operation that directly edits the baseline openclaw-sandbox.yaml file on disk.
[Expected Result]
-
The Static Changes section should describe only baseline (static) policy modifications, for example:
-
Editing
openclaw-sandbox.yaml directly, and/or -
Merging preset entries into the baseline file, then applying them by re-running
nemoclaw onboard or using nemoclaw rebuild so the updated baseline is picked up on sandbox recreation.
-
If
nemoclaw policy-add is mentioned in this context, the docs should make it explicit that:
-
policy-add modifies the live policy of a running sandbox, not the baseline YAML file. -
To make a preset-based change static / persistent across sandbox recreation, the corresponding entries must be:
-
Included in
openclaw-sandbox.yaml, or -
Shipped as a preset under
policies/presets/ and reapplied on rebuild/onboard.
[Actual Result]
-
In Static Changes → Edit the Policy File, the doc:
-
First instructs the user to edit
openclaw-sandbox.yaml. -
Then immediately says:
“If you only need one of the built-in presets, use nemoclaw policy-add instead of editing YAML by hand:
nemoclaw my-assistant policy-add”
This presentation suggests that policy-add is simply an alternative to editing YAML by hand to achieve a static change, implying that the preset becomes part of the baseline.
Elsewhere, however, policy-add is described as operating on the running sandbox’s policy via “get live policy → merge preset → set live policy,” with no indication that it writes back to openclaw-sandbox.yaml.
There is no explicit warning in this section that:
policy-add alone does not modify openclaw-sandbox.yaml, and
Its effects may be lost when the sandbox is stopped or recreated, unless the preset or YAML file is also updated.
[Impact / Notes]
Suggested Fix:
-
In the Static Changes section, either:
-
Remove
policy-add from that section entirely and discuss it only under Dynamic Changes, or -
Explicitly annotate its behavior. For example:
“To apply a built-in preset to a running sandbox without editing YAML, use nemoclaw policy-add. This updates the sandbox’s live policy only. To make that change static (so it survives sandbox recreation), also include the preset’s entries in openclaw-sandbox.yaml or ship the preset file in policies/presets/ and re-run nemoclaw onboard or nemoclaw rebuild.”
-
Add a short cross-link or callout to the Dynamic Changes and Policy Presets sections so the live-versus-baseline distinction is explicit.
Bug Details
| Field |
Value |
| Priority |
Unprioritized |
| Action |
Dev - Open - To fix |
| Disposition |
Open issue |
| Module |
Machine Learning - NemoClaw |
| Keyword |
NemoClaw, NemoClaw_Docs, NEMOCLAW_GH_SYNC_APPROVAL |
[NVB#6188952]
Description
[Description]
On the “Customize the Sandbox Network Policy” page, the “Static Changes” section mixes instructions for editing the baseline
openclaw-sandbox.yamlfile with usage ofnemoclaw policy-add. The wording implies thatpolicy-addis an alternative way to perform static changes (i.e., edits that persist across sandbox recreation), while in realitypolicy-addoperates on the live policy of a running sandbox. This makes it easy for readers to assume that runningpolicy-addhas permanently updated the sandbox’s baseline network policy, when it only affects the current sandbox session unless they also update the YAML or ship a preset file.[Environment]
policy-add,policy-remove, andpolicy-list.[Steps to Reproduce]
nemoclaw-blueprint/policies/openclaw-sandbox.yaml.Compare this wording to the CLI/Network Policy documentation, which describes
policy-addas:Applying a preset to a running sandbox, by fetching the live policy via
openshell policy get --full, merging in the preset’snetwork_policies, and writing the merged result back withopenshell policy set.Not described as an operation that directly edits the baseline
openclaw-sandbox.yamlfile on disk.[Expected Result]
openclaw-sandbox.yamldirectly, and/ornemoclaw onboardor usingnemoclaw rebuildso the updated baseline is picked up on sandbox recreation.nemoclaw policy-addis mentioned in this context, the docs should make it explicit that:policy-addmodifies the live policy of a running sandbox, not the baseline YAML file.openclaw-sandbox.yaml, orpolicies/presets/and reapplied on rebuild/onboard.[Actual Result]
openclaw-sandbox.yaml.This presentation suggests that
policy-addis simply an alternative to editing YAML by hand to achieve a static change, implying that the preset becomes part of the baseline.Elsewhere, however,
policy-addis described as operating on the running sandbox’s policy via “get live policy → merge preset → set live policy,” with no indication that it writes back toopenclaw-sandbox.yaml.There is no explicit warning in this section that:
policy-addalone does not modifyopenclaw-sandbox.yaml, andIts effects may be lost when the sandbox is stopped or recreated, unless the preset or YAML file is also updated.
[Impact / Notes]
nemoclaw my-assistant policy-addhas permanently updated the sandbox’s static network policy, when it has only changed the live policy for the current sandbox session.
The confusion is reinforced by the section title “Static Changes” and the immediate juxtaposition of “edit YAML” and “or use
policy-addinstead,” without clarifying their different persistence semantics.The same page later describes dynamic changes and
openshell policy setas session-scoped, but it doesn’t clearly tiepolicy-addin the Static section back to those dynamic semantics, which can lead to incorrect assumptions about durability.Suggested Fix:
policy-addfrom that section entirely and discuss it only under Dynamic Changes, orBug Details
[NVB#6188952]