Skip to content

chore(deps): upgrade OpenShell runtime and SDK to 0.1.9 - #12603

Draft
nvidia-nemopatch-writer[bot] wants to merge 428 commits into
mainfrom
nemopatch/openshell-upgrade/13d8cdd37d346ace/migration
Draft

nvidia-nemopatch-writer[bot] wants to merge 428 commits into
mainfrom
nemopatch/openshell-upgrade/13d8cdd37d346ace/migration

Conversation

@nvidia-nemopatch-writer

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

Copy link
Copy Markdown
Contributor

Outcome

Upgrade NemoClaw’s OpenShell CLI, SDK, gateway, sandbox, and supervisor to 0.1.9. Preserve existing agents, providers, platforms, GPU behavior, credential custody, ownership, rollback, and legacy data. Full qualification remains incomplete; this PR remains a draft.

Reason

The user retargeted this upgrade to 0.1.9. The release changes destination CA support, credential placement, private-IP policy, and runtime sessions. Maintainer direction requires compatible runtime paths and preserved GPU behavior.

Related issues

Refs #12558, #13338, and accepted registered-provider selection #12333. Controller prerequisites #14642, #14700, and #14715 are integrated. Additional inherited build prerequisites remain tracked in #14751 and #14752. This branch includes #14752’s package-pin repair and authenticated paired Pi receipts while preserving the migration and Vim Snapshot changes.

Changes

  • Pin official OpenShell 0.1.9 archives, SDK, supervisor, installers, and image inputs. Verify hashes and identities. Preserve historical mappings and the frozen Hermes lifecycle v1 API’s 0.1.5 contract.
  • Preserve native provider attachment, credential scoping, TLS identity, ownership checks, legacy-data refusal, stopped-source recovery, rollback, config export, and Landlock checks.
  • Repair migration qualification for legacy endpoint flags, readiness output, bounded compressed transcripts, native tool identifiers, MCP cancellation, Discord capture, and the pinned Switchyard schema.
  • Refresh exact Debian security pins and inventory fixtures. Preserve custom zlib backports, runtime probes, matched libc packages, negative compatibility cases, and executed positive Hermes compatibility cases on both architectures.
  • Settle Hermes Tirith startup writes under the sandbox identity before gateway launch. Retain mutation refusals. Weather failure diagnostics use a bounded no-follow proof-file read and retain the original failure.
  • The latest batch routes Ollama health checks through the existing bounded no-follow token reader. A real symbolic-link regression test verifies refusal before curl starts.
  • Hermes forbidden callbacks accept HTTP 403 or an observed native connect/EACCES denial. A scoped socket observer binds the reported syscall to the caught error object. Support tests reject connection refusal, timeout, generic permission failures, and a different later error. Healthy-peer, zero forbidden requests, wrong-handle, recovery, credential, cleanup, and deadline assertions remain.
  • OpenClaw MCP lifecycle failures capture existing bounded, redacted sandbox diagnostics before destructive cleanup. Cleanup-order and success-suppression tests protect this ordering. This adds evidence for the unresolved Doctor timeout.

Verification

Commit under review: a3e3a40ea7526e03c137d37bea96cd170ecdee7d. Trusted base: 860c09d3c810ef926d49f0dd4e33ad2fab426f62.

  • Latest repair: the provider-failure probe requires exactly two failed answer attempts, first efficient, then capable, matching pinned native Switchyard fallback with per-client retries disabled. Two duplicate passthrough assertions were removed: the existing denial predicate already requires those same values. Credential, denial, recovery, cleanup, and timeout coverage remains. No budget limits changed.
  • The 36 owning Hermes support tests passed in credential-free isolated validation. The final live assertion census, formatting, lint, and diff checks passed. Normal installed commit and publication hooks ran through the qualified isolation boundary. Current source CI passed all 12 CLI shards, merged coverage, and the required checks. Current Advisor passed all nine specialists with zero findings; their artifacts, summaries, and recommendations were authenticated and read. Current image CI passed all eight producers and AMD64, ARM64, and rootless Podman activation. The trusted resolver authenticated the current image cohort. Full live qualification remains incomplete.
  • Current focused E2E run selects exactly two original standard rows: Hermes main E2E and standard installation. Its authenticated dispatch receipt binds a3e3a40e, base/workflow 860c09d, the actor, selectors, and disabled opt-ins. Both rows passed actual tests with no skips and authenticated candidate artifacts. Hermes completed all nine phases, including the exact fallback order and count, recovery, credential and egress boundaries, and seven cleanup callbacks. The standard Ubuntu installer completed installation, CLI usability, and all three cleanup actions; its log confirms OpenShell CLI and SDK 0.1.9. Hermes config export produced its expected explicit unsupported refusal; this does not establish export support or the separate Hermes security-posture row. The trusted controller accepted qualified b5ee87d20 ancestor images: all 82 declared image-input patterns are unchanged, ancestry is verified, and all three image contracts were authenticated. The latest test sources run at a3e3a40e; image CI also passed independently. This supported image reuse changes no checks, assertions, deadlines, or the full 79-row objective. The installer row checks CLI usability; it does not qualify Spark hardware. The Ollama auth-proxy job is green, and its authenticated artifact confirms authentication, OpenAI/native inference, token restart, and the Docker boundary. It does not qualify the owned daemon lifecycle: the setup shell terminated before stopping the installer service, and the test-spawned daemon failed to bind. The fixture is identical on main. Draft #14766 repairs installer-service shutdown and verifies that the test-owned daemon stays alive. Its live validation remains pending.
  • Current tunnel lifecycle passed all eleven product phases and cleanup. Authenticated artifacts prove the configured non-default ports, public HTTP 200 with OpenClaw HTML, URL removal on stop, and five passive Ready observations over 60 seconds. The tunnel PID and URL survive restart and sandbox destruction. Final tunnel stop passed; sandbox cleanup confirmed the already-destroyed sandbox was absent. These observations do not establish browser/WebSocket authentication or Ollama cleanup.
  • Current legacy gateway fixtures passed all three original variants: v0.0.89, v0.0.123, and v0.0.141. Each completed all eight phases with no skips or errors; cleanup callbacks passed (5/6/6). Authenticated artifacts verify installer refusal, credential nonexposure, authenticated turns, workspace markers, preserved forward identity, stopped-state restoration where applicable, and preserved v0.0.123 systemd identity. The trusted receipt binds the current commit and all three selected rows. These fixtures prove safe refusal and preservation for unsupported legacy cutover; they do not prove automatic data migration.
  • Current qualification covers six of the original 79 rows: Hermes main, standard Ubuntu installation, tunnel lifecycle, and the three legacy fixtures. Ollama is API-green with the inherited ownership gap above. The other 72 rows remain unqualified on this commit.
  • Parent b5ee87d20: source CI passed all 12 shards, builds, typechecks, installer integration, plugin tests, SDK packaging, audits, and real OpenClaw distribution tests. CLI/plugin line coverage was 87%/97%. Managed images passed all eight producers and AMD64, ARM64, and rootless Podman activation.
  • Parent b5ee87d20: six E2E components passed actual tests with no skips and authenticated candidate artifacts: gateway authentication, WhatsApp QR rendering, snapshot CLI, sessions and agents, sandbox operations, and concurrent gateways. All owned cleanup passed. Concurrent gateways completed all 11 lifecycle phases plus cleanup, including gateway-scoped sessions, GC, survivor restart, and both uninstall paths. These are component passes, not full-suite qualification.
  • Parent full standard E2E selected all 79 coverage rows and remains red. Its initial SDK testing-admission rejection was corrected with an exact SHA/base-bound marker. No staging Brev, Jetson, or DGX Spark opt-ins were added.
  • The subsequent three-component run is terminal: concurrent gateways passed; Hermes main failed the exact provider-attempt count addressed here; Hermes security failed earlier because the pinned Rust parser could not find protoc. Both full failed logs and the aggregate were reviewed. Seven cleanup callbacks passed in each Hermes job. The inherited compiler setup is owned by unmerged fix(e2e): install pinned protoc for standard targets #14751; its image dependency fix(images): refresh available Trixie security package pins #14752 is also unmerged. No unchanged retry is planned for either deterministic failure.
  • Existing paired Pi receipts from run 44940293729 remain authenticated with strict source parity. Selected image evidence does not establish full-suite qualification.

The diff contains no secrets, API keys, or credentials. Historical qualification evidence remains in the full-suite disposition.

Review notes

Sensitive paths include runtime and SDK pins, installer trust, credentials, workflows, and image publication. The publishing agent reviewed the diff and affected contracts. Independent acceptance remains required.

Current a3e3a40e Advisor passed all nine specialists with zero findings. All nine findings files, summaries, and E2E recommendations were authenticated and read. The six component passes above do not qualify the full standard selection. The inherited Ollama daemon-ownership gap is tracked in #14766; that repair still needs live validation.

CodeRabbit skipped this bot-authored PR because a review seat was unavailable. An organization or CodeRabbit administrator owns that blocker; no repeated request is planned without a changed prerequisite.

Remaining qualification blockers include the inherited Ollama E2E daemon-ownership gap, Hermes native GPU compatibility, Personal/custom-provider policy composition, Telegram credential placement, unmerged #14751, automatic legacy data migration, a cold anonymous model-download stall, and the MCP Doctor deadline. OpenShell 0.1.9 lacks Telegram support added by upstream #5675; #4662 remains open. These constraints remain unresolved.

Standing task authorization covers in-scope CI/E2E work. It does not establish independent review, successful qualification, or merge authority. Preserve the historical opening report. No merge or release is requested.


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

@copy-pr-bot

copy-pr-bot Bot commented Oct 3, 2026

Copy link
Copy Markdown

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

coderabbitai Bot commented Oct 3, 2026 •

Copy link
Copy Markdown
Contributor

Important

Review skipped

Bot user detected.

Ask an organization administrator to verify that this bot has an active assigned CodeRabbit review seat. See the seat-assignment guide. After the seat is confirmed, request a review again.

Run configuration

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

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

@wscurran wscurran added area: packaging Packages, images, registries, installers, or distribution area: security Security controls, permissions, secrets, or hardening integration: hermes Hermes integration behavior needs: unblock Blocked item needs dependency or decision resolved platform: brev Affects Brev hosted development environments platform: container Affects Docker, containerd, Podman, or images labels Oct 6, 2026
Base automatically changed from nemopatch/openshell-upgrade/13d8cdd37d346ace/trust to main October 7, 2026 14:57
@github-actions

github-actions Bot commented Oct 7, 2026

Copy link
Copy Markdown
Contributor

@github-code-quality

github-code-quality Bot commented Oct 7, 2026 •

Copy link
Copy Markdown
Contributor

Code coverage overview

Languages: TypeScript

TypeScript / code-coverage/plugin

main 63002cd nemopatch/openshell-... a3e3a40 +/-
96% 97% +1%
View coverage of the 10 most impacted files.
File main 63002cd nemopatch/openshell-... a3e3a40 +/-
nemoclaw/src/onboard/config.ts 98% 96% -2%
nemoclaw/src/ru...time-context.ts 100% 99% -1%
nemoclaw/src/bl...print/runner.ts 95% 95% 0%
nemoclaw/src/sh...cy-boundary.cts 100% 100% 0%
nemoclaw/src/bl...ime-identity.ts 97% 97% 0%
nemoclaw/src/bl...est-fixtures.ts 98% 99% +1%
nemoclaw/src/index.ts 94% 99% +5%
nemoclaw/src/de...search-state.ts 0% 100% +100%
nemoclaw/src/on...native-route.ts 0% 100% +100%
nemoclaw/src/search-config.ts 0% 100% +100%

TypeScript / code-coverage/cli

main 63002cd nemopatch/openshell-... a3e3a40 +/-
84% 87% +3%
View coverage of the 10 most impacted files.
File main 63002cd nemopatch/openshell-... a3e3a40 +/-
src/lib/onboard...model-router.ts 80% 0% -80%
src/lib/actions...ess-recovery.ts 65% 83% +18%
src/lib/onboard...mage-runtime.ts 52% 74% +22%
src/lib/inferen...ollama/proxy.ts 41% 64% +23%
src/lib/actions...ry/operation.ts 0% 76% +76%
src/lib/domain/...y/dimensions.ts 0% 76% +76%
src/lib/actions...try/snapshot.ts 0% 84% +84%
src/lib/inferen...er/lifecycle.ts 0% 88% +88%
src/lib/inferen...er/blueprint.ts 0% 93% +93%
src/lib/hermes-...yard-routing.ts 0% 95% +95%

ℹ️ Calculations represent line coverage. Results may change as calculations are updated.


Updated November 22, 2026 04:44 UTC

@prekshivyas

Copy link
Copy Markdown
Collaborator

@coderabbitai review

@coderabbitai

coderabbitai Bot commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor
⚠️ Action not completed

Review skipped.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.

prekshivyas added a commit that referenced this pull request Oct 9, 2026
…2847)

## Outcome

PR E2E uses the upgrade commit's reviewed dependency inputs instead of
forcing main's versions. This change consolidates #12848's Deep Agents
controller fix and extends candidate selection through protected CPU/GPU
image builds and the OpenShell host runtime.

The acceptance target is at most two PRs for upgrades to existing
supported dependencies: normally one upgrade PR, plus one
trust/bootstrap prerequisite when necessary. Ordinary version bumps must
not need a third workflow-fix PR.

## Reason

OpenShell #12603 needs candidate runtime qualification, and Deep Agents
#12376 needs its candidate base in the protected controller. Selecting
candidate artifacts only in ordinary jobs leaves protected consumers on
main's dependencies. Pi also needs to publish source before it can
attach the resulting qualification receipts.

## Changes

- Resolve a reviewed OpenShell release from fixed files at the candidate
commit. Run this resolver only for its selected gateway consumers.
Candidate source remains inert input to pin and operational-template
verification.
- Derive image prerequisites from the existing workflow dependency
graph. Gateway-only and native-producer selections do not wait for
unused image publication. Image consumers and full release qualification
retain their gates.
- Reuse the existing candidate image path for OpenClaw, Hermes, Deep
Agents, Pi, OpenShell and shared npm inputs. Published base reuse now
checks Dockerfiles, parser inputs, ignore rules and copied files for all
three managed agents.
- Build candidate OpenClaw, Hermes and Deep Agents bases on native CPU
runners. Export each as OCI content with its agent, source, workflow,
platform and run identity. Protected GPU consumers verify those bytes
and identities before offline use; they reject published-base
substitution for PR runs.
- Project only reviewed OpenShell modules and release literals into the
trusted GPU controller. Select the reviewed candidate SDK archive,
retain the trusted dependency graph, then build and verify the CLI.
Restore the seven projected source files after qualification. These
steps reuse existing pin, archive and dependency-resolution controls
because arbitrary candidate host code cannot run with protected
credentials.
- Reuse main's Pi `--publication` check for local source publication,
including later repairs in the same PR. The same PR then adds both
published receipts and matching authority. CI remains strict; partial
refreshes and source drift fail.
- Retain verified-byte receipt parsing, staging Launchable opt-in,
failure propagation, resource cleanup and full-release gates. Update the
owning E2E guidance.
- Refresh both Pi qualification receipts from the successful candidate
image publication. Correct the managed-base fixture's platform and
image-label inputs. Validate and read protected files through the same
descriptor, with symlink and hardlink rejection tests. Add fixed,
non-secret approval-selector failure categories without changing its
acceptance rules.

## Verification

Current commit: `da4ce1d219e538cab23cb519aa65e90fda55c8e7`; base:
`6a02aac7f0053978a20eaed73159386e0ad2aff9`.

- [Core CI](https://github.com/NVIDIA/NemoClaw/actions/runs/37865126626)
passed, including all twelve CLI shards, coverage, static checks and the
final gate. [Security
scanning](https://github.com/NVIDIA/NemoClaw/actions/runs/37865126621)
passed.
- [Managed-image publication and
activation](https://github.com/NVIDIA/NemoClaw/actions/runs/37865126620)
passed. Both native Pi producers and all three managed-agent producers
succeeded. Docker and Podman artifacts bind the current commit and image
digests. Each reports 28 controlled-inference turns, zero activation
builds and 13 successful cleanup callbacks.
- [Self-hosted GPU
qualification](https://github.com/NVIDIA/NemoClaw/actions/runs/37865127484)
passed on attempt 2. The initial selector timed out before managed
publication completed; one affected-job rerun after publication
succeeded passed selection and the live test. Verified evidence binds
`da4ce1d`, records full GPU offload, an authenticated real OpenClaw
agent turn, public runtime destruction and provider resources already
absent. All three cleanup callbacks passed.
- [Focused manual
qualification](https://github.com/NVIDIA/NemoClaw/actions/runs/37869966568)
passed from trusted main, including Relevant E2E. All five selected jobs
passed: cloud onboarding, full OpenClaw E2E, Hermes E2E, and
OpenClaw/Hermes security posture. Verified artifacts bind the current PR
commit, workflow and dispatch correlation; each reports one passed test
with no failures, skips or errors. All recorded cleanup callbacks
passed. These tests do not establish candidate OpenShell 0.1.2 coverage.
Jetson, DGX Spark queue and staging Launchable opt-ins were false; full
Release qualification was not selected.
- Focused local validation of the expanded repair passed 13 files / 571
distinct tests. Projection tests cover immutable commits, reviewed
templates, SDK integrity, dependency metadata, tampering, filesystem
redirection, rollback and restoration. Executable SDK fixtures exercise
installation/import without lifecycle scripts or package credentials.
- Image handoff tests cover three agents and both native platforms. They
reject wrong-agent, source, workflow, run, platform and digest evidence.
Dependency-input regressions select candidate artifacts without workflow
edits and reject fallback to main after failed candidate publication.
- OpenShell #12603 package and lock inputs at
`f473315ca9b64cab56305f3bf4a15a877f0904b2` select reviewed SDK `0.1.2`
through the pure projector. This is not live SDK/runtime qualification.
- All four Pi checker/runner conflicts are resolved; those files match
the recorded main commit. The isolated checker/runner suite passed 57
tests. Normal commit hooks, isolated pre-push publication validation and
CLI TypeScript passed. All six PR commits are GitHub Verified.
- Local broad coverage could not finish under the approved macOS
isolation. The current Linux CI coverage result above supplies the
completed broad gate; no local broad-check pass or waiver is claimed.
- The diff contains no secrets, API keys or credentials.

## Review notes

Sensitive paths include `.github/workflows/e2e.yaml`, changed
`scripts/**`, `tools/e2e/**` and
`tools/pr-review-advisor/REVIEW-QUEUE.md`. Source self-review covers
candidate admission, protected execution, prerequisite propagation,
OCI/SDK verification and cleanup.

The [actual Advisor
run](https://github.com/NVIDIA/NemoClaw/actions/runs/37866718854)
reviewed current commit `da4ce1d` against base `6a02aac7`. All nine
specialists completed with no findings. Their artifacts and published
review were read. The deterministic E2E floor remains required; clear
reviews do not waive it.

[CodeRabbit's actual
review](#12847 (comment))
covers the final repair from `2cde38759` to `da4ce1d` and reports no
actionable comments. All three recorded review threads are resolved.
Earlier parity, receipt, fixture, pathname-race and diagnostic findings
were repaired; current core CI and security scanning passed afterward.

All reported PR checks and the focused manual E2E run are green on
`da4ce1d`. GitHub reports no merge conflicts, but human review remains
required. The protected qualification boundary below remains unresolved;
no automated review grants human approval or merge authority.

### Remaining qualification boundary

Current trusted main accepts only the v1 protected activation contract;
this PR introduces the v2 controller needed by candidate builds. Running
that known-rejected protected path before adoption would not qualify it.
Live protected candidate OpenShell/SDK projection therefore still needs
the adopted trusted controller, followed by testing the existing upgrade
PR. No third workflow PR should be needed for that supported path.

The completed Docker/Podman tests do not establish OpenShell 0.1.2
coverage or every dependency combination. Both current native Pi
publications and strict CI passed; Pi source publication and receipt
attachment remain within this PR. New release trust, SDK dependency
graphs, permissions or controller protocols may require the one allowed
trust/bootstrap prerequisite.

This PR does not claim full release qualification, waive missing
evidence, resolve the separate inference epic, or authorize merge.

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

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

* **Reliability**
* E2E checks can skip managed-image publication when selected checks
don’t require those images, while still enforcing publication for
dependent checks.
* Protected runtime checks verify candidate image artifacts and use
reviewed OpenShell sources and SDK versions.
* Admin approval selection errors now produce specific diagnostics
without exposing sensitive details.
* **Compatibility**
* OpenShell gateway checks resolve the version from the selected
candidate rather than relying on a fixed version.
* **Documentation**
* Added guidance on managed-image handling, E2E prerequisites, and
protected runtime checks.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
@prekshivyas

Copy link
Copy Markdown
Collaborator

@coderabbitai full review

@coderabbitai

coderabbitai Bot commented Oct 11, 2026 •

Copy link
Copy Markdown
Contributor
⚠️ Action not completed

Review skipped.

prekshivyas added a commit that referenced this pull request Oct 12, 2026
…13041)

## Outcome

Trust the OpenShell 0.1.2 supervisor operational template that results
from main's #12928 thread-group process reader, so the base-trusted
installer-pin verifier accepts the 0.1.2 upgrade (#12603).

## Reason

#12597 trusted the 0.1.2 release manifests and template hashes, but
#12928 later changed the supervisor operational template. The 0.1.2
selection now hashes to
`31e6923609575d64ac36943e58218c801be366b782def8d33406df6e938ef5dd`,
which `scripts/checks/extract-installer-pins.mts` does not list. Because
`installer-hash-check.yaml` runs the verifier from the PR base, #12603
cannot approve this itself: `check-hash` fails, and the manual E2E
planner stops at "Resolve reviewed candidate OpenShell version".

## Changes

- Append the exact hash to the 0.1.2 `runtimeTemplateSha256` list. No
existing trust record is changed or removed.

## Verification

- Against #12603's tree (`0b34b45f6`), main's parser fails with
`supervisor runtime operational template is not base-trusted ...
actual_sha256=31e69236…`. With this change the same command succeeds and
emits the 0.1.2 `release-tsv` records.
- Added `test/install/installer-hash-check.test.ts` coverage for the
0.1.2 thread-group template: it accepts the exact template and rejects
it after the runtime bytes are altered. Without the trust line the
acceptance case fails with `actual_sha256=31e69236…`. File: 96 passed,
at its 1500-line budget.
- Commit hooks passed, including secret scan and growth guardrails;
pre-push typechecks passed.
- This does not touch #12603's other blocker (the activation E2E
inference-route observation), which is tracked separately in #12558.

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

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


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

* **Chores**
* Updated verification data for the OpenShell 0.1.2 runtime templates.
No user-facing behavior changed.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
@prekshivyas

Copy link
Copy Markdown
Collaborator

@coderabbitai full review

@coderabbitai

coderabbitai Bot commented Oct 15, 2026 •

Copy link
Copy Markdown
Contributor
⚠️ Action not completed

Review skipped.

@prekshivyas prekshivyas changed the title chore(deps): upgrade OpenShell runtime and SDK to 0.1.2 chore(deps): upgrade OpenShell runtime and SDK to 0.1.3 Oct 15, 2026
prekshivyas added a commit that referenced this pull request Oct 15, 2026
…dits (#13183)

## Outcome

Let an independently reviewed upgrade PR supply OpenShell installer
trust records and an SDK identity as data, while keeping verification
code on the trusted base. Preserve reviewed records after merge,
distribute verified SDK archives to ordinary PR CI, and build candidate
managed images on native AMD64 and ARM64 runners.

Main remains pinned to OpenShell 0.0.116. This prerequisite prepares
support for #12603; it does not activate or qualify OpenShell 0.1.3.

## Reason

Ordinary upgrades should reuse the existing qualification infrastructure
instead of requiring a new workflow PR for every version or template
change. The accepted scope uses at most the existing prerequisite and
upgrade PRs. Candidate code must not gain shared protected-host
authority.

### Related issues

Refs #12603. Inference migration remains tracked separately in #12558.

## Changes

- Add a bounded OpenShell trust registry to the existing verifier.
Preserve historical artifact identities, complete manifest sets and
template revocations. Authenticate changed candidate trust against an
exact upstream PR head/base and an independent maintainer's current
approval. Approval does not replace upstream artifact hashing.
- Connect installer-hash checks and E2E version selection to that
admission path. Refresh the complementary commit status using
trusted-base code and inert API data. Reject stale identities, edited or
revoked approvals, forks and incomplete observations.
- Extend the existing SDK producer and npm verifier to approval-bound
official SDK archives. Validate matching manifests/locks, archive bytes,
workflow provenance and unchanged npm/audit policy. Remove API
credentials before candidate dependency installation. Support GitHub's
verified merge checkout without executing the PR source as the
controller.
- Extend managed-image production to native AMD64 and ARM64, retaining
rootless Podman on AMD64. Validate platform-specific digests, catalogs
and lifecycle contracts. Add a Pi receipt collector that verifies both
architecture artifacts from one successful run and emits a patch without
applying it or dispatching work.
- Close the reviewed installer bypass: a newer installed CLI no longer
skips inspection of retained legacy state. Revoke the unsafe prospective
template fingerprint and retain real shell-process denial regressions.
- Prepare the prospective OpenShell 0.1.3 trust record and SDK identity
while retaining the active 0.0.116 producer.
- Make dependency-pin consistency checking use the existing strict
registry parser and resolver instead of scanning the old inline release
table. This lets registry-only 0.1.3 records be checked while retaining
legacy identities and rejecting malformed, duplicate, incomplete or
conflicting records. Regression tests cover these boundaries and
consumer/formula drift.
- Consolidate duplicated fixture and npm-preparation code within the
existing architecture and growth budgets. No budget limit, required gate
or live assertion is weakened.

## Verification

Latest registry-checker repair
`6ac688cfeaf7dc9b2f00fc198d9092da86d604b0`:

- Dependency-pin integration tests: 19/19 passed.
- Existing candidate trust/admission support tests: 41/41 passed.
- Repository checks: all 18 passed, including dependency pins,
architecture, assertion ratchet, Pi receipt refresh and growth checks.
- Formatting and `git diff --check` passed.
- Normal commit hooks and pre-push publication validation passed,
including CLI TypeScript. GitHub marks the new signed commit Verified.
An independent documentation writer reviewed the two-file diff and found
no documentation impact.
- Broad `npm run check` exited 1 in the initial isolated test wrapper.
Focused reproductions found its long `HOME` broke a Unix socket, its
temporary root inherited the wrong group, and its file policy denied a
test-owned `/private/tmp` directory. Corrected isolated runs passed the
pairing and snapshot files; the unrelated Google Chat proxy file passed
8/8 with an empty environment outside the sandbox. The full broad check
has not been repeated, so no local broad-pass claim is made.
- Latest-head [core CI
38703064782](https://github.com/NVIDIA/NemoClaw/actions/runs/38703064782)
and [managed images
38703065346](https://github.com/NVIDIA/NemoClaw/actions/runs/38703065346)
completed successfully. The managed-image run includes direct OpenClaw,
Hermes and Deep Agents Code startup on AMD64 and ARM64 and real rootless
Podman activation. GitHub reports 49 successful checks, 14 skipped
checks, and no failed or pending checks for this head. These runs use
the prerequisite's active OpenShell 0.0.116.

Previous published head `d1cf7ea4203144f688b23f8176aee0fa222d60c7`:

- [Core CI
38684851255](https://github.com/NVIDIA/NemoClaw/actions/runs/38684851255)
passed.
- [Managed images
38684851036](https://github.com/NVIDIA/NemoClaw/actions/runs/38684851036)
passed, including Docker AMD64/ARM64 and rootless Podman activation.
These results use the prerequisite's active OpenShell 0.0.116, not the
0.1.3 upgrade.
- Both Pi receipts were acquired from successful [managed-images run
38678295208](https://github.com/NVIDIA/NemoClaw/actions/runs/38678295208)
on `0394cd43a`. Archive/receipt hashes, both platform identities and
common cohort were verified. Receipt tests and the refresh guard passed.
These are managed-image receipts, not full Pi lifecycle qualification.
- Earlier installer, registry, admission, SDK provenance, workflow,
platform/catalog and receipt suites passed their recorded focused
checks. Their historical results do not establish latest-head CI or full
runtime qualification.

Repository checks and the actual Git hooks ran in a credential-free
filesystem/network sandbox. A non-secret out-of-boundary canary was
denied; GitHub token and SSH-agent environment variables were absent.
Git signing and SSH transport remain outside that boundary. Diff
inspected: no secrets or credentials. No live 0.1.3 E2E result is
claimed.

## Review notes

Sensitive paths include `.github/workflows/**`, `.github/actions/**`,
`scripts/**` and `tools/e2e/**`. Self-review covers immutable artifacts,
independent admission, credential isolation, fail-closed errors,
merged-source provenance, lifecycle and cleanup. For the latest two-file
checker repair, self-review checked that the inspected registry is
parsed as data, historical identities cannot be replaced, complete
manifests remain required, and formula/consumer comparisons remain
enforced. This is consistency checking, not independent authorization of
candidate trust.

Actual CodeRabbit review of `d1cf7ea42` completed without actionable
findings. No CodeRabbit review is posted for `6ac688c`. All nine
specialist findings artifacts in the [latest-head Advisor run
38705459213](https://github.com/NVIDIA/NemoClaw/actions/runs/38705459213)
bind `6ac688c` and contain no findings; its blocker gate passed. A human
review of the latest head is still required.

The human [changes-requested
review](#13183 (review))
identified the installer early-return bypass, repaired in `1edc1af4e`
with retained-state regressions. That human review has not been
dismissed. Earlier CodeRabbit registry-subset, documentation, SDK
merge-checkout and Git PATH findings are repaired; all five observed
review threads were resolved on `d1cf7ea42`. Independent acceptance
remains required.

Remaining qualification boundaries: protected hosts still use finite
trusted projections; full isolated candidate product/helper execution
and SDK/transitive coverage there are incomplete. Jetson's operator
protocol, complete Pi lifecycle qualification, consolidated preflight
and full live 0.1.3 qualification are not established. This PR is not
proof that every dependency upgrade can use at most two PRs.
Image-publication gates remain enforced. Inference, unrelated and
baseline exclusions are not passing E2E results.

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

---------

Signed-off-by: Prekshi Vyas <prekshiv@nvidia.com>
prekshivyas added a commit that referenced this pull request Oct 15, 2026
## Outcome

Upgrade admission accepts a same-repository PR authored by a GitHub App
bot. It still requires an approval from a different human with current
`maintain` or `admin` permission, bound to the exact head and base
commits.

## Reason

The trusted verifier on `main` rejects
[#12603](#12603) before it reads
approvals because its author is `nvidia-nemopatch-writer[bot]`. This
blocks candidate admission and check-hash even though the PR's
repository, head, and base match the expected values.

### Related issues

Refs #12603.

## Changes

- Accept the GitHub App `[bot]` suffix only when parsing the PR author's
login. Keep the existing human login and account-type checks for
approvers.
- Add regression tests for bot-authored PR admission with and without
independent approval, and for a malformed bot login.
- Clarify the bot-author and human-approval rule in the E2E README.

## Verification

- `npx vitest run --project e2e-support
test/e2e/support/candidate-upgrade-admission.test.ts` — 46 tests passed
after rebasing onto current `main`.
- `npm run checks:repository` — all 18 checks passed.
- `npm run docs` — 0 errors, 2 warnings.
- `git diff --check` — passed.
- Normal `pre-commit`, `commit-msg`, and `pre-push` hooks — passed. The
pre-push hook ran through a credential restricted macOS sandbox while
SSH transport stayed outside it.
- The three-file diff contains no secrets, API keys, or credentials.

## Review notes

`tools/e2e/candidate-upgrade-admission.mts` is a sensitive enforcement
path. An independent documentation writer reviewed the code, tests, and
README for writing and documentation accuracy: PASS, no findings. The
trusted verifier change still awaits independent maintainer review. This
draft does not claim approval or successful live E2E.

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


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

## Summary by CodeRabbit

* **New Features**
* Candidate upgrade pull requests authored by GitHub App bots are
accepted when an independent human maintainer approves them.
* **Bug Fixes**
* Bot-authored pull requests are rejected if their author login is
malformed or the required approval is removed.

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve 50904c2 150284c

@prekshivyas

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve 4b25b20 c474038

@github-actions

Copy link
Copy Markdown
Contributor

PR Review Advisor finished for commit 4b25b20. 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

Wait for the pinned background installer during non-root gateway preparation.
Preserve explicit commands, unsafe marker refusals and concurrent config mutation checks.

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve 82d0e70 c474038

@github-actions

Copy link
Copy Markdown
Contributor

PR Review Advisor finished for commit 82d0e70. 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

Wait for the pinned background installer under the sandbox identity during root gateway preparation.
Preserve explicit commands, unsafe marker refusals and the existing startup deadline.

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve b60a89c c474038

@github-actions

Copy link
Copy Markdown
Contributor

PR Review Advisor finished for commit b60a89c. 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

Copy link
Copy Markdown
Collaborator

Reviewed all nine specialist findings, summaries and E2E recommendations from Advisor run 44890316262 for b60a89c. Eight reports are clear. The Hermes Tirith startup findings are cleared.

The remaining P1, F-operability-recovery-8c0504dcda6776e9a5c4, repeats the inherited tunnel-stop Ollama cleanup concern previously recorded here. Fresh comparison against accepted main c474038df67fcb22182298ac9a696c81e63f0ee2 confirms that the complete command, stop implementation, both test owners and Ollama producer are byte-identical. Both revisions preserve the sandbox gateway and retain the existing Ollama cleanup default. The command reference explicitly permits that cleanup. The producer releases resident models through /api/generate with keep_alive: 0; the inspected source alone does not prove that the next inference must fail.

Disposition: inherited accepted-main behavior, with no candidate change under the repository follow-up contract. This records the evidence; it does not make the hosted Advisor gate green, waive the finding or establish independent approval.

All E2E recommendations map to tunnel-lifecycle, already included in the prepared full 79-row standard suite. Current source CI and portable E2E are green. Managed-image publication and activation must complete before full-suite dispatch. No full-suite qualification or merge is claimed.

@prekshivyas

prekshivyas commented Nov 21, 2026 •

Copy link
Copy Markdown
Collaborator

Full standard E2E qualification remains red for b60a89c40e5d0b8a6f3ba128437b4c67b4120916, using trusted main c474038df67fcb22182298ac9a696c81e63f0ee2.

Run 44895808012, attempt 1, selected all 79 standard coverage rows. It finished with 67 successful jobs, 14 execution failures, one failed aggregate, and 20 skipped jobs. These are job counts, not test counts; skips provide no passing evidence. The failed-job logs have been read completely.

Two inherited prerequisites have separate fixes:

  • fix(images): refresh available Trixie security package pins #14752 now publishes signed, GitHub-verified d5866854cbe7a3b19c6f1130a469709b95fe821f. The previous managed-image run 44923815295 on 3acbec2 passed all eight agent/platform producer jobs and Docker AMD64, Docker ARM64, and rootless Podman activation jobs. Authenticated activation artifacts were inspected; each lane reports 13 phases and successful registered cleanup. Podman reports rootless execution and Docker absent before and after. Portable rootless E2E 44923815388 also passed its selected test and 13 phases. These results remain bound to 3acbec2, not the new commit or this full suite. The new four-file batch publishes genuine paired Pi receipts from that successful run, their authority hashes, and positive/negative Hermes pinned-Python verifier coverage for both architectures. Strict source parity, 13 candidate-selection tests, 154 security-package integration tests, and the actual installed hooks passed in isolation. CodeRabbit marked the pinned-Python finding addressed in d586685; its fresh review remains pending. New source CI 44933628090, managed images 44933628014, and portable E2E 44933627990 are running. The prior Advisor skipped because source CI failed; no passing review for the new commit is claimed.
  • fix(e2e): install pinned protoc for standard targets #14751 installs the pinned protoc for standard E2E targets. It remains open at c310ceb8f83dcca723070342d50c0d48305394f3. Its CI and automated reviews were collected; independent approval remains pending. Its manual image qualification failed on the inherited package pins before exercising the compiler setup. Those failed jobs were not rerun unchanged.

Other full-run failures remain unresolved: Personal-policy/custom-provider refusals; native GPU startup compatibility; the weather agent's command deadline; Hermes's callback returning no observed HTTP 403; Telegram credential placement; channel restart readiness; and the MCP test's 45-minute deadline. The channel artifact also records a PID-namespace mismatch and network-broker errors, so Telegram alone is not an established cause of its readiness failure. MCP artifacts show the suite deadline killed the stopped rebuild while it waited for a maintenance receipt; they do not identify the slow internal stage.

Weather artifacts contain aggregate tool counts but no per-exec HTTP results, so the timeout's cause remains unknown. A diagnostic-only Hermes numeric-errno patch is prepared locally: 27 focused support tests passed, with one unrelated loopback-server test unselected. Its live assertion-budget check passed unchanged at 1,315 direct assertions across 76 files. This unpublished diagnostic does not classify arbitrary transport errors as policy denial or replace the HTTP 403 assertions. No new live diagnostic run is claimed.

The pinned OpenShell 0.1.9 lacks the Telegram path-subtree capability added by upstream #5675. The existing token-custody and credential-placement checks remain required.

Passing source shards, portable E2E, or image producers do not qualify the failed full suite. No assertion, deadline, isolation control, or repository limit has been weakened. No unchanged full-suite rerun, merge, or release approval is claimed.

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve 2061535 c474038

@github-actions

Copy link
Copy Markdown
Contributor

PR Review Advisor finished for commit 2061535. 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

Copy link
Copy Markdown
Collaborator

Collected all nine specialist summaries, findings and E2E recommendations from Advisor run 44938903386 for 2061535. Eight reports are clear.

The remaining P1, F-migration-completion-a8cabc7581512c4c0882, repeats the tunnel-stop/Ollama cleanup concern already recorded here. Fresh comparison against accepted main c474038df67fcb22182298ac9a696c81e63f0ee2 confirms that the complete command, stop implementation, both test owners and Ollama producer are byte-identical. Both revisions preserve the sandbox gateway and retain resident-model cleanup. The command reference explicitly allows that cleanup; the producer uses /api/generate with keep_alive: 0, which does not itself prove that subsequent inference fails.

Disposition: inherited accepted-main behavior; no unrelated tunnel change in this upgrade repair. The hosted Advisor gate remains failed, and this comment supplies no waiver or independent approval. The original PR's CodeRabbit bot-author seat is still unavailable; no repeated review request was sent.

Source CI, all 12 CLI shards, coverage and the Podman CPU proof passed. Nine managed-image execution jobs and the portable build failed before runtime on unavailable Debian package pins. Their logs are reviewed; an eleven-file pin/inventory/base-reference repair passed 165 owning package/native-builder tests and six Hermes resolver tests in isolation. Genuine repaired-source image production and fresh Pi qualification receipts remain required.

The tunnel-lifecycle recommendation remains in the full-suite selection. Historical full-suite failures remain open; no new full-E2E qualification, approval or merge is claimed.

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve 889c29e c474038

Record both architecture receipts from the successful current image run.
Preserve strict source parity and candidate selection authority checks.

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve 7580b95 c474038

Integrate canonical main for the trusted SDK admission base identity.
Preserve the OpenShell 0.1.9 migration and current Pi qualification receipts.

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve 6e1d747 860c09d

@github-actions

Copy link
Copy Markdown
Contributor

PR Review Advisor finished for commit 6e1d747. 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

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

Copy link
Copy Markdown
Contributor

PR Review Advisor finished for commit b5ee87d. 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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve b5ee87d 860c09d

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

Copy link
Copy Markdown
Collaborator

NemoClaw-Upgrade-E2E: approve a3e3a40 860c09d

@github-actions

Copy link
Copy Markdown
Contributor

PR Review Advisor finished for commit a3e3a40. 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

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area: ci CI workflows, checks, release automation, or GitHub Actions area: docs Documentation, examples, guides, or docs build area: e2e End-to-end tests, nightly failures, or validation infrastructure area: install Install, setup, prerequisites, or uninstall flow area: observability Logging, metrics, tracing, diagnostics, or debug output area: onboarding Onboarding FSM, provider setup, sandbox launch, or first-run flow area: packaging Packages, images, registries, installers, or distribution area: providers Inference provider integrations and provider behavior area: sandbox OpenShell sandbox lifecycle, runtime, config, or recovery area: security Security controls, permissions, secrets, or hardening chore Build, CI, dependency, or tooling maintenance integration: dcode LangChain Deep Code integration behavior integration: hermes Hermes integration behavior integration: openclaw OpenClaw integration behavior needs: unblock Blocked item needs dependency or decision resolved platform: brev Affects Brev hosted development environments platform: container Affects Docker, containerd, Podman, or images platform: macos Affects macOS, including Apple Silicon

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants