Resolve the two output threads in #3456 left after the core dead-loop fix
landed via #3459 + #3434:

Sub-bug #3 — `src/lib/onboard.ts` printed
  `nemoclaw <name> destroy --yes && nemoclaw onboard --gpu`
with a literal `<name>` placeholder, and assumed at least one sandbox
was registered. When the GPU-passthrough mismatch hit on the State B
re-run path with an empty registry (the dead-loop case), the hint was
not actionable. Replace with a registry-aware helper at
`src/lib/onboard/gpu-recovery.ts` that renders the right shape:
  - empty registry → suggest `nemoclaw uninstall && nemoclaw onboard --gpu`
  - one sandbox → suggest destroy --yes --cleanup-gateway for that name
  - multiple sandboxes → list each, only the last gets --cleanup-gateway

Sub-bug #4 — `src/lib/actions/uninstall/run-plan.ts` printed
  `Destroyed gateway 'nemoclaw' skipped`
when the openshell destroy no-op'd (gateway already gone) — the
"Destroyed … skipped" wording was self-contradictory. Extend
`runOptional` with an `onSkip` option; route the gateway destroy to
emit `Gateway 'nemoclaw' already removed or unreachable` on no-op.

Tests:
- `src/lib/onboard/gpu-recovery.test.ts` (6 tests): forbid literal
  `<name>` placeholder anywhere in the output; cover empty / single /
  multi-sandbox cases; defensive filter on whitespace names so a
  `nemoclaw  destroy` rendering can never happen.
- `src/lib/actions/uninstall/run-plan.test.ts`: assert the new
  "already removed or unreachable" wording and the absence of the
  "Destroyed gateway 'nemoclaw' skipped" string.

The core dead loop itself (sub-bugs #1, #2 and State B GPU mismatch)
is already addressed by #3459 + #3434 + #3483; #3456 will close once
this lands. See the #3456 status comment for the full mapping.

Refs #3456. Mirrors (and tightens) the approach in the closed PR #3464,
which left the literal `<name>` placeholder in tests per CodeRabbit
feedback that was never addressed.

Signed-off-by: Charan Jagwani <charjags100@gmail.com>

…e test

Why: rebasing onto main pulled in NVIDIA#3434's new
"can override the sandbox inference base URL" test, which was written
against the pre-WeChat 13-parameter signature of patchStagedDockerfile.
The WeChat branch added a wechatConfig: LooseObject parameter at slot
13, so the test's `false` (intended for darwinVmCompat) was landing in
wechatConfig and TypeScript flagged the boolean-to-LooseObject mismatch.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

## Summary
- Bump the docs metadata and version switcher to `0.0.41`.
- Add v0.0.41 release notes plus operator guidance for OpenShell
pinning, Docker bridge reachability, Local Ollama proxy reachability,
and Docker GPU onboarding diagnostics.
- Refresh generated `nemoclaw-user-*` skills from the updated docs.

## Source summary
- #3434 -> `docs/reference/commands.md`,
`docs/reference/troubleshooting.md`, `docs/about/release-notes.md`:
Document Linux Docker-driver GPU onboarding behavior, diagnostics,
cleanup guidance, and the `NEMOCLAW_DOCKER_GPU_PATCH` troubleshooting
escape hatch.
- #3483 -> `docs/about/release-notes.md`: Note that `nemoclaw uninstall`
removes all installer-managed OpenShell helper binaries unless
`--keep-openshell` is passed.
- #3446 -> `docs/reference/commands.md`,
`docs/reference/troubleshooting.md`, `docs/about/release-notes.md`:
Document blueprint-driven OpenShell install pin resolution and fallback
behavior.
- #3472 -> `docs/inference/use-local-inference.md`,
`docs/reference/troubleshooting.md`, `docs/about/release-notes.md`:
Document sandbox-side Local Ollama auth proxy reachability checks and
firewall remediation.
- #3459 -> `docs/reference/commands.md`,
`docs/reference/troubleshooting.md`, `docs/about/release-notes.md`:
Document Docker-driver sandbox-to-gateway reachability checks and
firewall remediation.

## Test plan
- `python3 scripts/docs-to-skills.py docs/ .agents/skills/ --prefix
nemoclaw-user`
- `make docs`
- `git diff --check`
- `npm run build:cli`
- `npm run typecheck:cli`
- pre-commit hooks during `git commit`

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **New Features**
* Added `nemoclaw inference get` command to check current inference
settings
* Improved gateway health validation with Linux firewall remediation
guidance

* **Bug Fixes**
  * Enhanced proxy readiness validation with sandbox network path probes
  * Improved local Ollama route onboarding with rerun-safe fixes
  * Better sandbox-to-gateway connectivity detection

* **Documentation**
* Expanded troubleshooting guidance for firewall and connectivity issues
* Updated CLI reference with new command and environment variable
documentation
  * Added gateway binding and Docker-driver GPU compatibility guidance

<!-- review_stack_entry_start -->

[![Review Change
Stack](https://storage.googleapis.com/coderabbit_public_assets/review-stack-in-coderabbit-ui.svg)](https://app.coderabbit.ai/change-stack/NVIDIA/NemoClaw/pull/3531)

<!-- review_stack_entry_end -->

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

…3520)

> **Draft for visibility.** Issue-autopilot Stages 4-5 of #3456. Will
mark ready once batch self-review + CI complete.

## Summary

Closes the two remaining output threads in #3456 after the core
dead-loop fix already landed on `main` (via #3459, #3434, #3483). Full
sub-bug mapping in the [#3456 status
comment](#3456 (comment)).

- **Sub-bug #3** — `nemoclaw <name> destroy --yes` recovery hint
replaced with a registry-aware helper.
- **Sub-bug #4** — `Destroyed gateway 'nemoclaw' skipped`
self-contradictory wording replaced with `Gateway 'nemoclaw' already
removed or unreachable`.

## Acceptance criteria mapping

| Sub-bug | Resolution | Evidence |
|---|---|---|
| #1 dead loop | Already fixed on main (#3459) | out of scope |
| #2 firewall diagnostic | Already fixed on main (#3459) | out of scope
|
| **#3** literal `<name>` placeholder | **This PR** |
`src/lib/onboard/gpu-recovery.ts` + `onboard.ts:10387-10405` |
| **#4** misleading "skipped" wording | **This PR** |
`src/lib/actions/uninstall/run-plan.ts:210-228, 407-414` |
| #5 uninstall residuals | Already fixed on main (#3483) | out of scope
|

## Behavior matrix

`gpuPassthroughRecoveryLines(names)`:

| Input | Suggestion |
|---|---|
| `null` / `[]` | `nemoclaw uninstall && nemoclaw onboard --gpu` |
| one sandbox | `nemoclaw <name> destroy --yes --cleanup-gateway &&
nemoclaw onboard --gpu` |
| many sandboxes | each `destroy --yes`, only the last gets
`--cleanup-gateway` |

## Test plan

```
npm run typecheck:cli
npx vitest run src/lib/onboard/gpu-recovery.test.ts src/lib/actions/uninstall/run-plan.test.ts
```

22 tests pass (6 new + 16 existing).

## Notes for reviewers

- This is the work [#3464
attempted](#3464); that PR was
closed without merging after CodeRabbit asked for the `<name>`
placeholder to be forbidden in tests via negative assertion. This PR
adopts that refinement.
- `runOptional` extension is backwards-compatible — existing callers
without `onSkip` get the original wording.

Closes #3456 once merged.

---------

Signed-off-by: Charan Jagwani <charjags100@gmail.com>
Co-authored-by: Charan Jagwani <charjags100@gmail.com>
Co-authored-by: Carlos Villela <cvillela@nvidia.com>

…d for proc/comm write (#3515)

## Summary

The Docker-driver GPU patch
([src/lib/onboard/docker-gpu-patch.ts](src/lib/onboard/docker-gpu-patch.ts))
recreates the OpenShell-managed sandbox container with `--gpus all` plus
a reconstruction of the baseline container's flags, then runs a
three-step GPU proof in
[src/lib/onboard/initial-policy.ts](src/lib/onboard/initial-policy.ts) —
including `PROC_COMM_WRITE_PROBE`, which writes to
`/proc/<pid>/task/<tid>/comm`.

On DGX Spark hosts (#3511) and other Docker-driver Linux hosts where the
OpenShell-created baseline container's `CapAdd` does not include
`SYS_PTRACE` and its `SecurityOpt` does not include
`apparmor=unconfined`, the recreate inherits a flag set the kernel/LSM
stack rejects for that proc write. The proof aborts onboarding:

```
✗ GPU proof failed: /proc/<pid>/task/<tid>/comm write
Error: GPU proof failed: /proc/<pid>/task/<tid>/comm write (status 2):
  sh: 1: cannot create /proc/<pid>/task/<tid>/comm: Permission denied
```

The reporter confirmed a bare `docker run --rm --gpus all ubuntu sh -c
"echo test > /proc/1/task/1/comm"` succeeds on the same host, so the
kernel itself allows the write under the right Docker flags — the
problem is which flags reach the recreated container.

This PR makes the GPU recreate self-sufficient for the operations the
GPU proof checks, regardless of the non-GPU baseline:

- **Always inject `--cap-add SYS_PTRACE`** into the clone, deduped via a
`Set` so baselines that already have it stay flat.
- **Inject `--security-opt apparmor=unconfined`** only when the baseline
did *not* pin a specific apparmor profile. Docker rejects duplicate
`--security-opt apparmor=…` entries, so a baseline that explicitly chose
`apparmor=docker-default` (or similar) is respected — this stays scoped
to the GPU recreate path and does not override deliberate operator
choices.

## Related Issue
Resolves #3511

Related context:

- #3434 — PR that introduced the Docker-driver GPU recreate flow and the
`/proc/comm` write probe. This PR hardens the recreate to satisfy what
the probe checks across more baselines.

## Changes
- `src/lib/onboard/docker-gpu-patch.ts`: dedup `CapAdd` / `SecurityOpt`
via `Set`, always inject `SYS_PTRACE`, and inject `apparmor=unconfined`
only when no apparmor profile is pinned by the baseline.
- `src/lib/onboard/docker-gpu-patch.test.ts`: four new cases —
SYS_PTRACE always added; SYS_PTRACE deduped when baseline already has
it; apparmor=unconfined injected on empty baselines; baseline-pinned
apparmor profile preserved.

## Type of Change

- [x] Code change (feature, bug fix, or refactor)
- [ ] Code change with doc updates
- [ ] Doc only (prose changes, no code sample modifications)
- [ ] Doc only (includes code sample changes)

## Verification

- [x] `npx prek run --all-files` passes
- [x] `npm test` passes
- [x] Tests added or updated for new or changed behavior
- [x] No secrets, API keys, or credentials committed
- [ ] Docs updated for user-facing behavior changes
- [ ] `make docs` builds without warnings (doc changes only)
- [ ] Doc pages follow the [style
guide](https://github.com/NVIDIA/NemoClaw/blob/main/docs/CONTRIBUTING.md)
(doc changes only)
- [ ] New doc pages include SPDX header and frontmatter (new pages only)

Signed-off-by: Tinson Lai <tinsonl@nvidia.com>


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **Bug Fixes**
* GPU-enabled containers are now configured with the necessary kernel
security capabilities during clone/recreate operations.
* Security profile handling was fixed to preserve existing security
options and avoid unintended overrides or duplicate entries.

* **Tests**
* Added tests covering GPU container clone behavior across various
capability and security-profile scenarios to prevent regressions.

<!-- review_stack_entry_start -->

[![Review Change
Stack](https://storage.googleapis.com/coderabbit_public_assets/review-stack-in-coderabbit-ui.svg)](https://app.coderabbit.ai/change-stack/NVIDIA/NemoClaw/pull/3515)

<!-- review_stack_entry_end -->
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Tinson Lai <tinsonl@nvidia.com>
Co-authored-by: Carlos Villela <cvillela@nvidia.com>

)

## Summary

Fixes `host.openshell.internal` DNS resolution inside GPU sandboxes.
Closes #3562, #3568.

## Root cause

PR #3434 added the Docker GPU patch, which recreates the
OpenShell-managed sandbox container with GPU flags after OpenShell
creates it without `--gpu`. The recreation gated `--add-host` injection
on `networkMode !== "host"`. But `getDockerGpuPatchNetworkMode` defaults
to `"host"` (and `NEMOCLAW_DOCKER_GPU_PATCH_NETWORK` is never set), so
every GPU recreate silently dropped the OpenShell-injected
`host.openshell.internal` entry.

Symptom on DGX Spark: `LLM request failed: network connection error` —
inference can't reach the Ollama auth proxy because the hostname doesn't
resolve in the recreated container.

## The fix

`--add-host` writes to `/etc/hosts` in the mount namespace, not the
network stack — safe to preserve with `--network=host`. Removed the
conditional. `--dns` / `--dns-search` guard kept (those genuinely
conflict with host networking).

Also flipped `local.test.ts:618` to match the current validator output
(pre-existing test drift).

## Tests

- All 16/16 `docker-gpu-patch.test.ts` pass.
- Existing test 'can switch the recreated sandbox to host networking for
OpenShell callbacks' was asserting the bug; now asserts `--add-host` is
preserved.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **Bug Fixes**
* Docker GPU container networking now defaults to "preserve" unless
explicitly set to host, improving predictable network behavior.
* Host-to-container callback mapping is always preserved, ensuring
external callback endpoints remain reachable after container recreation
and improving endpoint resolution.

<!-- review_stack_entry_start -->

[![Review Change
Stack](https://storage.googleapis.com/coderabbit_public_assets/review-stack-in-coderabbit-ui.svg)](https://app.coderabbit.ai/change-stack/NVIDIA/NemoClaw/pull/3623)

<!-- review_stack_entry_end -->
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Charan Jagwani <cjagwani@nvidia.com>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Co-authored-by: Carlos Villela <cvillela@nvidia.com>