chore(deps): upgrade OpenShell runtime and SDK to 0.1.3 by nvidia-nemopatch-writer[bot] · Pull Request #12603 · NVIDIA/NemoClaw

nvidia-nemopatch-writer · 2026-10-03T22:52:29Z

Outcome

Not ready for approval. Latest PR commit: ae7a0517080934dbc83cf87b1ca8f11e694861a4, including main d087dbcad90ee639016ba0d259556f024340acb5. This includes #13481’s routed gateway startup wait. Scoped repairs remain local and unpublished.

This PR migrates NemoClaw to OpenShell 0.1.3, including native inference and runtime-owned networking. Final-candidate CI and full PR E2E remain required. GPU qualification remains on hold for Docker Desktop WSL and Jetson/Tegra.

Reason

Preserve existing inference, lifecycle, credential, ownership, GPU, and restoration contracts across the runtime upgrade. Preserve legacy volumes without automatic copying.

Related issues

Refs #12558 and #13338. Merged inference dependencies are consumed through main.

Changes

Adapt workspace selection, SDK requests, provider pagination, authentication, policy validation, and forward ownership. Reject incomplete inventories and ambiguous resource identities.
Migrate hosted and local inference to native providers. Keep managed authenticated vLLM on its loopback-published endpoint and retain runtime-owned network identity.
Preserve existing registered-provider selection without taking ownership of operator providers. Record the selected provider identity, profile digest, endpoint, API, and credential environment because a provider name alone cannot prove the attachment. Transaction tests cover reuse, replacement refusal, publication failure, and rollback with conflicting credential environments.
Carry the registered selection through status, doctor, deployment verification, start, resume, rebuild, and export. Reset refuses to delete a provider still selected by a sandbox. Rebuild refreshes scoped credential handles after restoring configuration, preserving model settings and file-integrity checks.
Separate image configuration data from host-side admission modules and include the required payloads in Docker and portable contexts. Collect genuine Pi qualification receipts from successful AMD64 and ARM64 image jobs.
Adapt E2E helpers and obsolete shared-route fixtures to the native contracts. Preserve ownership, credential, state-restoration, and failure checks. Fit within repository budgets without raising limits.
Integrate main's approved GPU-controller fix from #13486. Normalize its test fixture for the candidate's existing registry-level revocations without changing trust checks.
Integrate main’s #13498 fixture repair. The copied headless check receives the repository root; all five existing rejection tests now pass locally.

Verification

Previously published validation

These results apply to the repair batch already published before the current local changes.

Targeted source and E2E-support suites passed for the repaired behaviors. The main controller integration passed all 133 tests in its two support suites.
npm run typecheck:cli, CLI build, architecture checks, and all seven growth checks passed.
Normal pre-commit and commit-message hooks passed. Normal pre-push publication validation and CLI typecheck passed.

Documentation validation passed: 125 routes, zero Fern errors, and two existing warnings.

The reviewed diff contains no secrets, API keys, or real credentials. Local tests and source review do not establish live inference or GPU qualification.

Published evidence and limits

Current CI for ae7a051 is running. Published image and live E2E workflows have not been approved to rerun while known compatibility blockers remain.
The preceding commit, 98ab1cf, passed ordinary CI.
Its full default PR E2E run failed: 57 jobs passed, 26 failed, and 19 were skipped. The failure count includes the aggregate qualification job. These results do not qualify the latest PR commit or the local repairs.
Its managed-image workflow built the images, but AMD64 activation failed during a guarded Hermes rebuild. ARM64 and rootless Podman activation passed. The repair adds bounded diagnostics and preserves the integrity guard; the writer responsible for the metadata change remains unknown.
Advisor found two valid defects on 98ab1cf: headless instructions and an inference test that did not exercise the agent consumer. Local repairs address both. The registered-provider scope finding was checked against accepted refactor(hermes): remove dashboard shadow config #12333 and [Epic] Complete pre-work for the OpenShell 0.1.2 upgrade #12558; it describes existing supported selection. Current-candidate Advisor evaluation remains pending.
CodeRabbit skipped substantive review because it detected a bot author. This is not an approval.

Unpublished repairs and remaining qualification

Local repairs address native provider creation and reuse, credential expiry and publication ordering, policy comparison, GPU container identity, strict network-denial evidence, actual inference consumers, and affected fixtures. Targeted tests and independent component reviews are recorded locally. Final combined review and publication hooks remain pending.

Full qualification also requires compatible runtime or trusted-workflow behavior:

The protected GPU workflow still runs base-owned helpers that use legacy gateway configuration, routing, and container selection. Candidate helper repairs alone do not change that trusted execution. fix(e2e): admit approved OpenShell templates in GPU controller #13486 is already merged and consumed; it is not a pending controller PR.
Personal mode requires native support for its hostless DNS and private-destination policy contract. The observed policy-update failure lacks the raw policy and command error, so its first cause remains unresolved. Source inspection also found incompatible downstream behavior; connection refusal is not accepted as policy denial.
The accepted Spark image and qualification plan still use legacy inference. No replacement image is claimed qualified.
Docker Desktop WSL and Jetson/Tegra GPU qualification remains on hold. Existing support requirements remain in scope.

No failure is excluded as inherited from main without concrete base evidence. Standing authorization covers routine E2E approval records and dispatches. No successful final-candidate full E2E run is claimed.

Review notes

Sensitive-path review context below covers the previously published repair batch on 98ab1cf, compared with main 8890083. The latest bot merge is ae7a051; additional local repairs remain unpublished and need final review.

The PR changes workflow, installer/runtime, sandbox/policy, credential/inference, and enforcement paths. These include .github/**, scripts/**, agents/**, managed-inference/**, tools/mcp-tool-discovery-runtime/**, nemoclaw-blueprint/**, nemoclaw/src/blueprint/**, nemoclaw/src/shared/**, src/lib/policy/**, src/lib/sandbox/**, src/lib/private-networks.ts, schemas/sandbox-policy.schema.json, src/lib/inference/**, src/lib/messaging/**, src/lib/onboard.ts, src/lib/onboard/**, and tools/e2e/**.

Independent reviewers inspected the repair batch's source, security boundaries, and writing. They checked provider identity, rollback, native credential projection, rebuild ordering, export evidence, and restoration guards against focused test results. The reviewed file hashes match the committed files; the controller integration has a separate review. No blocking finding remains in those scoped reviews. This context does not replace publication gates, full PR review, required CI, or maintainer acceptance.

Before approval:

Qualify the repaired inference routes using final-candidate images and full default PR E2E. Exclude an inherited failure only with concrete base evidence.
Preserve Linux GPU, Docker Desktop WSL, and Jetson/Tegra requirements. The pinned OpenShell runtime lacks the required WSL/Jetson device/runtime contract. Do not remove support or externally recreate the workload to bypass that limitation.
Collect required CI and automated reviews and obtain the required independent acceptance.

Collect automated-review results for the latest commit. No approval, waiver, or merge is claimed. Separable improvements remain deferred.

Documentation Writer Review

The following receipt applies to the previously published 98ab1cf changes. It does not certify the unpublished repair batch or the new bot merge.

Independent documentation writers reviewed the changed sources and documentation, including user-visible errors, recovery instructions, and test meaning. Findings were resolved and final source hashes matched the review receipts. The main controller integration and the #13498 test helper repair received separate independent reviews.

Documentation writer subagent reviewed the completed changes
Result: docs-updated
Evidence: docs/inference/switch-providers.mdx, docs/reference/commands.mdx, and test/e2e/README.md. Independent review checked source behavior, terminology, recovery instructions, code examples, and generated OpenClaw/Hermes variants.
Agent: Codex Desktop

Signed-off-by: Prekshi Vyas prekshiv@nvidia.com

Patch-Walker-Manifest: sha256:77da0e97a8176f6fde7a669021d9becdafe2ae4deaed1546dfcf152873603b67 Patch-Walker-Action: sha256:7f075ffda8178283fad4ff2b0f7100a86c0b6985c8f59e09fbb6900ab258369c Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

copy-pr-bot · 2026-10-03T22:52:33Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

coderabbitai · 2026-10-03T22:52:38Z

Important

Review skipped

Bot user detected.

To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository: NVIDIA/NemoClaw/.coderabbit.yaml
Review profile: CHILL
Plan: Advanced
Run ID: cb20fce8-b0c7-4fd8-b6a9-b02083d8856b

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

🔍 Trigger review

Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

_{Comment @coderabbitai help to get the list of available commands.}

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

…eipts Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

…e details Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

prekshivyas · 2026-10-19T16:34:56Z

NemoClaw-Upgrade-E2E: approve 8014b34 b01bb7f

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

prekshivyas · 2026-10-19T18:22:02Z

NemoClaw-Upgrade-E2E: approve f0e06d3 b01bb7f

+  return spawnSync(command[0], command.slice(1), {
+    encoding: "utf8",
+    env: { ...process.env, PATH: `${directory}:${process.env.PATH}`, ...env },
+  });


Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

prekshivyas · 2026-10-19T19:11:25Z

NemoClaw-Upgrade-E2E: approve 33282d7 af1b779

prekshivyas · 2026-10-19T19:41:40Z

NemoClaw-Upgrade-E2E: approve d4f2aaa ae0dcf2

## Outcome The protected GPU controller can project independently approved OpenShell installer and runtime templates from the candidate commit. Candidates covered by main's existing trust records retain their existing path. ## Reason The controller rejected the approved OpenShell upgrade in #12603 before GPU qualification started. It validated candidate templates only against main's registry, although the installer check already supported commit-bound candidate admission. Historical failure: [protected GPU job in run 39206820005](https://github.com/NVIDIA/NemoClaw/actions/runs/39206820005/job/117922700270). ### Related issues Refs #12603. ## Changes - Reuse the existing independent maintainer approval check for new candidate trust during prepare, verify, and restore. - Compare the admitted registry and installer consumers with immutable candidate Git blobs before changing the trusted checkout. - Pass the admitted releases to both installer and feature-gate validation; retain artifact identity, revocation, SDK archive, and projection restrictions. - Give only the four admission steps a read-only GitHub token. Verify approval before installing the reviewed SDK or building the controller, then recheck before live qualification. - Preserve rejection without mutation, partial-write rollback, and ownership checks. Document restoration failure when approval is revoked. ## Verification - `node node_modules/vitest/vitest.mjs run --project e2e-support` with the protected workspace, projection, workflow, candidate trust, and upgrade admission test files — 241 tests passed. - Workspace regressions rerun after fixture restructuring — 22 tests passed. - Revision `95cad4745bec8bf5d0d1b0d6a69d37d77858aaaa`: the focused protected-runtime workflow guard suite passed 111 tests, including rejection of SDK installation before approval verification. - `npm run docs` — 125 routes checked; Fern reported zero errors and two warnings. - Normal pre-commit, commit-msg, and pre-push hooks — passed, including publication validation and CLI/plugin/JS-config TypeScript checks. - Exact-head manual PR E2E passed: `jobs=cloud-onboard,full-e2e,security-posture` ([run 39302613910](https://github.com/NVIDIA/NemoClaw/actions/runs/39302613910)) and `jobs=managed-image-protected-runtime` ([run 39305366189](https://github.com/NVIDIA/NemoClaw/actions/runs/39305366189)). See the [E2E evidence comment](#13486 (comment)) for selectors, tested commit, and receipts. - No live E2E assertions or budgets changed. No secrets, API keys, or credentials added. ## Review notes This repairs candidate admission only. The full protected GPU suite passed on this PR head with the candidate OpenShell `0.0.116` inputs recorded in its artifact. That result does not qualify the separate OpenShell 0.1.3 upgrade in #12603. Its trusted harness still needs the `inference set` and Docker bridge migrations before that upgrade can be qualified. The workflow change must land on main before #12603's trusted controller can use it. This PR does not authorize merging #12603 or claim 0.1.3 GPU platform qualification. Independent code, security, and documentation review covered commit `18b4a45c22aa066f6ad6d7e1a44499f14200614b`. Exact-head CI and CodeRabbit review completed on `95cad4745bec8bf5d0d1b0d6a69d37d77858aaaa`; the self-hosted llama.cpp GPU check passed on its second attempt. The Advisor reported three P1 findings, all rejected in the [maintainer disposition](#13486 (comment)); its blocker gate remains red. Independent human review and maintainer acceptance remain required. If approval is revoked or GitHub admission becomes unavailable after preparation, verification and restoration fail without changing projected files. The isolated checkout remains for diagnosis; stale approval is not reused. --- Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com> Signed-off-by: Rebecca Sliter <571084+rsliter@users.noreply.github.com>  ## Summary by CodeRabbit * **Security** * Protected E2E runtime operations now check candidate approval before projecting OpenShell sources when the candidate is not already trusted. Approval checks apply to preparation, verification, and restoration. * OpenShell verification now runs before SDK download and remains in place after installation. The workflow has read-only access to pull request details for approval checks. * **Documentation** * Clarified approval requirements, token scope, verification order, failure outcomes, and recovery guidance. Approval does not permit new SDK dependency graphs on protected E2E hosts, qualify Pi, or pass candidate catalogs to the Jetson operator.  --------- Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com> Signed-off-by: Rebecca Sliter <571084+rsliter@users.noreply.github.com> Co-authored-by: Rebecca Sliter <571084+rsliter@users.noreply.github.com>

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

prekshivyas · 2026-10-19T20:50:45Z

NemoClaw-Upgrade-E2E: approve c260dc0 e7edcc4

prekshivyas · 2026-10-19T21:34:30Z

NemoClaw-Upgrade-E2E: approve 98ab1cf 8890083

github-actions · 2026-10-19T22:01:59Z

PR Review Advisor finished for commit 98ab1cf. Include the Advisor findings in the complete PR feedback collection. Verify and group valid findings before repair.

Request review only when Require no Advisor blockers is green.

All previous runs

prekshivyas · 2026-10-19T23:16:51Z

NemoClaw-Upgrade-E2E: approve ae7a051 d087dbc

prekshivyas added 3 commits October 1, 2026 10:41

chore(deps): trust openshell 0.1.2 manifests

440f288

Patch-Walker-Manifest: sha256:77da0e97a8176f6fde7a669021d9becdafe2ae4deaed1546dfcf152873603b67 Patch-Walker-Action: sha256:7f075ffda8178283fad4ff2b0f7100a86c0b6985c8f59e09fbb6900ab258369c Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

chore(deps): trust OpenShell 0.1.2 SDK and runtime templates

7c6d1b3

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

chore(deps): upgrade OpenShell runtime and SDK to 0.1.2

a8e02ec

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

prekshivyas added 11 commits October 4, 2026 23:11

chore(deps): sync OpenShell trust prerequisite with main

34f97f8

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

chore(deps): sync OpenShell migration with trust prerequisite

c2dab1b

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): align OpenShell 0.1.2 component and credential contracts

55ebb60

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): refresh OpenShell trust receipts and installer template

9f63a89

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): align OpenShell 0.1.2 gateway and receipt contracts

0ae3b1d

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): synchronize OpenShell trust with current main

7183e3f

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): omit retired OpenShell gateway zero TTL

580478d

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): refresh OpenShell trust Pi receipts

b9a8928

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): align OpenShell guest TLS ownership and Pi receipts

b7d69b2

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): cover OpenShell 0.1.2 trust and consolidate baselines

f83676e

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): carry OpenShell trust review fixes into upgrade stack

cf41f01

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

prekshivyas added 8 commits October 6, 2026 21:35

fix(deps): merge main into OpenShell trust prerequisite

98174c0

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): merge refreshed trust and fix OpenShell inventory parsing

4620ff7

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): align SDK archive and refresh trust Pi receipts

6375e46

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): retain OpenShell upgrade with corrected archive and Pi rec…

86de56e

…eipts Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): make OpenShell cleanup fixture executable

b227971

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): detach persistent Podman fixtures and retain Hermes failur…

91e4bb4

…e details Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): expose extractor subprocess execution to test classification

1b2b320

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(deps): carry extractor test correction through the OpenShell stack

05631cb

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

prekshivyas added 12 commits October 19, 2026 09:56

fix(inference): align managed GPU networking with OpenShell

838a7e9

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

test(e2e): use native inference in Station smoke

b108520

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(inference): migrate OpenClaw compatible routes to native providers

f25bff5

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(inference): migrate Model Router to native providers

d8d2e02

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(inference): preserve compatible local API and endpoint ownership

9732bae

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(inference): bind Local NIM to its native provider lifecycle

18bf3f4

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

docs(inference): align platform matrix with native NIM routing

f4d5c2a

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

test(e2e): verify native local routes and GPU workload identity

49d7636

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

test(e2e): rotate the native DCode provider credential

24c2910

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(inference): bind portable Ollama to native provider transactions

ef8567c

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

fix(inference): preserve retained Hermes endpoint ownership

462bfee

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

test(e2e): verify remaining native inference routes

f0e06d3

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

github-advanced-security Bot found potential problems Oct 19, 2026

View reviewed changes

Comment thread test/e2e/support/registered-inference.test.ts

Comment on lines +61 to +64

return spawnSync(command[0], command.slice(1), {

encoding: "utf8",

env: { ...process.env, PATH: `${directory}:${process.env.PATH}`, ...env },

});

prekshivyas added 2 commits October 19, 2026 12:02

fix(openshell): repair native image and inference validation

4af4ef8

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

merge(main): retain Station resume support in OpenShell upgrade

33282d7

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

merge: resolve conflicts with main

d4f2aaa

test(openshell): align native inference validation fixtures

e4e96ba

prekshivyas added 2 commits October 19, 2026 13:39

fix(inference): preserve registered providers across native lifecycle

d6fbe70

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

merge(main): integrate approved GPU controller

c260dc0

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>

merge(main): integrate headless fixture repair

98ab1cf

merge: resolve conflicts with main

ae7a051

Navigation Menu

Repository navigation

chore(deps): upgrade OpenShell runtime and SDK to 0.1.3 - #12603

chore(deps): upgrade OpenShell runtime and SDK to 0.1.3#12603
nvidia-nemopatch-writer[bot] wants to merge 186 commits into
mainfrom
nemopatch/openshell-upgrade/13d8cdd37d346ace/migration

nvidia-nemopatch-writer Bot commented Oct 3, 2026 •

edited by prekshivyas

Loading

Uh oh!

copy-pr-bot Bot commented Oct 3, 2026

Uh oh!

coderabbitai Bot commented Oct 3, 2026 •

edited

Loading

Review skipped

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

github-actions Bot commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

nvidia-nemopatch-writer Bot commented Oct 3, 2026 • edited by prekshivyas Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Outcome

Reason

Related issues

Changes

Verification

Previously published validation

Published evidence and limits

Unpublished repairs and remaining qualification

Review notes

Documentation Writer Review

Uh oh!

copy-pr-bot Bot commented Oct 3, 2026

Uh oh!

coderabbitai Bot commented Oct 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Review skipped

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

github-actions Bot commented Oct 19, 2026

Uh oh!

prekshivyas commented Oct 19, 2026

Uh oh!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

nvidia-nemopatch-writer Bot commented Oct 3, 2026 •

edited by prekshivyas

Loading

coderabbitai Bot commented Oct 3, 2026 •

edited

Loading