fix(status): probe inference.local route for cloud providers in status/doctor#6203
Closed
harjothkhara wants to merge 128 commits into
Closed
fix(status): probe inference.local route for cloud providers in status/doctor#6203harjothkhara wants to merge 128 commits into
harjothkhara wants to merge 128 commits into
Conversation
Contributor
|
Important Review skippedToo many files! This PR contains 1383 files, which is 1083 over the limit of 300. To get a review, narrow the scope: ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: ⛔ Files ignored due to path filters (4)
📒 Files selected for processing (1383)
You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 WalkthroughWalkthroughThe inference gateway-chain health probe, previously restricted to ChangesGateway Chain Probing Expansion
Estimated code review effort: 2 (Simple) | ~15 minutes Possibly related issues
Poem
✨ Finishing Touches🧪 Generate unit tests (beta)
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
🧹 Nitpick comments (1)
src/lib/actions/sandbox/doctor.ts (1)
743-763: 🚀 Performance & Scalability | 🔵 Trivial | 💤 Low valueGateway-chain probe now runs regardless of sandbox reachability.
Unlike
status-snapshot.ts, which gates the equivalent probe onlookup.state === "present", this call toprobeSandboxInferenceGatewayHealth(sandboxName)runs whenevercurrentProvideris known, even if the sandbox itself was never confirmed reachable (sandboxReachableis false, e.g. dead/absent sandbox). The probe's contract returnsnullon exec failure so this won't crash, but it does add an unnecessary exec+network round trip (curl with 5s timeout) on every doctor run against a dead sandbox, and is inconsistent with the sibling implementation in the same cohort.🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@src/lib/actions/sandbox/doctor.ts` around lines 743 - 763, Gate the `probeSandboxInferenceGatewayHealth` call in `doctor.ts` the same way the sibling `status-snapshot.ts` logic does, so it only runs when the sandbox is actually reachable/present. Use the existing `sandboxReachable`/lookup-state check around the `gatewayChain` probe before appending to `inferenceHealth.subprobes`, and keep the current `probeSandboxInferenceGatewayHealth` handling unchanged otherwise.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@src/lib/actions/sandbox/doctor.ts`:
- Around line 743-763: Gate the `probeSandboxInferenceGatewayHealth` call in
`doctor.ts` the same way the sibling `status-snapshot.ts` logic does, so it only
runs when the sandbox is actually reachable/present. Use the existing
`sandboxReachable`/lookup-state check around the `gatewayChain` probe before
appending to `inferenceHealth.subprobes`, and keep the current
`probeSandboxInferenceGatewayHealth` handling unchanged otherwise.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: efa50985-5236-47fc-b18b-b1203746d225
📒 Files selected for processing (3)
src/lib/actions/sandbox/doctor-flow.test.tssrc/lib/actions/sandbox/doctor.tssrc/lib/actions/sandbox/status-snapshot.ts
…s/doctor `status` and `doctor` reported inference healthy for cloud/managed providers by probing only the upstream provider endpoint, never the `inference.local` route the agent actually uses. When `inference.local` was broken inside the sandbox they still reported healthy (exit 0), contradicting `connect`. Widen the inference.local gateway-chain probe (added for local providers in NVIDIA#3265) to run for every provider. In doctor this removes the local-only gate in collectInferenceSubprobes; in status it drops the provider clause on the snapshot probe. A broken route now surfaces as `[fail] Provider health (gateway)` and flips doctor to fail (exit 1). Refs NVIDIA#6192 Signed-off-by: harjoth <harjoth.khara@gmail.com> Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
harjothkhara
force-pushed
the
fix/6192-status-doctor-probe-inference-local
branch
from
6a45c01 to
ec8bd67
Compare
Contributor
Author
|
Thanks for the review! On the gateway-chain reachability nitpick — the probe is already gated on reachability in if (!sandboxReachable) return [...existing, skippedInferenceGatewayProbe()];
const gateway = await probeSandboxInferenceGatewayHealth(sandboxName);So an unreachable/dead sandbox short-circuits to the "skipped" subprobe with no The one real difference from the |
Contributor
Author
|
Hi @wscurran — thanks for the reviews on my earlier PRs. This one (#6203, fixes #6192) is blocked on the It's a small, tested runtime fix: |
Collaborator
|
Thanks—removing the local-provider gate is the right direction. Before merge, could we make the in-sandbox Currently, upstream healthy + Please:
|
<!-- markdownlint-disable MD041 --> ## Summary This PR restores LangChain Deep Agents Code inference by replacing the sandbox-create host proxy seed with OpenShell's managed runtime proxy before dcode runs. It keeps `inference.local` on the proxy path instead of allowing direct DNS resolution, without changing OpenShell route provisioning or other agent runtimes. ## Related Issue Fixes NVIDIA#6191. ## Changes - Pin the validated `NEMOCLAW_PROXY_HOST` / `NEMOCLAW_PROXY_PORT` build values in root-owned, read-only image files. The dcode startup and direct-exec launcher reject missing, linked, writable, or non-root-owned files and ignore process-level proxy-host overrides. - Normalize uppercase and lowercase HTTP proxy variables for every dcode entry path, set runtime bypasses to loopback plus the managed proxy host only, and persist the same normalized values for interactive and login shells. Inherited corporate proxy URLs, credentials, and `inference.local` bypass entries are not retained. - Make the dcode-only `connect --probe-only` route check use the login-shell proxy contract and fail if route repair still reports `inference.local` unhealthy. OpenClaw, Hermes, and unknown-agent probe behavior is unchanged. - Add focused coverage for inherited credential-bearing proxies, exact `NO_PROXY` values, trusted host/port overrides, hostile runtime overrides, file ownership/mode, validator parity, login-shell persistence, direct dcode execution, and broken-route connect behavior. - Require the live Deep Agents check to observe direct DNS state, verify the trusted proxy files, force the managed proxy for `https://inference.local/v1/models`, and return `PONG` with dcode exit 0 through both login-shell and direct-exec paths. Connection, DNS, timeout, provider, and ambiguous failures fail the check. - An empty `getent hosts inference.local` is intentionally valid for dcode. Direct DNS or `/etc/hosts` provisioning is outside this fix because OpenShell's managed L7 proxy owns the route; the sandbox-create host proxy seed remains unchanged for host-side chaining. - Existing Deep Agents Code sandboxes must be rebuilt after upgrading because the corrected startup and launcher scripts are baked into the image. - Exact-head [`ubuntu-repo-cloud-langchain-deepagents-code` E2E](https://github.com/NVIDIA/NemoClaw/actions/runs/28616070748) passed at `22d42e2fa26d17bb8f64d45eb9ba853c8feb9e51`: onboarding and Ready state, absent direct DNS, managed `/v1/models` HTTP 200, login/direct `PONG`, `connect --probe-only`, and clean teardown all passed. - Reporter-class macOS/Colima and DGX Spark confirmation remains pending. A local macOS/Colima attempt stopped before sandbox creation on provider HTTP 401 and cleaned up its isolated gateway state; any DGX Provisioning/GPU lifecycle failure remains a separate blocker. ## 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) ## Quality Gates <!-- Check all that apply. For any "covered by existing tests", "not applicable", or waiver entry, add a brief justification on the same line or in the Changes section. --> - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: New regression coverage was required and added. - [ ] Tests not applicable — justification: Tests apply and were added. - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: This restores the managed `inference.local` contract and rebuild workflow already documented in the Deep Agents quickstart, inference options, troubleshooting, and sandbox lifecycle pages. - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: Pending maintainer sensitive-path review; no waiver claimed. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: No acceptance claimed. The local repo-wide `test-cli` coverage hook hit unrelated baseline subprocess-loader and timeout failures; targeted suites, scoped hooks, and exact-head GitHub CI/E2E are authoritative for this change. ## Verification <!-- Check each item you ran and confirmed. Leave unchecked items you skipped. Doc-only changes do not require npm test unless you ran it. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes - [x] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Aaron Erickson <aerickson@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a Deep Agents Code launcher that normalizes managed proxy settings (host/port, `HTTP_PROXY`/`HTTPS_PROXY`, and `NO_PROXY`) before running entry points. * Introduced agent-aware inference route probing during sandbox connect. * **Bug Fixes** * Hardened proxy environment contract with stricter host/port validation and consistent `NO_PROXY` behavior across uppercase/lowercase. * Reduced risk of proxy credential leakage in outputs and generated artifacts. * **Tests** * Expanded end-to-end and unit coverage to enforce the login-shell proxy contract, probe wiring, and improved inference connection/error classification. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Aaron Erickson <aerickson@nvidia.com>
…kup (NVIDIA#6199) <!-- markdownlint-disable MD041 --> ## Summary Pre-upgrade `backup-all` aborted the `curl | bash` installer whenever a running sandbox's in-sandbox SSH endpoint did not answer, with no override, looping the upgrade forever. This classifies such a sandbox as unreachable and adds `NEMOCLAW_SKIP_UNREACHABLE_SANDBOX_BACKUP=1` to skip it so the upgrade proceeds and onboarding recovers it from its latest validated backup. ## Related Issue Fixes NVIDIA#6188 ## Changes - `src/lib/state/sandbox.ts`: add `unreachable` to `BackupResult` and set it on an SSH transport-level dir-check failure (exit 255, timeout, spawn error) via a new `isSshTransportFailure` predicate. - `src/lib/actions/maintenance.ts`: an unreachable running sandbox is skipped when `NEMOCLAW_SKIP_UNREACHABLE_SANDBOX_BACKUP=1`, otherwise it still fails but prints actionable guidance before exit. - `scripts/install.sh`: reword the pre-upgrade backup abort to name the override and the recovery path. - `docs/reference/commands.mdx`: document the flag in the `backup-all` section. - Tests: `maintenance.test.ts` gains skip-with-flag, fail-with-guidance, and flag truth-table cases; new `sandbox.test.ts` covers `isSshTransportFailure`. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes - [x] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) - [x] 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 * **New Features** * `backup-all` now detects running sandboxes with an unreachable in-sandbox SSH endpoint and can skip them when `NEMOCLAW_SKIP_UNREACHABLE_SANDBOX_BACKUP=1`. * Skipped sandboxes are recovered during onboarding from the latest validated backup; any uncommitted state since then is not preserved. * **Bug Fixes** * Improved failure handling and remediation when SSH transport/unreachability occurs, including clearer guidance and installer/upgrade retry behavior. * **Documentation** * Updated `nemoclaw` and `nemohermes` `backup-all` docs to state the default abort behavior and the skip flag’s exact `=1` requirement. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Tinson Lai <tinsonl@nvidia.com> Signed-off-by: Charan Jagwani <cjagwani@nvidia.com> Co-authored-by: Charan Jagwani <cjagwani@nvidia.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
wscurran
added
area: cli
Contributor
|
✨ Thanks for the PR. This fixes a meaningful gap where status/doctor gave false healthy signals for cloud providers by not probing the actual inference.local route. Ready for maintainer review. Related open issues:
Related open issues: |
<!-- markdownlint-disable MD041 --> ## Summary This PR makes release-prep docs a small explicit prerequisite before cutting a NemoClaw tag. Maintainers should run `/nemoclaw-contributor-update-docs for vX.Y.Z` before `release:plan`, and regenerate the plan if `origin/main` changes afterward. ## Related Issue None. ## Changes - Add the explicit `/nemoclaw-contributor-update-docs for vX.Y.Z` pre-tag invocation to the docs-update skill. - Add a short pre-tag docs note to `nemoclaw-maintainer-evening` before it loads `cut-release-tag`. - Add the same docs-before-plan precondition to `nemoclaw-maintainer-cut-release-tag` and the release-train policy reference. - Add a narrow maintainer-skill policy regression test for the docs-before-plan ordering. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [x] Doc only (includes code sample changes) ## Quality Gates <!-- Check all that apply. For any "covered by existing tests", "not applicable", or waiver entry, add a brief justification on the same line or in the Changes section. --> - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: this updates contributor and maintainer workflow guidance, not user-facing product documentation. - [ ] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each item you ran and confirmed. Leave unchecked items you skipped. Doc-only changes do not require npm test unless you ran it. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [ ] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes — latest cleanup commits were pushed with `--no-verify` per maintainer request; relying on remote CI. - [x] Targeted tests pass for changed behavior — `npm test -- test/maintainer-skills-policy.test.ts` - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) — ran `npm run docs`; it completed successfully, but Fern reported 1 hidden warning, so this is left unchecked. - [ ] 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Miyoung Choi <miyoungc@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a required pre-tag documentation step before generating final release plans. * Clarified when release-prep documentation should be run for a versioned release. * **Bug Fixes** * Prevented release plans from being created too early. * Ensured release plans are regenerated if release-related changes land afterward. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
<!-- markdownlint-disable MD041 --> ## Summary This PR carries the final proxy-review follow-ups that reached the source branch seconds after NVIDIA#6204 was squash-merged. It tightens the Deep Agents Code managed-proxy contract without changing its routing behavior. Existing Deep Agents Code sandboxes must still be rebuilt after upgrading because the corrected startup and launcher scripts are baked into the image. ## Related Issue Follow-up to NVIDIA#6204; relates to NVIDIA#6191. ## Changes - Make the credential-free runtime proxy environment read-only (`0444`) and verify its ownership and mode in focused and live checks. - Clear inherited `ALL_PROXY`/`all_proxy` state at every dcode runtime boundary, including persisted login-shell state, while preserving the sandbox-create host proxy seed used by OpenShell. - Extract the connect inference-route probe into a focused module and test, keeping existing non-dcode behavior unchanged. - Require strict exit-zero standalone `PONG`, preserve the dcode `routeHealthy: null` smoke path, and validate the real login/direct/connect paths. - Extend canonical secret detection for complete multi-segment LangSmith v2 keys and keep the shell/runtime mirrors under parity tests. - Record the agent-local `0444` runtime-file risk acceptance and its compensating controls in `SECURITY.md`. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: [Charan Jagwani approval](NVIDIA#6206 (review)). - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [ ] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes — exact affected hooks, CLI build/typecheck, focused tests, pre-push checks, and all exact-head GitHub CI passed; the broad local macOS hook exposed pre-existing Node 22.16 child-process and BSD `script` incompatibilities outside this diff. - [x] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) ### Runtime evidence - NVIDIA#6204 exact-head E2E: [ubuntu-repo-cloud-langchain-deepagents-code](https://github.com/NVIDIA/NemoClaw/actions/runs/28616070748) passed on `22d42e2f`, including onboarding/Ready, managed `/v1/models` HTTP 200, login-shell and direct-exec `PONG`, connect probe, and clean teardown. - NVIDIA#6206 final exact-head E2E: [ubuntu-repo-cloud-langchain-deepagents-code](https://github.com/NVIDIA/NemoClaw/actions/runs/28623778067) passed on `de004fd6`, including onboarding/Ready, cleared `ALL_PROXY`/`all_proxy`, trusted root-owned proxy source comparison, absent direct DNS, managed `/v1/models` HTTP 200, exit-zero standalone-line login/direct `PONG`, connect acceptance, clean credential scan, and clean teardown. --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Aaron Erickson <aerickson@nvidia.com> --------- Signed-off-by: Aaron Erickson <aerickson@nvidia.com> Signed-off-by: Charan Jagwani <cjagwani@nvidia.com> Co-authored-by: Charan Jagwani <cjagwani@nvidia.com> Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> Co-authored-by: Apurv Kumaria <akumaria@nvidia.com>
apurvvkumaria
requested changes
Jul 2, 2026
apurvvkumaria
left a comment
apurvvkumaria
left a comment
Collaborator
There was a problem hiding this comment.
Requesting changes at head ec8bd679. Removing the local-provider gate is the right direction, but the published patch still does not make the route the agent actually uses authoritative:
status-snapshot.tsretains the upstream provider result as rootinferenceHealthand appendsinference.localonly as a subprobe.statuscan therefore printInference: healthy, emit root JSONok: true, and exit0while the gateway subprobe is broken.- Both
doctorandstatusskip the route entirely when no direct provider-health probe is registered. Supported providers such asnvidia-routerandhermes-providertherefore remain untested. probeSandboxInferenceGatewayHealthtreats every positive HTTP code as healthy, including 503. The intended contract is 1xx-4xx reachable, 5xx unhealthy, and000broken.- A
null/unavailable in-sandbox probe is silently omitted, allowing an upstream result to produce a false green instead of failing closed.
Please probe inference.local independently, use it as the primary status/doctor result, retain upstream checks as labeled diagnostics, and make failed or unavailable authoritative probes produce nonzero status. Add regression coverage for status text and JSON, exit codes, providers without direct probes, HTTP 503, and an unavailable probe.
This formalizes the earlier maintainer comment; the current PR head has not yet incorporated that follow-up.
## Summary This PR resolves the maintainer-owned docs gaps behind the noisy contributor PR set without reusing that branch history. It adds missing recovery guidance and a host-side state reference while keeping OpenClaw-only gateway guidance out of the Hermes variant. ## Related Issue Closes NVIDIA#5326 — `docs/get-started/windows-preparation.mdx`: adds Cursor Run Mode / Legacy Terminal Tool troubleshooting for Windows starter-prompt installs, including fallback-file and Docker Desktop readiness guidance. Closes NVIDIA#6027 — `docs/reference/troubleshooting.mdx`: adds `Kubernetes namespace not ready` recovery steps that clean failed setup state before retrying install, with preserved-user-data notes. Closes NVIDIA#6028 — `docs/get-started/prerequisites.mdx`, `docs/reference/troubleshooting.mdx`, `ci/platform-matrix.json`, `docs/reference/platform-support.mdx`: documents that Homebrew Colima users must install both Colima and the Docker CLI and verify `docker info`. Closes NVIDIA#6030 — `AGENTS.md`: clarifies that `nemoclaw/` registers `/nemoclaw` OpenClaw TUI slash commands and that the `openclaw nemoclaw <cmd>` shell subcommand path is descoped. Closes NVIDIA#6031 — `docs/reference/troubleshooting.mdx`: adds OpenShell/OpenClaw gateway startup-order guidance for the OpenClaw variant and replaces stopped-sandbox guidance with a lighter recovery ladder before rebuild. Closes NVIDIA#6088 — `docs/reference/host-files-and-state.mdx`, `docs/manage-sandboxes/lifecycle.mdx`, `docs/index.yml`: adds a unified `~/.nemoclaw/` host files and state reference, including current `sandboxes.json` registry wording and uninstall preservation behavior. ## Changes - Add Windows Cursor recovery guidance to the Windows preparation page instead of expanding the starter prompt. - Document Homebrew Colima needing the Docker CLI, Kubernetes namespace cleanup, OpenShell/OpenClaw gateway order, and `sandbox_container_stopped` recovery. - Add a `Host Files and State` reference page, wire it into both OpenClaw and Hermes navigation, and link related docs. - Clarify the descoped `openclaw nemoclaw <cmd>` path in the agent architecture table. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [x] Doc only (includes code sample changes) ## Quality Gates - [ ] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [x] Tests not applicable — justification: docs-only change; validated with docs generators, link checks, and Fern docs build. - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [ ] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes - [ ] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) - [x] Doc pages follow the [style guide](https://github.com/NVIDIA/NemoClaw/blob/main/docs/CONTRIBUTING.md) (doc changes only) - [x] New doc pages include SPDX header and frontmatter (new pages only) Commands run: - `python3 scripts/generate-platform-docs.py --check` - `npm run docs:check-agent-variants` - `bash test/e2e/e2e-cloud-experimental/check-docs.sh --only-links --local-only docs/get-started/windows-preparation.mdx docs/resources/agent-skills.mdx docs/reference/troubleshooting.mdx docs/reference/host-files-and-state.mdx` - `git diff --check` - `npm run docs` — passed with 0 errors; Fern reported 1 warning. - `npm run build:cli` — run so the pre-push TypeScript hook could resolve `dist/` imports. --- Signed-off-by: Miyoung Choi <miyoungc@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a new reference page explaining NemoClaw host-side storage, safe-to-delete items, and how uninstall preserves or destroys data. * Added the new “Host Files and State” page to the user-guide navigation for both agent variants. * **Documentation** * Improved macOS Apple Silicon + Homebrew Colima setup by explicitly requiring the Docker CLI and verifying with `docker info`. * Expanded Windows troubleshooting guidance and added clearer recovery steps for onboarding and stopped sandboxes, including the correct startup order for gateways. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Miyoung Choi <miyoungc@nvidia.com>
<!-- markdownlint-disable MD041 --> ## Summary <!-- 1-3 sentences: what this PR does and why. --> Define exact-SHA E2E evidence as an explicit precondition of the release process. The candidate commit's `.github/workflows/e2e.yaml` remains the sole source of truth: every declared E2E test needs at least one green execution or an itemized maintainer exception before the exact tag confirmation is requested. ## Changes <!-- Bullet list of key changes. --> - Freeze the candidate SHA with the release plan, then build an evidence ledger across workflow runs, reruns, and attempts. - Require green evidence for every E2E test declared by that SHA's workflow, including explicit-only and expanded matrix executions, without maintaining a second test inventory. - Preserve maintainer discretion through itemized exceptions and invalidate both evidence and exceptions whenever the candidate SHA changes. - Carry the gate through the cut-tag, evening, daily-flow, and maintainer-cadence guidance while keeping overnight QA as additional post-tag validation. - Add a maintainer-policy contract test for the exact-SHA, flaky-run, exception, and confirmation-order invariants. ## 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) ## Quality Gates <!-- Check all that apply. For any "covered by existing tests", "not applicable", or waiver entry, add a brief justification on the same line or in the Changes section. --> - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: this changes internal maintainer policy and skills only; documentation review found no user-facing CLI, configuration, API, or runtime change. - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: maintainer-directed release-policy design received an independent diff review; both hardening findings were addressed before commit. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each item you ran and confirmed. Leave unchecked items you skipped. Doc-only changes do not require npm test unless you ran it. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes - [x] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Carlos Villela <cvillela@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Documentation** * Strengthened release workflow guidance to require pre-tag, exact commit-based E2E evidence before confirming a tag. * Added an evidence-ledger approach using the E2E workflow as the single source of truth, including detailed green evidence (counts plus run/job links and attempt numbers), explicit itemized exceptions for non-green tests, and rules to regenerate evidence if the candidate SHA changes. * **Tests** * Added automated coverage to ensure the updated evidence and confirmation gating rules are enforced across the maintainer release workflows. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Carlos Villela <cvillela@nvidia.com>
<!-- markdownlint-disable MD041 --> ## Summary This PR prepares the user-facing documentation for v0.0.73 before the release plan is frozen. It adds release notes for the merged runtime changes and closes documentation gaps around DNS-backed HTTPS endpoint validation and LangChain Deep Agents Code proxy recovery. ## Changes - Add the `v0.0.73` release-note section with links to the detailed command, inference, recovery, lifecycle, platform, and setup documentation. - Correct the custom endpoint guidance so DNS-backed HTTPS rejection and the supported alternatives match the fail-closed runtime behavior. - Document the managed `inference.local` proxy boundary and rebuild requirement for existing LangChain Deep Agents Code sandboxes. - Add troubleshooting guidance for the DNS-backed HTTPS validation error. - Validate with `npm run docs:sync-agent-variants` and `npm run docs`; Fern completed with 0 errors and 2 existing warnings. - Source summary: - [NVIDIA#6139](NVIDIA#6139) -> `docs/about/release-notes.mdx`, `docs/inference/inference-options.mdx`, `docs/reference/commands.mdx`, `docs/reference/commands-nemohermes.mdx`, and `docs/reference/troubleshooting.mdx`: Document fail-closed DNS-backed HTTPS endpoint handling and recovery options. - [NVIDIA#6142](NVIDIA#6142) -> `docs/about/release-notes.mdx`: Summarize native OpenShell GPU injection and compatibility-path diagnostics. - [NVIDIA#6197](NVIDIA#6197) -> `docs/about/release-notes.mdx`: Summarize agent-aware messaging preset rejection. - [NVIDIA#6199](NVIDIA#6199) -> `docs/about/release-notes.mdx`: Summarize the unreachable-sandbox backup opt-in, restore behavior, and data-loss boundary. - [NVIDIA#6204](NVIDIA#6204) and [NVIDIA#6206](NVIDIA#6206) -> `docs/about/release-notes.mdx` and `docs/get-started/quickstart-langchain-deepagents-code.mdx`: Document the corrected managed proxy contract and required sandbox rebuild. - [NVIDIA#6213](NVIDIA#6213) -> `docs/about/release-notes.mdx`: Summarize the merged setup, recovery, and host-state documentation. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [x] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates <!-- Check all that apply. For any "covered by existing tests", "not applicable", or waiver entry, add a brief justification on the same line or in the Changes section. --> - [ ] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [x] Tests not applicable — justification: documentation-only release preparation; the Fern docs build validates the changed pages and routes. - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [ ] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each item you ran and confirmed. Leave unchecked items you skipped. Doc-only changes do not require npm test unless you ran it. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes - [ ] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) - [x] 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Carlos Villela <cvillela@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Documentation** * Added a new **v0.0.73** release notes section with six highlights at the top of the changelog. * Expanded **Custom Endpoint URL Validation** guidance in inference option docs, including explicit acceptance/rejection rules for HTTP vs DNS-backed HTTPS and how validated IPs are stored. * Updated command references (`nemohermes inference set`, `$$nemoclaw inference set`) to match the new validation behavior. * Added troubleshooting documentation for unsupported **DNS-backed HTTPS endpoints**, plus clarified Deep Agents Code routing and post-upgrade sandbox rebuild guidance. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Carlos Villela <cvillela@nvidia.com>
<!-- markdownlint-disable MD041 --> ## Summary Adds an agent-first install path to the NemoClaw docs front door and quickstarts so users can copy a prompt into Cursor, Claude Code, Codex, Copilot, or another local coding agent. The prompt now lives in one shared docs source, powers the copy button, and renders as a manual fallback for users whose browser or agent UI cannot use clipboard copy. ## Related Issue Fixes NVIDIA#5048 ## Changes - Added `docs/_components/StarterPrompt.tsx` as the shared source for the starter prompt plus a manual-copy fallback. - Updated `docs/_components/StarterPromptButton.tsx` to copy the shared prompt instead of embedding a duplicate prompt string. - Added the agent-supported install path near the top of the home page, OpenClaw quickstart, Hermes quickstart, and AI Agent Docs page. - Updated the prompt to tell agents to use NemoClaw skills when available and bootstrap `nemoclaw-user-guide` when missing. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [ ] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [x] Doc only (includes code sample changes) ## Quality Gates <!-- Check all that apply. For any "covered by existing tests", "not applicable", or waiver entry, add a brief justification on the same line or in the Changes section. --> - [ ] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [x] Tests not applicable — justification: docs-only CTA, prompt source, and docs-site component changes; no CLI/runtime behavior changed. - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [ ] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each item you ran and confirmed. Leave unchecked items you skipped. Doc-only changes do not require npm test unless you ran it. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes - [ ] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) - [x] 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) `npm run docs` passed with 0 errors; Fern reported the pre-existing light-mode accent color contrast warning. `fern check --warnings` confirmed the warning is unrelated to these docs content changes. --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Miyoung Choi <miyoungc@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a “Start from Your Coding Agent” flow across the getting-started docs and homepage onboarding. * Provided a copyable starter prompt experience with an in-UI manual copy fallback for clipboard-restricted environments. * **Documentation** * Updated quickstart and related guide pages with expanded, agent-friendly prompt instructions and behavior constraints (one question at a time, command approval, no secrets). * **Tests** * Added coverage to confirm prompt contents and validate the copy/manual fallback rendering across docs pages. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
…n secret guard (NVIDIA#6202) ## Summary Adds a `dcode status` identity command to the managed LangChain Deep Agents Code wrapper so a user connected into a sandbox can tell which sandbox the session is in, and fixes the wrapper secret guard falsely refusing to start when OpenShell injects its canonical TLS client-key path after a credential provider is attached. ## Related Issue Fixes NVIDIA#6189 Resolves NVIDIA#6186 ## Changes - `agents/langchain-deepagents-code/dcode-wrapper.sh`: add a `status` / `whoami` / `identity` subcommand that distinguishes the sandbox, NemoClaw harness, active dcode agent, inference route, upstream provider, model, endpoint, and runtime, then exits without launching Deep Agents Code; advertise the managed aliases in `dcode --help`. - `agents/langchain-deepagents-code/dcode-wrapper.sh`: allow only the exact runtime pair `OPENSHELL_TLS_KEY=/etc/openshell/tls/client/tls.key`; alternate paths, opaque values, PEM material, and provider tokens remain rejected. - Keep `.deepagents/.env` fail-closed: the OpenShell runtime exception does not apply to the user-mutable env file. - Add `tvly-` Tavily tokens to the secret-shape detection patterns. - `src/lib/onboard/sandbox-create-launch.ts`, `src/lib/onboard.ts`: forward `NEMOCLAW_SANDBOX_NAME` into the Deep Agents Code sandbox create env (gated to the `langchain-deepagents-code` agent). - `agents/langchain-deepagents-code/start.sh`: persist `NEMOCLAW_SANDBOX_NAME` into the shared runtime env file that connect shells and the wrapper source, so `dcode status` resolves the name. - `docs/get-started/quickstart-langchain-deepagents-code.mdx`: document `dcode status`. - Tests: wrapper identity and agent-preference resolution, managed help, exact runtime-pair acceptance, negative secret/path cases, mutable-env rejection, sandbox-create env injection, and start.sh serialization. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates <!-- Check all that apply. For any "covered by existing tests", "not applicable", or waiver entry, add a brief justification on the same line or in the Changes section. --> - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: reviewed against the repository security checklist; the exception is limited to one exact runtime name/value pair, values are never logged, mutable env files remain fail-closed, and Linux tests cover alternate paths, PEM material, opaque values, and provider tokens. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each item you ran and confirmed. Leave unchecked items you skipped. Doc-only changes do not require npm test unless you ran it. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [ ] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes — pre-push hooks and scoped checks pass; the broad pre-commit test hook encounters unrelated environment-sensitive baseline failures. - [x] Targeted tests pass for changed behavior - [x] Required live Deep Agents Code E2E passes on the PR branch - [ ] Full `npm test` passes (broad runtime changes only) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) — build passes with two pre-existing warnings. - [x] 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> Signed-off-by: Apurv Kumaria <akumaria@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added `dcode status`, `dcode whoami`, and `dcode identity` to show the active sandbox/session identity and exit without launching the interactive UI. * Sandboxes for supported Deep Agents Code runs can now propagate a sandbox name into the runtime environment. * **Bug Fixes** * Strengthened runtime secret/credential checks by allowing only OpenShell’s exact mounted TLS-key keypair, and correctly rejecting additional secret-shaped Tavily token formats. * **Documentation** * Updated the Quickstart “Use the Harness” guide with the new identity/status workflow and command aliases. * **Tests** * Added and expanded coverage for identity/status output, secret gating, TLS-key allowlisting behavior, and sandbox name propagation. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Co-authored-by: Apurv Kumaria <akumaria@nvidia.com> Co-authored-by: Prekshi Vyas <prekshiv@nvidia.com> --------- Signed-off-by: Tinson Lai <tinsonl@nvidia.com> Signed-off-by: Apurv Kumaria <akumaria@nvidia.com> Co-authored-by: Prekshi Vyas <prekshiv@nvidia.com> Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com> Co-authored-by: Apurv Kumaria <akumaria@nvidia.com>
## Summary Reverts NVIDIA#6202 from `main` so v0.0.73 retains the release boundary that was already documented and exercised by the release E2E run. PR NVIDIA#6202 remains targeted for v0.0.74. The resulting tree is byte-for-byte identical to commit `2276b2e1373548f9afa95b3a4f4bcd8db244874c` (`967ad0207b591bfc7f3398f37c230f199bb932ed`). ## Related Issue Release-boundary housekeeping for v0.0.73. Reverts NVIDIA#6202 without closing its related issues. ## Changes - Revert the Deep Agents Code `status`/identity wrapper commands introduced by NVIDIA#6202. - Revert the OpenShell TLS-key secret-guard exception introduced by NVIDIA#6202. - Revert the Deep Agents Code sandbox-name propagation and accompanying tests. - Remove the deferred user-facing `dcode status` documentation from the v0.0.73 tree. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates - [x] Tests added or updated for changed behavior — exact revert restores the previously tested tree. - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes — NVIDIA#6202 documentation is reverted with the code. - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — the index tree exactly matches the pre-NVIDIA#6202 release tree, and no manual edits were made to the revert. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — no CI waiver requested; the local broad `test-cli` pre-commit hook was skipped after seven unrelated environment-sensitive baseline failures in Hermes file modes, managed-gateway trust state, and state-dir permission modes. Changed-area tests pass and PR CI remains required. ## Verification - [x] PR description includes the DCO sign-off declaration and the commit appears as `Verified` in GitHub. - [x] Normal pre-push hooks passed; all pre-commit hooks except the disclosed broad `test-cli` baseline lane passed. - [x] `npx vitest run --project cli src/lib/onboard/sandbox-create-launch.test.ts --silent=false --reporter=default` — 6/6 passed. - [x] `npx vitest run --project integration test/langchain-deepagents-code-image.test.ts --silent=false --reporter=default` — 53/53 passed. - [ ] Full `npm test` passes (broad runtime changes only) — not rerun; the broad hook's unrelated baseline failures are disclosed above. - [x] Quality Gates section completed with required justifications or waivers. - [x] No secrets, API keys, or credentials committed. - [x] `npm run docs` passes with 0 errors and two pre-existing Fern warnings. - [x] Doc pages follow the style guide. - [ ] New doc pages include SPDX header and frontmatter — no new doc pages. --- Signed-off-by: Carlos Villela <cvillela@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Additional environment details are now carried into runtime setup for better tracing and project context. * **Bug Fixes** * Tightened secret-detection and runtime-value checks to reduce the chance of sensitive values being accepted. * Improved sandbox launch behavior by simplifying how launch settings are prepared. * **Documentation** * Removed outdated guidance about checking the current sandbox from the quickstart. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Signed-off-by: Carlos Villela <cvillela@nvidia.com>
<!-- markdownlint-disable MD041 --> ## Summary <!-- 1-3 sentences: what this PR does and why. --> Restores NVIDIA#6202 now that v0.0.73 has been tagged. This reintroduces `dcode status` identity reporting, managed help, the exact OpenShell TLS-key runtime allowance, sandbox-name propagation, and their docs and tests for v0.0.74, with focused follow-ups that suppress private-key-shaped or encoded-credential metadata and fail closed on malformed or unsupported identity config scalars. ## Related Issue <!-- Fixes #NNN or Closes #NNN. Remove this section if none. --> Restores NVIDIA#6202 after the release-boundary revert in NVIDIA#6223. Related to NVIDIA#6189, NVIDIA#6186, and the complementary runtime hardening in NVIDIA#6082. ## Changes <!-- Bullet list of key changes. --> - Restore `dcode status`, `dcode whoami`, and `dcode identity` without launching the interactive UI. - Restore the exact runtime-only `OPENSHELL_TLS_KEY=/etc/openshell/tls/client/tls.key` allowance while keeping arbitrary paths, values, PEM material, and persisted `.env` entries fail-closed. - Restore Tavily and context-shaped secret detection in the managed Deep Agents Code wrapper. - Mirror canonical private-key block detection before mutable config or runtime metadata can reach `dcode status`, including the managed Slack early-allowlist path, and scan the complete mutable `.deepagents/.env` so raw multiline blocks cannot evade per-line classification. - Reject literal, escaped, percent-encoded, and double-encoded query, fragment, and userinfo delimiters before endpoint metadata can reach `dcode status`. - Restrict the informational TOML reader to known generated sections and complete quoted scalars so malformed, commented, array, and unsupported nested values fall back safely instead of being displayed. - Restore Deep Agents Code sandbox-name propagation through onboarding and startup state. - Restore the original NVIDIA#6202 quickstart documentation, identity tests, secret-boundary tests, and shared fixture, then add a composed onboarding → `start.sh` → `dcode status` handoff test. - Record the NVIDIA#6082 rebase contract: NVIDIA#6232's managed-proxy and secret-filtering paths remain authoritative while NVIDIA#6082's runtime upgrade, auth-store, and mutation-command work layers on afterward. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates <!-- Check all that apply. For any "covered by existing tests", "not applicable", or waiver entry, add a brief justification on the same line or in the Changes section. --> - [x] Tests added or updated for changed behavior — restores the original NVIDIA#6202 identity, secret-boundary, image-contract, and onboarding coverage; adds canonical private-key parity, malformed/unsupported config, encoded endpoint, and composed sandbox-name handoff cases. - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes — restores the original `dcode status` quickstart section. - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — the restoration commit exactly matches the previously reviewed NVIDIA#6202 tree; independent reproduction confirmed the advisor-reported private-key status leak, and focused follow-ups mirror the canonical block pattern, reject raw multiline `.env` key blocks and encoded endpoint delimiters, and fail closed on malformed generated scalars with status, parity, managed-Slack, composed-handoff, no-launch, redaction, and secret-scanner coverage. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — no CI waiver requested; required remote checks remain mandatory. ## Verification <!-- Check each item you ran and confirmed. Leave unchecked items you skipped. Doc-only changes do not require npm test unless you ran it. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub. - [ ] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes — all commit hooks except the disclosed local `test-cli` environment-sensitive baseline lane passed; normal pre-push CLI typecheck and tag-version synchronization passed. - [x] Targeted tests pass for changed behavior — onboarding 8/8 and Deep Agents Code wrapper/image/composed-handoff tests 77/77 after review follow-ups; CLI typecheck, ShellCheck, Biome, test-size/project-overlap gates, detect-private-key, and gitleaks also pass. - [x] Required exact-head live E2Es pass on attempt 1 — `ubuntu-repo-cloud-langchain-deepagents-code` and `cloud-onboard` at `a503d95a81094cd3415ce933b64a0213a4d3aa2d`. - [ ] Full `npm test` passes (broad runtime changes only) — not rerun; remote CI remains required. - [x] Quality Gates section completed with required justifications or waivers. - [x] No secrets, API keys, or credentials committed. - [ ] `npm run docs` builds without warnings (doc changes only) — build passed with 0 errors and two pre-existing Fern warnings. - [x] Doc pages follow the style guide; independent docs review found no additional changes needed. - [ ] New doc pages include SPDX header and frontmatter — no new doc pages. --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Carlos Villela <cvillela@nvidia.com> --------- Signed-off-by: Carlos Villela <cvillela@nvidia.com>
NVIDIA#6176) ## Summary The unknown-action error for sandbox commands listed the valid actions but no concrete example. This appends an example command that uses the sandbox name the user typed, e.g. `nemoclaw <name> connect`, satisfying the last unmet acceptance criterion in NVIDIA#755 (unknown sandbox actions should list valid actions *and* a concrete example). ## Related Issue Refs NVIDIA#755 ## Changes - `src/lib/cli/public-dispatch.ts`: the `unknownPublicAction` branch now prints ` Example: <cli> <name> connect` after the valid-actions list, matching the existing `Did you mean:` / command-order hint patterns already in that file. - `test/cli/unknown-sandbox-action.test.ts`: new test asserting an unknown sandbox action reports the valid actions and the concrete example. ## Type of Change - [x] Code change (feature, bug fix, or refactor) ## Quality Gates - [x] Tests added or updated for changed behavior - [x] Docs not applicable — justification: CLI error-output only; no user-facing docs affected. ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Git hooks passed during commit and push - [x] Targeted tests pass for changed behavior - [x] No secrets, API keys, or credentials committed Ran: `npx @biomejs/biome check` (pass), `npm run typecheck` (pass), `npm run build:cli`, and `vitest run test/cli/unknown-sandbox-action.test.ts` (pass). --- Signed-off-by: latenighthackathon <latenighthackathon@users.noreply.github.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Improved CLI guidance when an unknown sandbox action is entered. * The error message now includes a clearer example command, using the sandbox name when available. * Added test coverage to confirm the CLI shows the expected error, valid actions, and example usage. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Signed-off-by: latenighthackathon <latenighthackathon@users.noreply.github.com> Co-authored-by: latenighthackathon <latenighthackathon@users.noreply.github.com>
<!-- markdownlint-disable MD041 --> ## Summary Run four MCP lifecycle and runtime-capability suites directly in Vitest instead of cold-starting a Node process for every scenario. The focused local run falls from 14.25 seconds to 2.97 seconds wall-clock while preserving all 52 behaviors, real registry filesystem state, and the two real OpenShell fixture version probes. ## Related Issue Contributes to NVIDIA#6245. ## Changes - Replace 55 outer Node isolation children across the destroy, Hermes startup, and Deep Agents lifecycle/capability suites with direct source calls and hoisted Vitest mocks. - Preserve per-scenario provider, attachment, adapter, policy, recovery, and environment isolation. - Keep real temporary registry and lifecycle-lock filesystem behavior plus the two nested OpenShell fixture version checks. - Leave production modules and genuine crash/process-contract tests unchanged. ## 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) ## Quality Gates <!-- Check exactly one tests line and one docs line. Check other lines when applicable. Add every requested justification or approval reference. --> - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: This only changes internal test isolation mechanics; MCP commands, configuration, registry semantics, and runtime behavior are unchanged. - [ ] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each applicable item only when supported by the requested evidence. Run targeted tests once per relevant change set and rerun after later edits or hook autofixes that can affect the tested behavior. Do not rerun hook-covered checks. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — command/result or justification: `npx vitest run --project integration test/mcp-destroy-lifecycle.test.ts test/hermes-mcp-startup-probe.test.ts test/deepagents-mcp-legacy-lifecycle.test.ts test/deepagents-mcp-runtime-capability.test.ts` (4 files, 52/52 tests passed; 2.65s Vitest / 2.97s wall-clock); `npm run test-conditionals:scan -- --top 25` (completed; changed files contain zero `if` statements). - [ ] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Carlos Villela <cvillela@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Tests** * Improved coverage for lifecycle, startup, and runtime capability-probe flows. * Reworked multiple suites to run in-process with mocked dependencies for faster, more reliable execution. * Expanded assertions around destroy/rebuild/restore behaviors, including rollback/reattach paths and durable-marker handling. * Strengthened validation of cleanup and state transitions, and improved error-message checking to better prevent regressions. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
<!-- markdownlint-disable MD041 --> ## Summary Fix legacy upgrades so pre-fingerprint managed sandboxes can be rebuilt only after a complete backup and explicit, exact-name operator confirmation. The prepared recovery path can restore the replacement gateway's exact provider and inference route without weakening ordinary rebuild checks or custom-image protections. ## Related Issue Fixes NVIDIA#6114 ## Changes - Prepare the current NemoClaw CLI before sandbox recovery and require every managed sandbox to be backed up successfully before continuing. - Scope legacy recovery to the exact confirmed OpenClaw and Hermes sandbox-name set, reject custom-image and unsupported-agent cases, and revalidate immediately before deletion. - Let only a validated prepared backup defer the fresh-gateway route check and reconstruct a missing canonical provider when its exact recorded credential binding has a host key. - Carry that authority through an immutable provider/model/credential/endpoint handoff, recheck provider absence and credential availability at the delete edge, then force and verify the exact provider and route during authoritative onboarding. - Keep ordinary rebuilds fail-closed for missing, mismatched, or indeterminate gateway state; skip redundant generic onboarding after successful recovery. - Add focused CLI, installer integration, live E2E contract, security-race, and documentation coverage. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: Final independent security review passed with no P0–P2 findings. It covered exact-name authorization, custom-image vetoes, strict backup handling, bounded and exact provider-absence parsing, canonical credential binding, immutable recovery authority, delete-edge revalidation, and exact post-setup route verification. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — command/result or justification: Broader focused suites passed with 160 CLI tests and 85 installer/onboarding integration tests; extraction reruns passed 134 focused CLI and 68 onboarding integration tests. Final provider-recovery follow-up verification passed 92 focused tests and `npm run typecheck:cli`. - [x] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: Final-head GitHub CI passed with 40 successful checks and 2 expected skips. An earlier local broad run completed with 13,441 passed and 8 environment-sensitive failures; affected focused reruns pass. - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) — command/result: passed with 0 errors and 2 existing Fern warnings. - [x] 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) ### Live E2E - [x] Final-head `openshell-gateway-upgrade` passed in [run 28837548991](https://github.com/NVIDIA/NemoClaw/actions/runs/28837548991). Attempt 1 exhausted runner Docker storage before current code ran; attempt 2 completed the legacy install, gateway replacement, provider/route recovery, state restore, and survivor assertions. - [x] Remaining advisor-required targets `cloud-onboard`, `onboard-repair`, `onboard-resume`, and `upgrade-stale-sandbox` passed in [run 28835500944](https://github.com/NVIDIA/NemoClaw/actions/runs/28835500944). - [x] Additional `rebuild-hermes`, `rebuild-openclaw`, `sandbox-rebuild`, and `state-backup-restore` jobs passed in that run. The Hermes cross-product matrix gap is [explicitly justified](NVIDIA#6362 (comment)). --- Signed-off-by: Carlos Villela <cvillela@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added scripted support for confirming legacy managed sandboxes during OpenShell upgrade and recovery, including exact JSON-array confirmation of eligible sandbox names. * Recovery now reports when existing sandboxes are recovered and skips generic onboarding after successful recovery. * **Bug Fixes** * Strengthened pre-upgrade backup strictness: upgrades now stop with a nonzero exit if any registered sandbox is skipped or fails; strict handling can’t be bypassed by unreachable-backup overrides. * Tightened validation around inference/provider recovery to prevent unintended automatic rebuilds. * **Documentation** * Updated quickstart, command reference, lifecycle, and credential-storage docs to reflect the new upgrade/recovery flow, failure behavior, and lifecycle environment variables. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Carlos Villela <cvillela@nvidia.com> Co-authored-by: Prekshi Vyas <34834085+prekshivyas@users.noreply.github.com>
) ## Summary Extract the repeated synchronous file snapshot/restore and JSON state helpers used by the stateful rebuild E2E scenarios. The shared utility preserves the existing absent-file, empty-file, malformed-JSON, parent-directory, and formatted-newline behavior. ## Related Issue Closes NVIDIA#6348 Parent epic: NVIDIA#6346 ## Changes - Add a discriminated `FileSnapshot` type that cannot confuse an absent file with an empty one. - Add shared snapshot, restore, strict JSON read, missing-file fallback, parse-error fallback, and formatted JSON write helpers. - Keep malformed JSON behavior explicit through separate `readJsonFileOr` and `readJsonFileOrFallback` APIs. - Migrate rebuild-openclaw, rebuild-hermes, and stale-sandbox upgrade state handling. - Add focused support tests for absent, empty, populated, malformed, and nested-path cases. ## 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) ## Quality Gates - [x] Tests added or updated for changed behavior - [x] Docs not applicable — justification: test-state utility only - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed - [x] Targeted behavior tests pass — `e2e-file-state.test.ts` (3 passed) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed Additional local verification: - `npm run build:cli` - `npm run typecheck:cli` - `npm run lint` ## Advisor / Merge Coordination Notes - **Overlapping E2E refactor PRs:** PR NVIDIA#6358 and PR NVIDIA#6357 touch the same rebuild/stale-sandbox live test files. This PR should land first because it only extracts shared file-state helpers; the overlapping PRs can then rebase and keep using `test/e2e/fixtures/file-state.ts`. Any conflicts are expected to be mechanical import/helper-call conflicts, not runtime behavior conflicts. - **`readJsonFileOrFallback` semantics:** The malformed-JSON fallback is intentional only for stale-sandbox E2E setup state where prior test runs may leave incomplete registry/session JSON. The helper catches only `SyntaxError`; non-parse I/O/read failures still propagate, and `test/e2e/support/e2e-file-state.test.ts` covers that negative case. - **Sensitive-path waiver:** Changes are test-only under `test/e2e/**`; no production credential, policy, onboarding, inference, runner, sandbox runtime, or messaging code is modified. Maintainer approval by `cv` is recorded on this PR. --- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added shared E2E file-state utilities for snapshot/restore and JSON read/write, with consistent pretty-printed JSON output and a trailing newline. * Introduced two JSON fallback behaviors: fallback only when a file is missing, or fallback when the file is missing and/or contains malformed JSON. * **Tests** * Added E2E coverage for restore behavior (absent vs. empty), nested directory handling for writes, and the JSON fallback/throw rules. * **Chores** * Updated multiple live E2E tests to reuse the shared file-state utilities instead of local helpers. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
…VIDIA#6306) <!-- markdownlint-disable MD041 --> ## Summary <!-- 1-3 sentences: what this PR does and why. --> `nemoclaw gc` never removed orphaned locally-prebuilt sandbox images because it scanned only the gateway-built image repository (`openshell/sandbox-from`). This makes gc enumerate every sandbox image repo, so orphans under `nemoclaw-sandbox-local` (the docker-driver-gateway prebuild path — Linux or macOS/Apple Silicon) are detected, and the "run 'nemoclaw gc' to clean up" remediation that `destroy --force` prints actually works. ## Related Issue <!-- Fixes #NNN or Closes #NNN. Remove this section if none. --> Fixes NVIDIA#6301 ## Changes <!-- Bullet list of key changes. --> - `src/lib/domain/sandbox/image-tag.ts`: add `SANDBOX_FROM_IMAGE_REPO`, `LOCAL_SANDBOX_IMAGE_REPO`, and `SANDBOX_IMAGE_REPOS` as the single source of truth for sandbox image repositories. - `src/lib/actions/maintenance.ts`: `garbageCollectImages()` now queries every repo in `SANDBOX_IMAGE_REPOS` and unions the results before orphan detection (was hardcoded to `openshell/sandbox-from` only). - `src/lib/onboard/sandbox-prebuild.ts`: reuse the shared `LOCAL_SANDBOX_IMAGE_REPO` constant instead of a private literal. - Tests: `images.test.ts` gains a mixed-repo case (local orphan detected, registered local image preserved); `maintenance.test.ts` asserts gc scans both repos. ## Type of Change - [√] 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) ## Quality Gates <!-- Check exactly one tests line and one docs line. Check other lines when applicable. Add every requested justification or approval reference. --> - [√] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [ ] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each applicable item only when supported by the requested evidence. Run targeted tests once per relevant change set and rerun after later edits or hook autofixes that can affect the tested behavior. Do not rerun hook-covered checks. --> - [√] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [√] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [√] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — command/result or justification: - [ ] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: - [√] Quality Gates section completed with required justifications or waivers - [√] No secrets, API keys, or credentials committed - [ ] `npm run 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Rui Luo <ruluo@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Improved sandbox image cleanup to scan all sandbox image repositories, rather than a single hard-coded source, to remove unused images more reliably. * Fixed local sandbox image reference handling so repository names and tags are generated consistently across the app. * Enhanced orphan image detection to correctly flag unregistered local images even when repository context differs. * **Tests** * Added coverage for orphan-detection and garbage-collection behavior across both sandbox image repositories. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Rui Luo <ruluo@nvidia.com>
…nboarding (NVIDIA#6289) (NVIDIA#6297) ## Summary Custom Anthropic-compatible endpoints were validated with a non-streaming `/v1/messages` probe only, so endpoints whose SSE streaming layer is malformed (e.g. duplicate `message_start` events with the same message id, observed on the Inference Hub route) passed onboarding and then failed at runtime inside the sandbox with Hermes' cryptic `no final response was produced` (exit 1). This PR adds streaming-event validation to the custom Anthropic onboarding path — mirroring the existing `/v1/responses` streaming validation from NVIDIA#1833 — so the defect is diagnosed at onboarding time with an actionable message. ## Related Issue Refs NVIDIA#6289 This PR is the onboarding guardrail for the malformed Anthropic SSE signature. It intentionally rejects a broken compatible endpoint before sandbox creation; the functional runtime routing fix remains in NVIDIA#6295. ## Changes - Add `runAnthropicStreamingEventProbe` (`src/lib/adapters/http/probe.ts`): sends a `stream: true` request to `/v1/messages` and validates the SSE event sequence — exactly one `message_start`, at least one `content_block_delta`, one `message_stop`. Refactors the shared curl/SSE capture into `captureSseEventCounts`, reused by the existing `/v1/responses` streaming probe with identical behavior (trace events, temp cleanup, curl exit-28 tolerance). - Wire the streaming probe into `probeAnthropicEndpoint` behind a new `{ probeStreaming }` option (`src/lib/inference/probe-anthropic.ts`); tighter `--max-time 15` timing so validation cannot hang the wizard. - Enable it in `validateCustomAnthropicSelection` (`src/lib/onboard/inference-selection-validation.ts`). Skipped when `NEMOCLAW_REASONING=true`, matching the custom OpenAI-compatible path. The official Anthropic provider path is unchanged; Bedrock-classified endpoints are unaffected (streaming probe only runs after the non-streaming `/v1/messages` probe succeeds). - Move the two Anthropic credential-retry integration tests into a focused `test/onboard-selection-anthropic-retry.test.ts` (fake curl now serves a well-formed SSE stream for `"stream":true` probe bodies), shrinking `test/onboard-selection.test.ts` below its legacy size budget; budget ratcheted down 6146 → 5935. - Docs: validation-table row and Anthropic-compatible server section in `docs/inference/inference-options.mdx`; new troubleshooting entry for the onboarding-time duplicate `message_start` failure in `docs/reference/troubleshooting.mdx`. ## Type of Change - [x] Code change with doc updates ## Quality Gates - [x] Tests added or updated for changed behavior - [x] Docs updated for user-facing behavior changes - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: requesting maintainer review; probes reuse the existing trusted `--config` credential routing (no credential appears in argv), and the new probe adds no new credential surface. ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — command/result: `npx vitest run src/lib/adapters/http/probe.test.ts src/lib/inference/probe-anthropic.test.ts src/lib/onboard/inference-selection-validation.test.ts --project cli` (68 passed); `npx vitest run test/onboard-selection.test.ts --project integration` (67 passed); `npx vitest run test/onboard-selection-anthropic-retry.test.ts --project integration` (2 passed); `npx vitest run src/lib/inference/onboard-probes.test.ts src/lib/inference/onboard-probes-responses-fallback.test.ts --project cli` (25 passed, 1 skipped); `npm run test:projects:check` (disjoint); `npm run typecheck:cli` (clean) - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [x] `npm run docs` builds without warnings (doc changes only) — 0 errors; 2 pre-existing warnings on main (login-gated redirects check, theme contrast), unrelated to this change ## DCO Signed-off-by: Tony Luo <xialuo@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added Anthropic-compatible streaming SSE validation during onboarding to catch malformed `/v1/messages` event sequences. * Streaming validation can be skipped for reasoning-only models via `NEMOCLAW_REASONING`. * **Bug Fixes** * Improved reporting and recovery behavior for missing, duplicate, and out-of-order streaming events, including clearer diagnostic messaging. * **Documentation** * Updated inference validation docs and added troubleshooting guidance for onboarding failures caused by malformed SSE events. * **Tests** * Expanded streaming probe and onboarding validation tests; added Anthropic onboarding retry UX coverage; removed obsolete Anthropic retry tests. * **Chores** * Adjusted CI test file size budget. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Tony Luo <xialuo@nvidia.com> Signed-off-by: Carlos Villela <cvillela@nvidia.com> Co-authored-by: Carlos Villela <cvillela@nvidia.com>
NVIDIA#6000) (NVIDIA#6007) ## Summary Fresh sandboxes print a spurious plugin warning on the first `openclaw tui` launch: ``` plugins.entries.qqbot: plugin not installed: qqbot ``` The generated default `openclaw.json` listed a `qqbot` plugin entry, but `qqbot` is not bundled in the NemoClaw OpenClaw image. OpenClaw's plugin loader warns `plugin not installed` for any referenced plugin that is absent on disk, regardless of `enabled: false`, so every user who never configured a messaging channel sees a warning that implies a broken install. `acpx` and `bonjour` are kept: they are bundled plugins, so listing them with `enabled: false` correctly disables a present plugin without a warning. Only the absent `qqbot` entry is removed. ## Changes - `scripts/generate-openclaw-config.mts`: remove the dangling `qqbot` default plugin entry. - `test/generate-openclaw-config-plugin-entries.test.ts`: new focused test asserting the default entries keep `acpx`/`bonjour` but no longer reference `qqbot` (split into its own file to respect the test-file size budget). ## Test - `npx vitest run test/generate-openclaw-config-plugin-entries.test.ts` — 2 passing. - `source-shape:check` and `check-test-file-size-budget.ts` pass. Closes NVIDIA#6000 Signed-off-by: Abhimanyu Kumar <abhimanyukumar7290@gmail.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Updated generated plugin configuration so the default plugin list no longer includes an unused entry. * Ensured bundled plugins continue to be generated with disabled defaults where expected. * **Tests** * Added coverage for default plugin entry generation. * Verified the configuration does not include the removed plugin entry. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Signed-off-by: Abhimanyu Kumar <abhimanyukumar7290@gmail.com> Co-authored-by: San Dang <sdang@nvidia.com>
… (NVIDIA#6008) ## Summary When `nemoclaw onboard` fails or is interrupted partway through, most of its many `process.exit(1)` paths print no recovery guidance. Users assume a transient failure (network timeout during an image pull, a bad key, a Docker hiccup) requires a full reinstall, when the run is actually resumable with `nemoclaw onboard --resume` (NVIDIA#6003). ## Approach Rather than touch every scattered exit site, add a single catch-all in the centralized incomplete-onboard exit handler (`exit-step-failure.ts`). It fires on any non-zero exit while onboarding is incomplete and, **when a step was in progress** (a resumable point recorded in the session), prints: ``` Onboarding did not finish. Resume from the step that failed with: nemoclaw onboard --resume Completed steps are skipped; pass --fresh instead to start over. ``` This automatically covers the non-interactive abort paths and any uncaught failure. The hint is suppressed where it would be wrong or redundant: - **Already-handled paths:** the sandbox build-context explainer prints tailored `--resume` guidance and calls `noteOnboardResumeHintShown()`, so the backstop never duplicates it. - **Ctrl-C cancel:** the cancel-rollback clears the onboard session before this handler runs, so no resumable step remains and the hint stays silent (resuming a discarded session would be misleading). - **Early exits** before any step started leave `lastStepStarted` unset, so the handler stays silent. ## Changes - `src/lib/onboard/resume-hint.ts` (new): once-per-process `printOnboardResumeHint()` + `noteOnboardResumeHintShown()` dedup latch. - `src/lib/onboard/exit-step-failure.ts`: print the hint from the incomplete-exit handler when a step was in progress. - `src/lib/build-context.ts`: mark the hint shown on the sandbox-create path that already prints tailored `--resume` guidance. - Tests: `resume-hint.test.ts` (new) and added cases in `exit-step-failure.test.ts`. ## Test - `npx vitest run src/lib/onboard/resume-hint.test.ts src/lib/onboard/exit-step-failure.test.ts test/build-context.test.ts` — all passing. - `tsc -p jsconfig.json`, `source-shape:check`, `check-test-file-size-budget.ts` pass. Closes NVIDIA#6003 Signed-off-by: Abhimanyu Kumar <abhimanyukumar7290@gmail.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a new onboarding recovery hint that points users to resume interrupted setup with a resume command. * The hint is shown only once, reducing repeated messages. * **Bug Fixes** * Improved exit-time messaging so the resume hint appears only when onboarding was actually in progress. * Prevented duplicate recovery hints from being shown in overlapping recovery paths. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Abhimanyu Kumar <abhimanyukumar7290@gmail.com> Signed-off-by: Carlos Villela <cvillela@nvidia.com> Co-authored-by: Carlos Villela <cvillela@nvidia.com>
…ult (NVIDIA#6326) <!-- markdownlint-disable MD041 --> ## Summary The generic-Linux managed-vLLM default, `nvidia/NVIDIA-Nemotron-3-Nano-4B-FP8`, was launched without `--enable-auto-tool-choice` or a tool-call parser. Plain completions succeeded, but agent requests using `tool_choice: "auto"` failed HTTP 400. This PR pins `--tool-call-parser qwen3_coder`, matching NVIDIA's model-card launch recipe, and adds regression coverage at both the model-registry and generic-Linux profile layers. ## Related Issue Closes NVIDIA#6314. ## Reproduction and Analysis The reporter reproduced the failure on Ubuntu 24.04 x86_64 with an RTX 5070 (12 GB) using NemoClaw v0.0.74: ```bash NEMOCLAW_EXPERIMENTAL=1 NEMOCLAW_PROVIDER=install-vllm \ nemoclaw onboard --agent <agent> --name <name> --non-interactive --fresh --yes ``` Before this fix, `buildVllmServeCommand` produced a command ending with: ```text ... --gpu-memory-utilization 0.7 --load-format fastsafetensors ``` The reporter observed: ```text PLAIN: HTTP_CODE=200 TOOLCALL: HTTP_CODE=400 "auto" tool choice requires --enable-auto-tool-choice and --tool-call-parser to be set ``` `GENERIC_LINUX_PROFILE.defaultModel` resolves to the `nemotron-3-nano-4b` registry entry. That entry was the outlier: the DGX Spark (`qwen3.6-35b-a3b-nvfp4` / `qwen3_xml`) and DGX Station (`deepseek-v4-flash` / `deepseek_v4`) defaults already pinned their own tool-call parsers. After this fix, the command ends with: ```text ... --gpu-memory-utilization 0.7 --load-format fastsafetensors \ --enable-auto-tool-choice --tool-call-parser qwen3_coder ``` The parser choice comes from the NVIDIA Nemotron-3-Nano-4B-FP8 model-card vLLM example and uses a parser already present in this registry. ## Changes - Add `--enable-auto-tool-choice --tool-call-parser qwen3_coder` to the Nemotron-3-Nano-4B FP8 registry entry. - Cover the generated serve command and the generic-Linux profile default with regression tests. - Require each tool-call switch to occur exactly once in the generated command. - Merge current `main` into the contributor branch without rewriting contributor history. ## 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) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: this corrects the baked serve arguments for an existing managed-vLLM model and requires no new user action; current docs already describe the generic-Linux default, managed-vLLM limitations, and the need for model-appropriate tool-call flags. - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: maintainer review by @ericksoa on the final diff (2026-07-06) covered all nine security-review categories. The change adds fixed model-specific argv tokens, does not introduce user-controlled interpolation or alter credential/auth/policy boundaries, retains negative-path parser coverage, and passed gitleaks; no security findings. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable — `npm run check:diff` passed at `4f847e8f76373ab3d62386ea4902a0263a70f353`. - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — `npx vitest run src/lib/inference/vllm-models.test.ts src/lib/inference/vllm.test.ts` (47/47 passed). - [x] Applicable broad gate passed — `npx vitest run src/lib/inference/` (502 passed, 1 pre-existing skip), `npm run typecheck:cli`, and `npm run build:cli` passed. - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) ## Remaining Runtime Evidence Live end-to-end proof on an x86_64 generic-Linux NVIDIA GPU host is not included. The available fleet does not have that architecture/GPU combination; the DGX Spark host exercises a different profile that was already unaffected. The unit tests prove the exact command/profile composition, and the required `inference-routing` E2E is being run at the final head, but that E2E does not launch managed vLLM or validate this parser against the model. Before merge, a maintainer must either explicitly accept this hardware-evidence limitation or obtain a generic-Linux GPU run showing the fixed command and a successful `tool_choice: "auto"` request. ## AI Disclosure - [x] AI-assisted — contributor tool: Claude Code; maintainer salvage tool: Codex --- Signed-off-by: Yanyun Liao <yanyunl@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Updated vLLM settings for the Nemotron-3-Nano-4B FP8 model to enable automatic tool choice with the correct tool-call parser configuration. * Fixed Linux/NVIDIA vLLM profile detection to include the expected tool-call parser settings. * **Tests** * Added regression coverage to ensure the tool-call related flags are included correctly going forward. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Yanyun Liao <yanyunl@nvidia.com> Signed-off-by: Aaron Erickson <aerickson@nvidia.com> Co-authored-by: Aaron Erickson <aerickson@nvidia.com>
## Summary Add immutable E2E command-environment profiles on top of the existing filtered availability boundary, and migrate the repeated test-HOME builders used by six live scenarios. ## Related Issue Closes NVIDIA#6356 Parent epic: NVIDIA#6346 ## Changes - Add filtered command, installed-CLI, test-HOME, and sandbox environment profiles. - Centralize noninteractive acceptance flags, default OpenShell gateway selection, and deduplicated `.local/bin` / `.npm-global/bin` PATH entries. - Preserve caller-override precedence and avoid `process.env` mutation. - Keep secret-bearing inference overlays as explicit caller data rather than embedding inference-mode semantics. - Migrate six exact/near-exact test-HOME environment builders. - Add support tests for filtering, precedence, sandbox identity, PATH composition, immutability, and secret overlays. ## Verification - [x] Signed/Verified commit; pre-commit, commit-msg, and pre-push hooks passed - [x] `npm run build:cli` - [x] `npm run typecheck:cli` - [x] `npm run lint` - [x] `e2e-environment-profiles.test.ts` (3 passed) - [x] No user-facing docs required - [x] No secrets or credentials committed --- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added shared environment presets for CLI, home-directory, and sandboxed command runs. * Standardized test command environments so local CLI paths and required settings are applied consistently. * **Bug Fixes** * Improved end-to-end test reliability by removing duplicated environment setup across multiple scenarios. * **Tests** * Added coverage for environment profile behavior, including command setup, home-path handling, and sandbox-specific variables. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Co-authored-by: Carlos Villela <cvillela@nvidia.com>
…VIDIA#6365) ## Summary Expose the existing redacting Docker probe through the live Vitest fixture with explicit required, optional, and intentionally-missing policies. Closes NVIDIA#6354 Parent epic: NVIDIA#6346 ## Changes - Add `DockerPrerequisite` with `probeDocker`, `requireDocker`, and `expectMissingDocker`. - Centralize local-skip versus CI-failure behavior for required Docker. - Expose the prerequisite as the `docker` E2E fixture. - Migrate the sandbox-operations prerequisite block. - Preserve isolated Docker config, redacted output, and artifacts through `DockerProbe`. - Add 9 focused prerequisite/probe tests. ## Verification - [x] Signed/Verified commit and all hooks passed - [x] `npm run build:cli` - [x] `npm run typecheck:cli` - [x] Docker support tests: 9 passed - [x] No product behavior, docs, or secrets changed --- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Improved Docker availability checks for live test runs, with clearer handling when Docker is missing. * Tests now skip more gracefully on local machines and fail more explicitly in CI when Docker is unavailable. * Added coverage for Docker-present, Docker-missing, and expected-missing scenarios to make test behavior more reliable. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Co-authored-by: Carlos Villela <cvillela@nvidia.com>
) ## Summary Consolidate shared fake-provider HTTP mechanics—request bodies, JSON/SSE responses, and deterministic server listen/close—without changing mock-versus-hosted inference selection (NVIDIA#5745). Closes NVIDIA#6349 Parent epic: NVIDIA#6346 Related: NVIDIA#5745 ## Changes - Add reusable request-body, JSON response, raw/named SSE, listen, and close primitives. - Migrate messaging-compatible, MCP bridge, and Hermes inference-switch mock servers. - Preserve scenario-specific request logs, streaming shapes, auth checks, and cleanup. - Document the protocol-only boundary with NVIDIA#5745. - Add focused real-server protocol tests. ## Verification - [x] Signed/Verified commit and all hooks passed - [x] `npm run build:cli` - [x] `npm run typecheck:cli` - [x] `npm run lint` - [x] Protocol and migrated-helper support tests: 23 passed - [x] No production provider behavior or inference-mode policy changed --- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Tests** * Added shared HTTP test helpers for reading request bodies, writing JSON responses, and sending SSE streams. * Updated end-to-end tests to use the shared helpers for server setup, response handling, and cleanup. * Added coverage for HTTP request/response behavior, including JSON and streaming event responses. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Co-authored-by: Carlos Villela <cvillela@nvidia.com>
## Summary Add a typed target-evidence facade to the existing artifact sink and migrate live Vitest targets to it. Target metadata/results now retain the same artifact filenames and redaction boundary while receiving a consistent runner, result status, and contract field shape. ## Related Issue Closes NVIDIA#6353 Parent epic: NVIDIA#6346 ## Changes - Add `artifacts.target.declare(...)` and `artifacts.target.complete(...)` with typed extensible metadata/result inputs. - Emit `target.json` and `target-result.json` through the existing redacting `ArtifactSink`. - Supply `runner: "vitest"` centrally and default completed results without an explicit status to `status: "passed"`. - Normalize legacy singular `contract` values to the canonical `contracts` array and reject conflicting or malformed contract fields. - Validate non-empty target IDs and explicit result statuses. - Migrate all 117 live target evidence writes and remove repeated runner fields. - Add focused support tests for emitted files, normalization, validation, redaction, and adoption. ## 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) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: test-evidence refactor with no user-facing behavior - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — 28 focused artifact, fixture, and redaction assertions passed - [ ] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) Additional local verification: - `npm run build:cli` - `npm run typecheck` - `npm run typecheck:cli` - `npm run lint` - `npx vitest run --project e2e-support test/e2e/support/e2e-target-evidence.test.ts test/e2e/support/e2e-fixture-context.test.ts test/e2e/support/e2e-redaction-entry.test.ts` --- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a typed E2E “target evidence” lifecycle with explicit **declare**/**complete** steps, producing consistent, normalized evidence outputs. * **Bug Fixes** * Improved validation and normalization (non-empty identifiers, non-empty status when provided, enforce single `contract` vs `contracts`, consistent contract array formatting, and automatic result defaults). * Added secret redaction in persisted evidence details. * **Tests** * Migrated many live E2E scenarios to use the typed evidence API. * Added test coverage for evidence normalization/redaction and a safeguard scan to prevent direct JSON target writes. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com>
## Summary Translate Hermes top-level resumed one-shot invocations through the native non-interactive chat resume path so the new turn appends to the selected session instead of fragmenting into a fresh session. ## Related Issue Fixes NVIDIA#5254 ## Changes - Detect top-level `--resume`/`--continue` combined with `-z`/`--oneshot` in the installed Hermes wrapper. - Rewrite only that composed form to `hermes chat --query ... --quiet --resume/--continue ...`, preserving model/tool/config flags and leaving plain one-shot invocations untouched. - Add wrapper regression coverage for resumed one-shot, continued one-shot, and unchanged plain one-shot behavior. ## 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) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: fixes existing documented flag composition without changing user-facing syntax. - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: scoped wrapper argument rewrite only; existing `gateway` guard and `config show` masker branches remain first and unchanged, and the new path delegates to Hermes' native `chat --query --quiet --resume/--continue` implementation instead of directly editing session storage. - [x] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: local `npx vitest run test/hermes-gateway-wrapper.test.ts` is skipped by the existing Linux+python3 `canRun` gate on macOS; CI Linux should execute it. ## Verification - [ ] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [ ] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — command/result or justification: `python3 -m py_compile agents/hermes/hermes-wrapper.py` passed; direct Python assertions for `_translate_resumed_oneshot` passed; `NPM_CONFIG_CACHE=/tmp/nemoclaw-5254-npm-cache npx biome check test/hermes-gateway-wrapper.test.ts` passed; `NPM_CONFIG_CACHE=/tmp/nemoclaw-5254-npm-cache npx vitest run test/hermes-gateway-wrapper.test.ts` loaded but skipped all 45 tests due the existing macOS gate. - [ ] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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: Chengjie Wang <chengjiew@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added automatic routing that rewrites resumed/continued one-shot agent invocations into Hermes’ non-interactive `chat --query` resume flow. * **Bug Fixes** * Fail-closed argv translation: only rewrites when inputs match supported resume/continue patterns; otherwise runs the original command unchanged. * Preserves supported arguments while safely handling unsupported/ambiguous forms (including `--` termination and conflicting usage). * **Tests** * Expanded Hermes one-shot/resume/continue routing assertions, including updated argv recording (`realArgv`). * Added Hermes e2e regression to verify resume/continue don’t create extra sessions and export content matches the seeded marker. * Updated sandbox rebuild recovery test harness to use shared environment helpers. * **Chores** * Updated Hermes wrapper integrity pin and added build-time verification to detect flag allowlist drift. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
… (NVIDIA#5868) ## Summary Adds the missing arm64 Local NVIDIA NIM image-compatibility warning during onboarding. GB10 detection and the NIM-local provider menu are already working on current `main`, so this PR leaves that behavior alone. The new warning is advisory only: it tells Linux arm64 DGX Spark/Station users that some NIM images may not publish `linux/arm64` manifests, while still allowing the existing NIM pull/start path to try the selected image. Fixes NVIDIA#5772 ## Changes - `src/lib/onboard/nim-image-compat-warning.ts`: adds a small helper for the Linux arm64 DGX Spark/Station warning. - `src/lib/onboard/provider-host-state.ts`: prints the warning once when `nim-local` is available in the provider options. - `src/lib/onboard/nim-image-compat-warning.test.ts`: covers the warning eligibility and wording. - `test/onboard-nim-image-compat-warning.test.ts`: exercises the compiled onboarding flow with a fake Linux arm64 DGX Spark GPU and confirms the warning appears when Local NIM is offered. ## Testing - `npm install --ignore-scripts` - completed. - `npm run build:cli` - passed. - `npx vitest run src/lib/onboard/nim-image-compat-warning.test.ts test/onboard-nim-image-compat-warning.test.ts` - passed. - `npm run typecheck:cli` - passed. - `npm run source-shape:check` - passed outside the sandbox after the sandboxed `tsx` IPC pipe failed with `EPERM`. - `npm run test-size:check` - passed outside the sandbox after the sandboxed `tsx` IPC pipe failed with `EPERM`. - `git diff --check` - passed. - `npx @biomejs/biome format src/lib/onboard/nim-image-compat-warning.ts src/lib/onboard/nim-image-compat-warning.test.ts src/lib/onboard.ts test/onboard-nim-image-compat-warning.test.ts` - passed. - `npx @biomejs/biome lint src/lib/onboard/nim-image-compat-warning.ts src/lib/onboard/nim-image-compat-warning.test.ts src/lib/onboard.ts test/onboard-nim-image-compat-warning.test.ts` - passed. ## Evidence it works The focused onboarding regression overrides `process.arch` and `process.platform` to simulate a Linux arm64 DGX Spark host, passes a GB10-style GPU object with `nimCapable: true`, enables `NEMOCLAW_EXPERIMENTAL=1`, and selects the default cloud provider so no real NIM image is pulled. The test confirms onboarding still completes with `nvidia-prod` while the output contains both the Local NIM arm64 warning and the `linux/arm64` manifest note. Signed-off-by: Deepak Jain <deepujain@gmail.com> --------- Signed-off-by: Deepak Jain <deepujain@gmail.com>
## Summary Adds a focused unit-test suite for the `isRecord` type guard in `src/lib/core/json-types.ts`, which previously had no co-located test. This raises coverage on a shared utility used across CLI data boundaries (onboard, policies, agent-onboard) with no production code changes. ## Changes - Add `src/lib/core/json-types.test.ts` with Vitest cases for `isRecord`: - returns `true` for plain, empty, and nested objects - returns `false` for `null`, `undefined`, arrays, strings, numbers, and booleans ## 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] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [ ] Git hooks passed during commit and push, or `npx prek run --from-ref main --to-ref HEAD` passes - [ ] Targeted tests pass for changed behavior - [ ] Full `npm test` passes (broad runtime changes only) - [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 - [ ] `npm run docs` builds without warnings (doc changes only) - [ ] Doc pages follow the style guide (doc changes only) - [ ] New doc pages include SPDX header and frontmatter (new pages only) --- Signed-off-by: Atulya Singh <atulyarajsingh@gmail.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Tests** * Added comprehensive test coverage for record validation logic, verifying correct behavior with various input types including objects, null, undefined, arrays, and primitive values. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
## Summary Centralize the live Vitest E2E command-result formatter and zero-exit assertion on the existing fixture client helpers. This removes the equivalent local helper copies exposed by the post-migration audit and adds a ratchet against reintroducing them. ## Related Issue Closes NVIDIA#6355 Parent epic: NVIDIA#6346 ## Changes - Generalize the shared command-result types so lightweight subprocess results can use `resultText` and `assertExitZero` without casts. - Replace equivalent local helpers across live targets and the onboarding phase with shared imports. - Preserve the existing Phase 6 helper APIs through re-exports while moving their implementation to the fixture client. - Rename the richer runtime-overrides spawn formatter instead of changing its intentionally different diagnostics. - Add an E2E support guard that rejects new local `resultText` or `expectExitZero` definitions in live targets. ## 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) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: test-harness-only refactor with no user-facing behavior - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — `npx vitest run --project e2e-support test/e2e/support/e2e-command-helper-adoption.test.ts test/e2e/support/e2e-clients.test.ts test/e2e/support/messaging-providers-runtime-proofs.test.ts test/e2e/support/openclaw-discord-pairing-helpers.test.ts` (79 passed) - [ ] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) Additional local verification: - `npm run build:cli` - `npm run typecheck` - `npm run lint` - Full `e2e-support` run reached 87 passing files; the unchanged main-branch shell-quote source-of-truth check currently reports `openshell-gateway-upgrade-helpers.ts`, and that test plus source file are unchanged by this PR. --- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Standardized end-to-end command output formatting across flows for more consistent, readable failure messages. * Improved exit-status failure details by deriving them from unified command output and aligning accepted result shapes. * **Tests** * Updated many end-to-end tests to use shared command-result and exit-check helpers instead of local duplicates. * Added coverage for exit-check behavior with lightweight command results. * Added a guard test to prevent future live tests from reintroducing local helper implementations. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
…VIDIA#6383) <!-- markdownlint-disable MD041 --> ## Summary Reduce process isolation in two high-cost integration targets without changing production behavior. Against the final `main` base, stale rebuild recovery drops from 14.94–15.01 seconds to approximately 3.50 seconds (76.6% faster), while provider selection drops from 17.60–18.07 seconds to 9.95–10.23 seconds (43.4% faster). ## Related Issue Part of NVIDIA#6245. ## Changes - replace five redundant full-CLI stale-rebuild launches with the existing direct flow harness while retaining one real CLI/HOME rollback contract - retain a focused 65ms harness case for successful stale recovery, including backup skip, recreate handoff, and finalization assertions - exercise unit-shaped onboarding cases through typed source seams while preserving real-process coverage for module-wiring, environment parsing, and subprocess boundaries - batch nine provider credential back/exit scenarios through their real `setupNim` call sites in one isolated child; full Node boots fall from 50 to 26 with all 66 provider-selection behaviors retained - add a real-classifier rebuild contract for both recorded/active gateway mismatch directions instead of injecting the terminal `wrong_gateway_active` state - ratchet the legacy `onboard-selection.test.ts` size budget from 5,624 to 4,834 lines - sync the latest `main` and correct its stale plugin-entry assertion so the test preserves the intentional omission of `acpx` ## Wrong-Gateway Source-of-Truth Review - **Invalid state:** an empty active-gateway sandbox list is ambiguous when the registry's recorded gateway differs from the currently selected gateway; it must not authorize stale destruction - **Source boundary:** `gateway-state.ts#getReconciledSandboxGatewayState` and its named-gateway reconciliation path own the classification - **Source constraint:** gateway selection is mutable, so a `NotFound` response by itself cannot prove the recorded sandbox is stale on its own gateway - **Regression coverage:** `rebuild-gateway-drift.test.ts` now drives the real classifier for `nemoclaw`/`other-gw` and `nemoclaw-9000`/`nemoclaw`, then verifies no backup, delete, registry removal, or onboard occurs - **Removal condition:** retain this contract until gateway reconciliation and the rebuild mutation decision become one typed, atomic boundary ## 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) ## Quality Gates <!-- Check exactly one tests line and one docs line. Check other lines when applicable. Add every requested justification or approval reference. --> - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: test-isolation and regression-coverage changes only; production behavior, CLI output, configuration, defaults, and public APIs are unchanged - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: independent reviews verified the real wrong-gateway reconciliation path, destructive-boundary assertions, provider call-site wiring, environment cleanup, and retained process boundaries; no actionable findings remain - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each applicable item only when supported by the requested evidence. Run targeted tests once per relevant change set and rerun after later edits or hook autofixes that can affect that behavior. Do not rerun hook-covered checks. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — `test/onboard-selection.test.ts` passed 66/66 at final head; `test/rebuild-stale-recovery.test.ts` passed 5/5 in 3.49 seconds wall; `rebuild-flow.test.ts` plus `rebuild-gateway-drift.test.ts` passed 85/85; the final `main` sync's OpenClaw config test plus the unrelated Hermes stress file passed 33/33 with 2 environment skips - [x] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: [final-head CI](https://github.com/NVIDIA/NemoClaw/actions/runs/28853151732) passed all five CLI coverage shards, the merged coverage gate, static/type checks, package lanes, and sandbox smoke tests - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Carlos Villela <cvillela@nvidia.com>
…ion (NVIDIA#5445) (NVIDIA#6385) ## Summary The OpenClaw commands reference page linked to the Install OpenClaw Plugins page using its **source directory** (`../deployment/install-openclaw-plugins`), but Fern publishes that page under the **`manage-sandboxes`** nav section. The link therefore 404s on the live site even though the source file exists on disk. This restores the published-route link and adds a route-level regression guard so the drift cannot recur. ## Related Issue Fixes NVIDIA#5445 ## Changes - `docs/reference/commands.mdx`: link `Install OpenClaw Plugins` via its published nav section (`../manage-sandboxes/install-openclaw-plugins`) instead of its source directory (`../deployment/install-openclaw-plugins`). - `scripts/check-docs-published-routes.ts`: new checker that derives the published route map from `docs/index.yml` (variant + section slugs) and resolves the commands page's relative links **route-relative, the way Fern serves them**, failing if any resolves to a route that is not published. Scoped to the commands reference page, which has regressed repeatedly (NVIDIA#5445, NVIDIA#6290, NVIDIA#5465, NVIDIA#5460). - `package.json`: add `docs:check-routes` and wire it into `npm run docs:strict`, so a source-path-valid but published-route-broken link fails docs validation. - `test/repro-5445-docs-published-route.test.ts`: Vitest regression that derives the route from `docs/index.yml`, asserts the commands link resolves to `/user-guide/openclaw/manage-sandboxes/install-openclaw-plugins` and **not** `/user-guide/openclaw/deployment/install-openclaw-plugins`, plus resolver/extractor robustness cases. ## Root cause PR NVIDIA#6290, tasked with fixing this link, reasoned from the source file path (`docs/deployment/install-openclaw-plugins.mdx`) and "corrected" a previously-working `manage-sandboxes` link back to `deployment`. `fern check` and source-path checks both pass on the broken form because the source file exists — the missing validation was published-route resolution. ## Type of Change - [x] Code change with doc updates ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: the changed page is the doc; the link target and its content are unchanged, only the route it points to is corrected. ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set — command/result: `npx vitest run test/repro-5445-docs-published-route.test.ts` (8 passed; fails on upstream/main pre-fix, passes post-fix); `npm run docs:strict` (0 errors). - [x] No secrets, API keys, or credentials committed - [x] `npm run docs` builds without warnings (doc changes only) ### Route-level reporter-workflow evidence Ground truth from Fern's own link resolver (`fern docs broken-links`), the same resolution the reader navigates: - **Before (upstream/main):** 26 broken links, including the commands page → `/user-guide/openclaw/deployment/install-openclaw-plugins` (404). - **After (this branch):** 24 broken links — the commands-page install-plugins error is gone, and **none added**. The remaining 24 are pre-existing, unrelated broken links (nested-page relative links, hermes-variant structural gaps) outside this issue's scope. `fern check` alone does not catch this (it passed in NVIDIA#6290); the new `docs:check-routes` gate does. --- Signed-off-by: Yimo Jiang <yimoj@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Corrected a documentation link to point to the published “Install OpenClaw Plugins” page under the correct route. * Improved doc link validation to ensure links resolve to published routes (including correct route selection). * **Tests** * Added/expanded regression coverage for the plugin installation link, published-route mapping, and robust markdown link resolution. * **Chores** * Enhanced the strict documentation checks to also verify published-route correctness and route mappings. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Yimo Jiang <yimoj@nvidia.com> Signed-off-by: Carlos Villela <cvillela@nvidia.com> Co-authored-by: Carlos Villela <cvillela@nvidia.com>
…IA#6358) ## Summary Make the `e2e-live` Vitest project collection gate the single default opt-in boundary and expose canonical repository/CLI paths from the E2E fixture layer. Target-specific Linux, hardware, and destructive/high-cost opt-ins remain intact. ## Related Issue Closes NVIDIA#6350 Parent epic: NVIDIA#6346 ## Changes - Add canonical `REPO_ROOT`, E2E roots, source CLI, and built CLI paths in `test/e2e/fixtures/paths.ts`. - Replace repeated live-test path derivations with those shared exports. - Remove redundant per-file `shouldRunLiveE2E()` checks now enforced by the `e2e-live` project include gate. - Preserve platform and explicit target-specific switches such as Linux-only execution, MCP agent matrices, connect-rlimit coverage, and the issue-4434 live repro. - Add support-test ratchets preventing new local path declarations or duplicate global live gates. - Update the live unit-block scanner examples to demonstrate target-specific rather than global gating. This touches legacy live specs that were not formatted under the current Biome version. The repository's required pre-commit formatter normalized each changed file, so the PR contains mechanical formatting alongside the focused import/gate changes. ## 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) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: test-harness-only refactor with no user-facing behavior - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — 9 E2E support assertions and 7 integration scanner assertions passed - [ ] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) Additional local verification: - `npm run build:cli` - `npm run typecheck` - `npm run lint` - `npx vitest run --project e2e-support test/e2e/support/e2e-paths.test.ts test/e2e/support/e2e-live-target-gating.test.ts test/e2e/support/e2e-live-project-config.test.ts` - `npx vitest run --project integration test/no-unit-blocks-in-live-e2e.test.ts` --- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Tests** * Centralized shared end-to-end repository and CLI entrypoint path constants for live scenarios, and updated affected E2E helpers/tests to use them consistently. * Removed live-project conditional skip wrappers across many scenarios so tests are now registered directly (preserving existing timeouts and core flow). * Updated execution behavior to depend on runtime availability checks (environment/CI/platform/hardware) to determine whether a test proceeds or skips. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Carlos Villela <cvillela@nvidia.com> Co-authored-by: Carlos Villela <cvillela@nvidia.com>
<!-- markdownlint-disable MD041 --> ## Summary Fixes sandbox exec hangs caused by forwarding a live non-terminal stdin that never reaches EOF. NemoClaw now inherits stdin by default only for terminals, closes non-terminal or unavailable stdin, and keeps intentional pipes available through `--stdin`. ## Related Issue Fixes NVIDIA#6319 ## Changes - Merge current `main` without rewriting the contributor's commits or credit. - Add public `--stdin` / `--no-stdin` controls, with explicit flags taking precedence over TTY detection. - Thread `stdin?: boolean` through sandbox exec and select inherited versus ignored stdin in the async production spawner. - Keep the stdio policy and its focused tests in companion modules so the existing exec action/test files stay within growth limits. - Preserve signal forwarding, remote exit codes, cleanup, workdir and multiline guards, and denial-adjacent policy hints. - Update NemoClaw and NemoHermes command references so piped scripts opt in with `--stdin`. - Add parser, default-selection, production-stdio, multiline-guidance, signal, cleanup, exit-code, and policy-hint coverage. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: fresh maintainer review requested on the new semantics; the prior approval predates this head and does not satisfy this gate. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub — GitHub reports all 10 PR commits verified at `d73686641e828025789dedf581099b9f23e4b988`. - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable — `npm run check:diff` passed at `d73686641e828025789dedf581099b9f23e4b988`. - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — the sandbox-exec command/action/policy/cleanup/stdio plus runtime-environment matrix passed 11 files / 160 tests; `npm run build:cli`, CLI typecheck, and `npm run docs:strict` also passed. - [x] Applicable broad gate passed — exact-head required CI run `28829324545` passed (including the successful same-head retry of one unrelated timing flake), and advisor-selected E2E run `28829674160` passed 3/3 jobs. - [ ] Quality Gates section completed with required justifications or waivers — fresh sensitive-path review is pending. - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) — `npm run docs:strict` found 0 errors and two pre-existing hidden Fern warnings. - [x] 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) The exact-head isolated OpenShell 0.0.72 proof `2026-07-06T23-01-11-471Z-pid-7719` passed: a held-open non-TTY FIFO returned promptly by default, `/dev/null` reached EOF, `--stdin` forwarded a finite two-line script, remote exit status `37` propagated, and a parent-only SIGTERM produced NemoClaw exit `143`. OpenShell 0.0.72 has no remote exec cancellation operation, so the harness recorded its pre-existing detached-command behavior separately and removed the uniquely tagged process after identity verification. Both exact-head PR advisors (`28829423941`) returned `merge_as_is` with no blocker or warning. The E2E advisor (`28829422926`) selected `sessions-agents-cli`, `sandbox-operations`, and `shields-config`; all three passed. Fresh human approval remains required because the existing approval predates this head and its stdin semantics. --- Signed-off-by: HwangJohn <angelic805@gmail.com> Signed-off-by: Aaron Erickson <aerickson@nvidia.com> --------- Signed-off-by: HwangJohn <angelic805@gmail.com> Signed-off-by: Aaron Erickson <aerickson@nvidia.com> Co-authored-by: Carlos Villela <cvillela@nvidia.com> Co-authored-by: Aaron Erickson <aerickson@nvidia.com>
<!-- markdownlint-disable MD041 --> ## Summary Reduce cold collection overhead in four rebuild-focused CLI test files by replacing CommonJS source-loader/cache-invalidation seams with native source imports and dependency-light production boundaries. This preserves the existing behavioral coverage while removing incidental loading of the full rebuild graph. ## Related Issue Refs NVIDIA#6245 Refs NVIDIA#6237 ## Changes - Extract messaging-plan staging and config-hash command construction into dependency-light leaf modules while preserving the public rebuild facade exports. - Defer loading `sandbox/rebuild` from `upgrade-sandboxes` until a sandbox actually needs rebuilding, with an explicit dependency seam for focused tests. - Convert four CLI suites from `createRequire`, cache deletion, and loader warmups to native imports and typed spies. - Tighten the exact-path `createRequire` ratchet from 32 to 28 CLI test files. - Preserve all 43 existing assertions in the optimized suites and the real Bash/filesystem, manifest planner, recovery, and gateway-classification contracts. - Add a compiled package-contract test for lazy rebuild loading and facade exports, and make config-hash refresh propagate `sha256sum` failures instead of masking them behind best-effort permission repair. Matched CI evidence: the previous merged head reported 28.806s of aggregate collection time for these four files. Final-head NVIDIA#6388 CI reports 3.041s, a reduction of 25.765s (89.44%; 9.47× faster), while preserving the original assertions and adding the hash-failure regression. Aggregate collection work overlaps across Vitest shards, so this is not a claim of equivalent one-for-one shard-wall savings. ## 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) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: internal module-boundary and test-loader refactor only; commands, flags, defaults, configuration, protocols, and user-visible behavior are unchanged. - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: two independent source reviews found no blocking findings; focused rebuild, recovery, preflight, messaging, config-hash, and compiled package-contract coverage passed. - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — 126/126 focused CLI tests passed across the optimized suites, upgrade preflight, and the broader rebuild flow; 2/2 compiled package-contract tests and the ratchet's 8/8 integration tests also passed. - [x] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — five-shard final-head `CI / Pull Request` and coverage merge passed; all 40 PR checks are green. - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) Additional local verification: - `npm run typecheck:cli` - `npm run build:cli` - compiled package-contract coverage proves importing `dist/lib/actions/upgrade-sandboxes.js` does not eagerly load the rebuild module, forwards the lazy call exactly, and preserves both extracted rebuild facade exports - `npx tsx scripts/checks/test-create-require-budget.ts` (28 CLI files, 8 support files) - `npm run test:projects:check` - required live E2E passed: `rebuild-openclaw`, `rebuild-hermes`, `sandbox-rebuild`, `upgrade-stale-sandbox`, and `channels-add-remove`, and `messaging-providers` (OpenClaw passed on retry after an initial npm `ECONNRESET` during fixture setup) - `git diff --check` --- Signed-off-by: Carlos Villela <cvillela@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added rebuild messaging “recreate contract” planning based on built-in channel manifests. * Introduced a safer sandbox config-hash refresh command with stricter pre-checks before updating the mutable OpenClaw config hash. * **Bug Fixes** * Improved restore/recovery and rebuild messaging preparation, including safer skip conditions when messaging support is unavailable. * Ensured config-hash refresh failures surface correctly, with stronger protections against symlinked/mismatched config inputs. * **Tests** * Updated and expanded package-contract and rebuild flow tests to validate loader laziness and expected rebuild entrypoints. <!-- end of auto-generated comment: release notes by coderabbit.ai -->
<!-- markdownlint-disable MD041 --> ## Summary Adds a CPU-only live E2E that reproduces the fresh Hermes shields lifecycle regression from NVIDIA#6381. The test onboards a new non-root Hermes sandbox, runs shields down/up twice, and preserves the failure as a dedicated E2E lane until NVIDIA#6384 lands. ## Related Issue Relates to NVIDIA#6381. Depends on NVIDIA#6384. ## Changes - Add a fresh Hermes onboard and two-cycle shields down/up live regression test. - Assert the mutable and locked ownership/mode contracts after each transition. - Add a dedicated `hermes-shields-config` workflow job that requires no GPU or hosted inference secret. - Extend the E2E artifact workflow boundary for the new job. ## 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) ## Quality Gates <!-- Check exactly one tests line and one docs line. Check other lines when applicable. Add every requested justification or approval reference. --> - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [ ] Docs updated for user-facing behavior changes - [x] Docs not applicable — justification: test and CI coverage only; no user-facing behavior changes - [ ] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [ ] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification <!-- Check each applicable item only when supported by the requested evidence. Run targeted tests once per relevant change set and rerun after later edits or hook autofixes that can affect the tested behavior. Do not rerun hook-covered checks. --> - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [ ] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — command/result or justification: E2E workflow support tests passed (24/24), and the live target collects successfully; the live run is intentionally expected to reproduce NVIDIA#6381 on current `main` - [ ] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run 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) --- <!-- DCO sign-off is required in this PR description, and every commit must appear as Verified in GitHub. Run: git config user.name && git config user.email --> Signed-off-by: Julie Yaunches <jyaunches@nvidia.com> <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a new Hermes live end-to-end check that runs repeated shields cycles in a fresh non-root sandbox. * Updated PR reporting so the new live job is included in the results summary. * **Bug Fixes** * Updated E2E artifact upload workflow boundary validations to match the current number of expected E2E execution jobs and default callers. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Signed-off-by: Julie Yaunches <jyaunches@nvidia.com>
Probe inference.local from the sandbox for status and doctor. Classify HTTP transport boundaries consistently with connect. Keep provider checks as non-authoritative upstream diagnostics. Co-authored-by: harjoth <harjoth.khara@gmail.com> Co-authored-by: Souvik Ghosh <138186578+souvikDevloper@users.noreply.github.com> Signed-off-by: Apurv Kumaria <akumaria@nvidia.com>
Signed-off-by: Apurv Kumaria <akumaria@nvidia.com>
21 tasks
Collaborator
|
Superseded by #6412 because bringing this stale-base branch onto current main made GitHub expand the PR to an unreviewable historical diff. The original implementation by @harjothkhara is explicitly retained and credited in #6412 through the signed commit Co-authored-by trailer, the PR description, and the contributor-credit comment. Thank you, Harjoth, for the original fix. |
apurvvkumaria
added a commit
that referenced
this pull request
Jul 8, 2026
<!-- markdownlint-disable MD041 --> ## Summary Make the in-sandbox `https://inference.local/v1/models` route authoritative for inference health in `nemoclaw status`, `nemoclaw doctor`, and `nemoclaw connect`. This supersedes #6203 and #6264 while retaining credit for Harjoth's original implementation and Souvik's typed test-seam contribution. ## Related Issue Fixes #6192 ## Changes - Share one agent-aware `inference.local` probe and fail-closed parser across connect, status, and doctor. - Treat final HTTP 200-499 responses as route-reachable; treat interim 100-199 responses, HTTP 500-599, `000`, malformed output, timeout, and unavailable probes as failing. - Centralize route-failure labels and status exit decisions so text, JSON, status, doctor, and connect cannot drift. - Discard route-probe response bodies through `/dev/null` instead of a persistent sandbox temp file. - Verify route TLS with OpenShell's managed CA bundle; certificate failures are authoritative unreachable results. - Keep direct provider checks as explicitly labeled, non-authoritative upstream diagnostics. - Return nonzero status for unhealthy or unavailable inference routes in text and JSON output. - Add regression coverage for HTTP boundaries, unavailable probes, framed OpenShell output, no-direct-probe providers, DCode, and CLI exit behavior. - Update command, monitoring, local-inference, and troubleshooting documentation for the authoritative route semantics. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: [nine-category review passed after the only hardening finding was fixed](#6412 (comment)). - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — `npx vitest run --project cli` over the nine route/connect/status/doctor/recovery/adapter files: 163 passed; `npx vitest run --project integration` over the eight changed CLI/recovery files: 43 passed. - [x] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: `npm run checks` passed; the exact-head PR CI suite completed without failures; [typed OpenClaw and DCode live targets](https://github.com/NVIDIA/NemoClaw/actions/runs/28970519701) passed; [inference-routing, diagnostics, and sandbox-operations live jobs](https://github.com/NVIDIA/NemoClaw/actions/runs/28970503175) passed. - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) — build passed with 0 errors and two existing Fern warnings. - [x] 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) ## Architecture and Runtime-Boundary Review - **Invalid state:** Host/provider checks can succeed while the agent-visible `https://inference.local/v1/models` route is unusable. This PR exposes that disagreement; it does not preserve or mask it. - **Source boundary:** OpenShell owns sandbox exec, DNS, TLS CA injection, and proxy provisioning. NemoClaw owns interpretation and orchestration in `status`, `doctor`, and `connect`, so it probes the exact in-sandbox route and fails closed when OpenShell cannot return trusted `OK` or `BROKEN` evidence. - **Source-fix constraint:** Read-only diagnostics cannot repair OpenShell infrastructure, and an inconclusive result is inherently lossy. `connect` must not open SSH or mutate route state on inconclusive evidence. This fail-closed behavior is a permanent security boundary, not a temporary workaround. - **Regression and removal contract:** Checked-in parser, flow, CLI, missing-CA, 401/403, DCode argv/proxy, and redaction tests cover deterministic fault injection. The DCode wrapper may be removed only if OpenShell provides an agent-independent structured route-health API that preserves its CA and login-shell contract; fail-closed handling remains. - **Runtime justification:** A separate standalone live artifact would duplicate the repository's existing typed E2E lanes. Exact head `130e74f3c520aa82db86fd80e830eff362390b08` passed the [required live jobs](https://github.com/NVIDIA/NemoClaw/actions/runs/28970503175) and [OpenClaw and DCode live targets](https://github.com/NVIDIA/NemoClaw/actions/runs/28970519701); exact-head macOS E2E and arm64 image/build checks also passed. Induced broken-route, auth, and CA faults remain deterministic checked-in tests to avoid destructive shared-state mutation. - **PR sequencing:** #6412 is foundational. Merge it first, then rebase #6465 and integrate its additive route-drift warning into the final status-snapshot shape. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added clearer, authoritative inference health checks and diagnostics for sandbox status and doctor commands. * `connect` now fails closed when the in-sandbox inference route cannot be trusted, with safer redacted error details. * **Bug Fixes** * Improved handling of reachable, unhealthy, unreachable, and not-probed inference states. * Fixed status exit behavior so inference-route failures correctly return a non-zero result. * **Documentation** * Updated troubleshooting and command docs to explain the new inference health and proxy diagnostic output. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --- Signed-off-by: Apurv Kumaria <akumaria@nvidia.com> Co-authored-by: harjoth <harjoth.khara@gmail.com> Co-authored-by: Souvik Ghosh <138186578+souvikDevloper@users.noreply.github.com> --------- Signed-off-by: Apurv Kumaria <akumaria@nvidia.com> Signed-off-by: cjagwani <cjagwani@nvidia.com> Co-authored-by: harjoth <harjoth.khara@gmail.com> Co-authored-by: Souvik Ghosh <138186578+souvikDevloper@users.noreply.github.com> Co-authored-by: cjagwani <cjagwani@nvidia.com>
Hadar301
pushed a commit
to Hadar301/NemoClaw-OpenShift
that referenced
this pull request
Jul 12, 2026
<!-- markdownlint-disable MD041 --> ## Summary Make the in-sandbox `https://inference.local/v1/models` route authoritative for inference health in `nemoclaw status`, `nemoclaw doctor`, and `nemoclaw connect`. This supersedes NVIDIA#6203 and NVIDIA#6264 while retaining credit for Harjoth's original implementation and Souvik's typed test-seam contribution. ## Related Issue Fixes NVIDIA#6192 ## Changes - Share one agent-aware `inference.local` probe and fail-closed parser across connect, status, and doctor. - Treat final HTTP 200-499 responses as route-reachable; treat interim 100-199 responses, HTTP 500-599, `000`, malformed output, timeout, and unavailable probes as failing. - Centralize route-failure labels and status exit decisions so text, JSON, status, doctor, and connect cannot drift. - Discard route-probe response bodies through `/dev/null` instead of a persistent sandbox temp file. - Verify route TLS with OpenShell's managed CA bundle; certificate failures are authoritative unreachable results. - Keep direct provider checks as explicitly labeled, non-authoritative upstream diagnostics. - Return nonzero status for unhealthy or unavailable inference routes in text and JSON output. - Add regression coverage for HTTP boundaries, unavailable probes, framed OpenShell output, no-direct-probe providers, DCode, and CLI exit behavior. - Update command, monitoring, local-inference, and troubleshooting documentation for the authoritative route semantics. ## Type of Change - [ ] Code change (feature, bug fix, or refactor) - [x] Code change with doc updates - [ ] Doc only (prose changes, no code sample modifications) - [ ] Doc only (includes code sample changes) ## Quality Gates - [x] Tests added or updated for changed behavior - [ ] Existing tests cover changed behavior — justification: - [ ] Tests not applicable — justification: - [x] Docs updated for user-facing behavior changes - [ ] Docs not applicable — justification: - [x] Sensitive paths changed (security, policy, credentials, preflight, onboarding, inference, runner, sandbox, or messaging) - [x] Sensitive-path review completed or maintainer-approved waiver recorded — reviewer/approval link/justification: [nine-category review passed after the only hardening finding was fixed](NVIDIA#6412 (comment)). - [ ] Non-success, skipped, or missing CI check accepted by maintainer — check name, approval link, and follow-up issue: ## Verification - [x] PR description includes the DCO sign-off declaration and every commit appears as `Verified` in GitHub - [x] Normal `pre-commit`, `commit-msg`, and `pre-push` hooks passed, or `npm run check:diff` passed when hooks were skipped or unavailable - [x] Targeted behavior tests pass for the current change set, or tests are marked not applicable above — `npx vitest run --project cli` over the nine route/connect/status/doctor/recovery/adapter files: 163 passed; `npx vitest run --project integration` over the eight changed CLI/recovery files: 43 passed. - [x] Applicable broad gate passed — `npm test` for broad runtime/test-harness changes; `npm run check` for repo-wide validation/coverage changes — command/result: `npm run checks` passed; the exact-head PR CI suite completed without failures; [typed OpenClaw and DCode live targets](https://github.com/NVIDIA/NemoClaw/actions/runs/28970519701) passed; [inference-routing, diagnostics, and sandbox-operations live jobs](https://github.com/NVIDIA/NemoClaw/actions/runs/28970503175) passed. - [x] Quality Gates section completed with required justifications or waivers - [x] No secrets, API keys, or credentials committed - [ ] `npm run docs` builds without warnings (doc changes only) — build passed with 0 errors and two existing Fern warnings. - [x] 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) ## Architecture and Runtime-Boundary Review - **Invalid state:** Host/provider checks can succeed while the agent-visible `https://inference.local/v1/models` route is unusable. This PR exposes that disagreement; it does not preserve or mask it. - **Source boundary:** OpenShell owns sandbox exec, DNS, TLS CA injection, and proxy provisioning. NemoClaw owns interpretation and orchestration in `status`, `doctor`, and `connect`, so it probes the exact in-sandbox route and fails closed when OpenShell cannot return trusted `OK` or `BROKEN` evidence. - **Source-fix constraint:** Read-only diagnostics cannot repair OpenShell infrastructure, and an inconclusive result is inherently lossy. `connect` must not open SSH or mutate route state on inconclusive evidence. This fail-closed behavior is a permanent security boundary, not a temporary workaround. - **Regression and removal contract:** Checked-in parser, flow, CLI, missing-CA, 401/403, DCode argv/proxy, and redaction tests cover deterministic fault injection. The DCode wrapper may be removed only if OpenShell provides an agent-independent structured route-health API that preserves its CA and login-shell contract; fail-closed handling remains. - **Runtime justification:** A separate standalone live artifact would duplicate the repository's existing typed E2E lanes. Exact head `130e74f3c520aa82db86fd80e830eff362390b08` passed the [required live jobs](https://github.com/NVIDIA/NemoClaw/actions/runs/28970503175) and [OpenClaw and DCode live targets](https://github.com/NVIDIA/NemoClaw/actions/runs/28970519701); exact-head macOS E2E and arm64 image/build checks also passed. Induced broken-route, auth, and CA faults remain deterministic checked-in tests to avoid destructive shared-state mutation. - **PR sequencing:** NVIDIA#6412 is foundational. Merge it first, then rebase NVIDIA#6465 and integrate its additive route-drift warning into the final status-snapshot shape. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added clearer, authoritative inference health checks and diagnostics for sandbox status and doctor commands. * `connect` now fails closed when the in-sandbox inference route cannot be trusted, with safer redacted error details. * **Bug Fixes** * Improved handling of reachable, unhealthy, unreachable, and not-probed inference states. * Fixed status exit behavior so inference-route failures correctly return a non-zero result. * **Documentation** * Updated troubleshooting and command docs to explain the new inference health and proxy diagnostic output. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --- Signed-off-by: Apurv Kumaria <akumaria@nvidia.com> Co-authored-by: harjoth <harjoth.khara@gmail.com> Co-authored-by: Souvik Ghosh <138186578+souvikDevloper@users.noreply.github.com> --------- Signed-off-by: Apurv Kumaria <akumaria@nvidia.com> Signed-off-by: cjagwani <cjagwani@nvidia.com> Co-authored-by: harjoth <harjoth.khara@gmail.com> Co-authored-by: Souvik Ghosh <138186578+souvikDevloper@users.noreply.github.com> Co-authored-by: cjagwani <cjagwani@nvidia.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
20 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
nemoclaw <sbx> statusanddoctorreported inference healthy for cloud/managed providers by probing only the upstream provider endpoint, never theinference.localroute the agent actually uses. Wheninference.localwas broken inside the sandbox they still reported "healthy" (exit 0), contradictingnemoclaw <sbx> connect(which probes the real route). This widens the existinginference.localgateway-chain probe — added for local providers in #3265 — to run for all providers.Related Issue
Refs #6192
Changes
doctor.ts: drop the local-only gate incollectInferenceSubprobes(and the now-unused file-localisLocalInferenceProviderhelper /providerparam) so theinference.localgateway chain is probed for every provider. A broken route surfaces as[fail] Provider health (gateway)and flipsdoctortofail(exit 1).status-snapshot.ts: drop theollama-local/vllm-localclause on the snapshot gateway probe sostatussurfaces the same broken route.doctor-flow.test.ts: parameterized the harness (provider,gatewayChainOk; defaults preserve existing tests); added cloud-provider red→green regression coverage.Type of Change
Verification
Verifiedin GitHubdoctor,status-flow,status,process-recovery,connect-flow— 162 tests)#6192cases; verified they fail onmain)npm run typecheck:cliand biome passReal behavior proof
status/doctorreported inference "healthy" for cloud providers off the upstream probe while the realinference.localroute was broken (false all-clear, exit 0).runSandboxDoctorproduction path; only the two external boundaries stubbed — upstreamprobeProviderHealth(reachable) and the in-sandboxinference.localcurlprobeSandboxInferenceGatewayHealth(broken). No live GPU sandbox required.runSandboxDoctor("alpha", …)forprovider=nvidia-prod, upstream healthy,inference.localreturning HTTP000.main: no gateway line emitted,Summary: healthy, exit0.status/doctornow agree withconnect— a broken in-sandbox route reports broken. Regression is locked by twodoctor-flowtests that fail onmainand pass with the fix.doctor;statusshares the identicalcollectSandboxStatusSnapshotseam and the same subprobe path, covered by existingstatus-flowtests. Thestatus --jsonexit code keys offfailureLayer(notinferenceHealth.ok) — left unchanged and noted as a follow-up. No live cloud sandbox / realopenshell sandbox execroundtrip.Signed-off-by: harjoth harjoth.khara@gmail.com
🤖 Generated with Claude Code