## Release range

- Previous release: `v0.0.136` at `29acfc54caed0a70d77632868ee8f4dc310c8a05`
- Candidate: `f00fb1dafe2113940b89bcd82f0561adf2696a7b`
- Candidate selection: current-main
- Scope after planning: Rebecca kept this frozen candidate after product PR [#13528](https://github.com/NVIDIA/NemoClaw/pull/13528) advanced `main` to `95d0ed24726403b717e62bc29a303ed9e8350ec4`. The planned commit remains its direct ancestor; #13528 is excluded from v0.0.137.
- Commits: 16
- Risky files detected: 66

## QA context

### Risky areas

- Workflow / enforcement
- Sandbox / policy / SSRF
- Installer / bootstrap
- Credentials / inference
- Onboarding / host glue

### Suggested test focus

- Fresh install and upgrade paths
- Onboarding wizard and sandbox creation
- Policy enforcement, network egress, and SSRF protections
- CI checks, pre-commit hooks, and DCO declarations
- Credential storage and inference provider routing

## Canonical release entry

- Path: `docs/changelog/2026-10-20.mdx` at `f00fb1dafe2113940b89bcd82f0561adf2696a7b`
- Entry:

## v0.0.137

NemoClaw v0.0.137 changes the default hosted model for new OpenClaw and Hermes NVIDIA Endpoints and OpenRouter onboarding to Nemotron 3.5 Super VL.
It also improves uninstall recovery, managed image security, policy recovery, agent input, and startup behavior.

- New NVIDIA Endpoints onboarding defaults to `nvidia/nemotron-3.5-super-vl-120b-a12b`.
  OpenRouter uses its catalog ID, `nvidia/nemotron-3.5-super-vl`.
  Explicit model selections and the Deep Agents Code Ultra default remain unchanged.
  Related change: [PR #13542](https://github.com/NVIDIA/NemoClaw/pull/13542).
- Onboarding preserves OpenRouter pricing-route IP pins when it applies the built-in pricing policy.
  If OpenShell refuses a policy write, `onboard --resume` reuses the saved preset selection for the same sandbox and confirms the live policy before clearing it.
  An unavailable saved preset stops recovery with a `--fresh` hint.
  Related change: [PR #13499](https://github.com/NVIDIA/NemoClaw/pull/13499).
- `nemoclaw <sandbox> agent` accepts piped messages without an inline `--message` and applies the input deadline to stalled pipes.
  Deep Agents Code readiness and its Auto classifier now use the selected managed Ultra route and required request parameters.
  Related change: [PR #13444](https://github.com/NVIDIA/NemoClaw/pull/13444).
- Pi restores its previous model catalog if settings synchronization fails and updates settings atomically.
  When applying a Pi configuration, managed startup still rejects symbolic links at the catalog and settings paths before the update.
  Related change: [PR #13370](https://github.com/NVIDIA/NemoClaw/pull/13370).
- New skill installations reject names longer than 64 characters with a specific validation error.
  Removal still accepts previously installed longer names.
  Related change: [PR #13488](https://github.com/NVIDIA/NemoClaw/pull/13488).
- Successful `nemoclaw uninstall` retires the selected managed gateway's native-provider authority after gateway state removal, so onboarding on the same port can recreate the provider.
  Failed cleanup, `--keep-openshell`, and concurrent registry changes retain authority for recovery.
  On rootless Podman, OpenClaw native-home cleanup retries as the sandbox user when root cannot remove the files.
  Related change: [PR #13504](https://github.com/NVIDIA/NemoClaw/pull/13504).
- Managed sandbox image builds include patched system libraries and updated Hermes HTTP and image libraries.
  Reviewed MCP SDK graphs use version 1.32.0, and image validation rejects stale package inventories.
  Existing sandboxes retain their prior images until rebuilt; after updating NemoClaw, run `nemoclaw upgrade-sandboxes` to inspect or rebuild them.
  A selected Podman runtime no longer triggers Docker host setup, Docker preflight checks, or Docker-only NIM cleanup.
  Related change: [PR #13513](https://github.com/NVIDIA/NemoClaw/pull/13513).
- Native local provider attachment now retains the expected OpenShell profile verified during inspection so NemoClaw can check it again after attachment.
  A changed profile, failed export, or timeout warns the gateway operator without rolling back a successful attachment.
  Related change: [PR #13564](https://github.com/NVIDIA/NemoClaw/pull/13564).

## Documentation coverage

- Latest included cumulative docs PR: [#13580](https://github.com/NVIDIA/NemoClaw/pull/13580), merged into this candidate.
- Final PR commit and merge commit: `c52aed19ab37e57ede5497a9df5225b390d5933e` and `f00fb1dafe2113940b89bcd82f0561adf2696a7b`.
- Final automated refresh coverage commit: `41f72b5ec8c0d07e46ae9ebfac9bf9efdff5b28b`, the main parent of verified automation commit `eaa2f6c4646c8eb4a17759b0a98c1139ce4ddc2a`.
- Later commits and merged PRs: first-parent history from the coverage point to the candidate contains only the #13580 documentation merge `f00fb1d`; no later product merge is included.
- Changed paths: only `docs/changelog/2026-10-20.mdx` and `docs/inference/understand-provider-validation.mdx`, both allowed documentation paths.
- Review and checks: Rebecca approved the exact final PR head; all required PR checks passed or were legitimately skipped. [Advisor 39551977030](https://github.com/NVIDIA/NemoClaw/actions/runs/39551977030) passed with nine clear specialists and no PR E2E selector. Independent documentation review and the local docs build passed on the final PR head. CodeRabbit skipped review of the bot-authored PR.
- Open managed docs PRs: none at coverage readback.
- Maintainer decision: Proceed with the candidate as shown.

## Base and managed image evidence

- Base-image candidate: `f00fb1dafe2113940b89bcd82f0561adf2696a7b`
- Evidence: [base-image-publication job 119068985403](https://github.com/NVIDIA/NemoClaw/actions/runs/39553713740/job/119068985403) completed successfully at `2026-10-21T06:12:42Z` on the exact candidate. Its owning [E2E workflow run 39553713740, attempt 1](https://github.com/NVIDIA/NemoClaw/actions/runs/39553713740) used `.github/workflows/e2e.yaml`, `main`, and `workflow_dispatch`. The exact-candidate check run, job, and workflow attempt identities agree.

## General E2E decision

- Newest identifiable full manual main run: [E2E full main 39541140084, attempt 1](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084), created and started `2026-10-21T03:54:36Z`, last updated `2026-10-21T04:46:19Z`. It was approximately 2 hours 22 minutes old at the `2026-10-21T06:16:32Z` inspection. Tested `c1dda570ebf6421467930d33d3f8ff9f82ad1327`, an ancestor rather than the planned `f00fb1d` candidate. Status `completed`, conclusion `failure`. [Release qualification](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119043607178) ran `04:45:50Z` to `04:45:54Z` and failed. [Exact staging Brev Launchable](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029300596) ran `03:56:02Z` to `04:25:20Z` and failed. Other failed jobs were protected managed-image startup on [amd64](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029781535) and [arm64](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029781642), [OpenClaw](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029782847) and [Hermes](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029783108) native NVIDIA inference, [Pi managed runtime on Linux AMD64](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029782998), [personal public fetch](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029783997), and [OpenClaw Discord pairing](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119029784659). Skipped jobs included the dependent [protected-image GPU and local inference](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119035513086) job and [Relevant E2E](https://github.com/NVIDIA/NemoClaw/actions/runs/39541140084/job/119043608809) aggregator; other intentionally skipped jobs are visible in the linked run.
- Later candidate-bound focused run: [workflow 39553713740, attempt 1](https://github.com/NVIDIA/NemoClaw/actions/runs/39553713740) on exact `f00fb1d` completed successfully. Its selected [onboarding policy-preset target](https://github.com/NVIDIA/NemoClaw/actions/runs/39553713740/job/119069471373) and `base-image-publication` passed. It is a focused PR E2E run, not a full manual release qualification.
- Automatic main E2E for exact `f00fb1d`: [run 39553308579](https://github.com/NVIDIA/NemoClaw/actions/runs/39553308579) was pending at `2026-10-21T06:16Z` and was cancelled at `2026-10-21T06:21:02Z` after #13528 advanced `main`. It did not produce a full candidate result.
- Subsequent automatic main E2E for `41f72b5`: [run 39549630528](https://github.com/NVIDIA/NemoClaw/actions/runs/39549630528) was still running at `2026-10-21T06:16Z`, with failed Jetson GPU and Hermes native NVIDIA inference jobs and three jobs still running. Its older-candidate result does not qualify the planned candidate.
- Requested additional E2E runs: none.
- Maintainer choice: Proceed with the status as shown. Rebecca requested no additional focused or full E2E run before tagging.

Exceptions: Proceeding with the failed older full manual E2E and cancelled final-candidate automatic main E2E because Nemotron Super 3 deprecation and Brev E2E fixes are coming in the next release.