…4628 review)

Address @cv's PR NVIDIA#4628 review feedback: the 700/600 heal was too generic.
If a future caller routes a mutable-sandbox OpenClaw config dir (which
uses the 2770/660 contract per NVIDIA#4538 / PR NVIDIA#4610) through
`ensureConfigDir`, the walker would silently normalize it to 700/600 —
exactly the NVIDIA#4538 EACCES regression.

Add three predicates:
- `hostNemoclawDir()` — canonical `${HOME}/.nemoclaw`
- `isHostNemoclawRoot(dirPath)` — gate for the sibling sweep
- `isMutableSandboxConfigPath(targetPath)` — gate for the file-level heal

Wire them through:
- `ensureConfigDir`: skip dir-mode chmod when path is mutable-sandbox.
  Run `healRootLevelFiles` only when path IS the host nemoclaw root.
- `readConfigFile`: skip per-file chmod when path is mutable-sandbox.

Tests:
- 2 new scope-boundary tests (negative: arbitrary dir, no heal;
  positive: host nemoclaw root, heal fires).
- Existing heal tests (multi-sibling + symlink positive control)
  updated to use the new `withHome` helper so they place files under
  `<HOME>/.nemoclaw` instead of an arbitrary tmp dir — preserves the
  NVIDIA#4546 acceptance while respecting the new scope boundary.

18/18 tests pass.

Co-authored-by: kagura-agent <kagura.agent.ai@gmail.com>
Signed-off-by: Charan Jagwani <cjagwani@nvidia.com>

## Summary
- Adds the `v0.0.60` section to `docs/about/release-notes.mdx` using the
dev announcement from discussion #4877.
- Fills the source-doc gaps found during release-prep review across
inference, policy tiers, command behavior, security boundaries, Hermes
dashboard/tooling, runtime context, and troubleshooting.
- Refreshes generated agent skills under `.agents/skills/` from the
current Fern docs output and upgrades Fern from `5.44.3` to `5.45.0`.

## Source summary
- #4037 -> `docs/reference/architecture.mdx`,
`docs/about/how-it-works.mdx`, `docs/about/release-notes.mdx`: Documents
system-only runtime context that stays out of visible chat.
- #4875 -> `docs/reference/architecture.mdx`,
`docs/about/how-it-works.mdx`, `docs/about/release-notes.mdx`: Documents
try-first sandbox network/filesystem guidance and clearer failure
classification.
- #4788 -> `docs/security/best-practices.mdx`,
`docs/about/release-notes.mdx`: Documents shared OpenClaw
device-approval policy for startup and connect.
- #4768 -> `docs/reference/network-policies.mdx`,
`docs/network-policy/integration-policy-examples.mdx`,
`docs/get-started/quickstart.mdx`,
`docs/get-started/quickstart-hermes.mdx`, `docs/reference/commands.mdx`:
Documents `weather`, `public-reference`, and Hermes managed-tool gateway
preset behavior.
- #3788 and #4864 -> `docs/reference/network-policies.mdx`,
`docs/reference/commands.mdx`: Documents non-interactive policy-tier
fail-fast behavior and interactive prompt fallback.
- #4756 and #4866 -> `docs/reference/commands.mdx`: Documents env-aware
default sandbox resolution for `list`, `status`, and `tunnel` commands.
- #4320 -> `docs/reference/commands.mdx`: Documents `$$nemoclaw tunnel
status` behavior.
- #4328 -> `docs/reference/commands.mdx`: Documents line-scoped policy
preset descriptions in `policy-list`.
- #4580 and #4748 -> `docs/reference/architecture.mdx`: Documents
package-managed OpenShell gateway service and Docker-driver
gateway-marker behavior.
- #4598 -> `docs/manage-sandboxes/lifecycle.mdx`: Documents concurrent
gateway/dashboard cleanup isolation by sandbox name and port.
- #4777 -> `docs/reference/troubleshooting.mdx`: Documents Docker GPU
patch rollback behavior.
- #4610 -> `docs/reference/troubleshooting.mdx`,
`docs/reference/commands.mdx`: Keeps mutable OpenClaw config permission
guidance aligned and removes skipped experimental wording.
- #4868 -> `docs/reference/commands.mdx`: Keeps `.dockerignore` handling
for custom `onboard --from <Dockerfile>` contexts in generated skills.
- #4870 -> `docs/reference/commands.mdx`,
`docs/manage-sandboxes/runtime-controls.mdx`: Documents
`NEMOCLAW_MINIMAL_BOOTSTRAP` and generated skill coverage.
- #4641 -> `docs/inference/inference-options.mdx`,
`docs/reference/troubleshooting.mdx`: Documents local NVIDIA NIM
platform-digest pulls and served-model id adoption.
- #4810 and #4867 -> `docs/inference/inference-options.mdx`: Documents
stable NGC managed-vLLM image lineage and DGX Station DeepSeek V4 Flash
coverage.
- #4852 -> `docs/inference/use-local-inference.mdx`,
`docs/reference/troubleshooting.mdx`: Documents Ollama model fit
filtering, 16K context floor, cold-load retry, and failed-model
exclusion.
- #4847 -> `docs/inference/switch-inference-providers.mdx`: Documents
API-family sync, Hermes `api_mode`, and Bedrock Runtime exception.
- #4800 -> `docs/inference/tool-calling-reliability.mdx`: Documents
Nemotron managed-inference native tool-search fallback.
- #4333 -> `docs/inference/switch-inference-providers.mdx`: Documents
interactive multimodal input prompting.
- #4086 -> `docs/reference/troubleshooting.mdx`: Keeps proxy bypass
normalization in generated troubleshooting coverage.
- #4811 and #4855 -> `docs/get-started/quickstart-hermes.mdx`: Documents
prebuilt Hermes dashboard assets and TUI recovery without runtime
rebuilds.
- #4854 -> `docs/inference/switch-inference-providers.mdx`,
`docs/reference/commands.mdx`: Documents Hermes proxy API-key
placeholder preservation during inference switches.
- #4248 -> `docs/manage-sandboxes/messaging-channels.mdx`,
`.agents/skills/`: Keeps messaging enrollment behavior aligned with
manifest-hook implementation.
- #4771 -> `docs/security/best-practices.mdx`,
`docs/security/credential-storage.mdx`: Documents Hermes
placeholder-only secret boundary for sandbox-visible runtime files.
- #4787 -> `docs/security/best-practices.mdx`,
`docs/about/release-notes.mdx`: Documents expanded memory scanner
examples for OpenAI project keys and Slack app-level tokens.
- #4848 -> `docs/reference/commands.mdx`: Documents OpenClaw skill
install mirroring into the agent home directory.
- #4790 -> `docs/about/release-notes.mdx`: Uses the prior release-prep
structure and generated `.agents/skills/` refresh as the template for
this release.

## Verification
- `python3 scripts/docs-to-skills.py docs/ .agents/skills/ --prefix
nemoclaw-user --doc-platform fern-mdx`
- `python3 scripts/docs-to-skills.py docs/ .agents/skills/ skills/
--prefix nemoclaw-user --doc-platform fern-mdx --dry-run`
- `npm run docs`
- `git diff --check`
- skip-term scan across `docs/`, `.agents/skills/`, and `skills/`
- `npm run build:cli`
- `npm run typecheck:cli`
- Commit and pre-push hook suites, including markdownlint, gitleaks,
env-var docs gate, docs-to-skills verification, and skills YAML tests

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

## Summary by CodeRabbit

## Release Notes

* **New Features**
* DeepSeek-V4-Flash now available as default inference model for DGX
Station.
* Hermes dashboard improved with dedicated port and OAuth-authenticated
tool gateway selection.
* Added weather and public-reference policy presets for expanded agent
capabilities.
* Enhanced Ollama model selection with GPU memory filtering and
automatic retry for timeouts.

* **Bug Fixes**
  * Improved policy tier validation to prevent invalid configurations.
* Better sandbox cleanup scoping by port to prevent conflicts across
deployments.
  * Added GPU patch failure recovery with automatic rollback.

* **Documentation**
* Expanded troubleshooting guides for inference, security, and sandbox
lifecycle.
  * Added .dockerignore best practices for custom deployments.

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

---------

Co-authored-by: Carlos Villela <cvillela@nvidia.com>

…box doctor --fix (NVIDIA#4538)

A raw connect-shell `openclaw doctor --fix` enforces OpenClaw's
single-user 700/600 layout, tightening /sandbox/.openclaw and
openclaw.json back from the NemoClaw mutable contract (2770 setgid +
group-writable dir, 660 config). That blocks the gateway UID — a member
of the sandbox group — from persisting config. PR NVIDIA#4610 only repaired
this via host-side NemoClaw wrapper commands (`nemoclaw doctor --fix`,
rebuild, startup); none run after the raw in-sandbox path the QA
reporter used, so the gateway stayed broken until the next restart.

Fix the missed path at the always-on in-sandbox guard: the emitted
openclaw() shell function (sourced by every interactive/login sandbox
shell) now re-asserts the contract after every routed openclaw command,
regardless of exit code.

- add _nemoclaw_restore_mutable_config_perms to the guard env; skips
  when shields are up (config dir root-owned), no-op via fast path when
  the contract already holds, and strips group-write from the read-only
  recovery baseline the recursive chmod could otherwise loosen
- drop errexit around the guarded call (mirroring the devices-approve
  branch) so a nonzero `doctor --fix` exit cannot abort the guard before
  the restore runs — the nonzero-exit case is the exact reported failure
- regression tests plus a gated real-image reporter-workflow E2E that
  drives raw `openclaw doctor --fix` and asserts 2770/660 is restored

Signed-off-by: Yimo Jiang <yimoj@nvidia.com>

…box doctor --fix (#4538) (#5017)

## Summary

A raw connect-shell `openclaw doctor --fix` (run directly inside a
mutable sandbox, outside any NemoClaw wrapper) enforces OpenClaw's
single-user 700/600 layout, tightening `/sandbox/.openclaw` and
`openclaw.json` back from the NemoClaw mutable contract (2770 setgid +
group-writable dir, 660 config). That blocks the gateway UID — a member
of the `sandbox` group — from persisting config. This fixes the missed
path at the always-on in-sandbox `openclaw()` guard so the contract is
restored after the raw command, even when `doctor` exits nonzero.

## Related Issue

Fixes #4538

## Changes

- **Always-on in-sandbox restore.** The emitted `openclaw()` guard
function (written into `/tmp/nemoclaw-proxy-env.sh`, sourced by every
interactive/login sandbox shell) now calls a new
`_nemoclaw_restore_mutable_config_perms` helper after every routed
openclaw command, re-asserting `2770` dir / `660` config (+
`.config-hash`). PR #4610 only repaired this via host-side wrapper
commands (`nemoclaw doctor --fix`, rebuild, startup), none of which run
after the raw in-sandbox path the reporter used.
- **Safe by construction.** The helper skips when shields are up (config
dir root-owned, so the lock is never weakened), is a no-op fast path
when the contract already holds, and strips group-write from the
read-only recovery baseline that the recursive `chmod` could otherwise
loosen in rootless mode.
- **Survives nonzero exit.** Errexit is dropped around the guarded call
(mirroring the existing `devices approve` branch) so a nonzero `doctor
--fix` exit — the exact reported failure (EACCES on a root-locked shell
init file) — cannot abort the guard before the restore runs.
- **Tests.** Regression tests for the helper (restore, shields-up no-op,
errexit survival, baseline protection) and a real-image
reporter-workflow E2E (`test/repro-4538-raw-doctor-perms.test.ts`) that
sources the actual emitted guard, runs raw `openclaw doctor --fix` in a
NemoClaw sandbox image, and asserts the contract is restored. The E2E is
gated behind `NEMOCLAW_RUN_DOCTOR_PERMS_DOCKER_E2E=1` so it runs where
docker + the image are available.

## Type of Change

- [x] Code change (feature, bug fix, or refactor)

## Verification

- [x] `npm test` passes locally (the only failures were 14 pre-existing,
load-induced 5s-timeout flakes in unrelated CLI `--help` oclif-dispatch
tests, e.g. `test/cli/snapshot-shields.test.ts`; they do not touch this
change). All CI checks are green: ShellCheck, static-checks, 5
cli-test-shards, plugin-tests, build-typecheck, CodeQL, macos/wsl-e2e,
dco-check, codebase-growth-guardrails.
- [x] Tests added or updated for new or changed behavior
- [x] No secrets, API keys, or credentials committed

### Reporter-workflow E2E evidence

The exact reported workflow — connect to a mutable sandbox, source the
always-on guard (as `/etc/bash.bashrc` does on connect), run the **raw
in-sandbox `openclaw doctor --fix`** — was reproduced end-to-end against
a real NemoClaw sandbox image (`nemoclaw-production:latest`), as the
`sandbox` user.

- **Test:** `test/repro-4538-raw-doctor-perms.test.ts`
- **Env / command:** `NEMOCLAW_RUN_DOCTOR_PERMS_DOCKER_E2E=1 npx vitest
run test/repro-4538-raw-doctor-perms.test.ts`

```text
=== BEFORE (baseline mutable contract) ===
/sandbox/.openclaw           2770 sandbox:sandbox
/sandbox/.openclaw/openclaw.json  660 sandbox:sandbox
=== reporter cmd: raw `openclaw doctor --fix` ===
(without the fix this leaves: dir 700 sandbox:sandbox, openclaw.json 600 sandbox:sandbox)
=== AFTER (with the always-on openclaw() guard) ===
/sandbox/.openclaw           2770 sandbox:sandbox
/sandbox/.openclaw/openclaw.json  660 sandbox:sandbox
RESULT dir=2770 file=660
PASS: mutable contract (2770/660) restored after raw openclaw doctor --fix
```

The before-fix regression (dir→700, file→600) was also reproduced
directly in the same image without the guard sourced, confirming the
gap. Additional non-docker regression tests in the same file cover the
restore helper, shields-up no-op, errexit survival on a nonzero `doctor
--fix`, and recovery-baseline protection.

---
Signed-off-by: Yimo Jiang <yimoj@nvidia.com>

Signed-off-by: Yimo Jiang <yimoj@nvidia.com>

## Summary

The "Selective E2E Results" comment posted by
`.github/workflows/nightly-e2e.yaml` bucketed job results into `passed`
/ `failed` / `skipped` but never accounted for `cancelled`. A job ending
in `cancelled` (e.g. when `cancel-in-progress` kills a stale run)
slipped past all three tallies and fell through to the default `"✅ All
requested jobs passed"` status, with the summary line reading `"0
passed, 0 failed, 0 skipped"` — masking that the run produced no signal
at all.

## Repro (in the wild)

PR #5085 commit `026802ba8`:

```text
### Selective E2E Results — ✅ All requested jobs passed
Run: 27305033204
Requested jobs: agent-turn-latency-e2e,cloud-inference-e2e
Summary: 0 passed, 0 failed, 0 skipped

| Job                    | Result        |
|------------------------|---------------|
| agent-turn-latency-e2e | ⚠️ cancelled |
| cloud-inference-e2e    | ⚠️ cancelled |
```

Both requested jobs were cancelled (cancel-in-progress superseding the
older run) yet the headline read green. Same pattern earlier on #4610.

## Root

`.github/workflows/nightly-e2e.yaml:2486-2495` (pre-fix):

```js
const passed  = ran.filter(([, v]) => v.result === 'success');
const failed  = ran.filter(([, v]) => v.result === 'failure');
const skipped = reportedEntries.filter(([, v]) => v.result === 'skipped');

const status =
  failed.length > 0 || missingRequested.length > 0 ? '❌ Some jobs failed'
  : skipped.length > 0 && passed.length === 0    ? '⚠️ No requested jobs ran'
  :                                                  '✅ All requested jobs passed';
```

A `cancelled` job is `!== 'success'`, `!== 'failure'`, `!== 'skipped'` —
falls through to the default.

## Changes

- Add a `cancelled` bucket derived from `ran` the same way the others
are.
- Insert a status branch between the failure case and the no-ran case:
when cancelled jobs are present and nothing passed, surface `⚠️ Run
cancelled — no signal` instead of falsely claiming success.
- Include cancelled count in the summary line so the bucket is visible
even when the other states are zero.

Successful, failed, and skipped runs continue to render exactly as
before — only the cancelled case changes from "false green" to "honest
yellow."

cc @jyaunches (flagged this earlier in slack)

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

## Summary by CodeRabbit

* **Tests**
* Improved nightly e2e test reporting in PR comments to better
distinguish cancelled jobs from other outcomes. PR comments now display
cancelled job counts and provide clearer status messaging when test runs
are cancelled.

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

Signed-off-by: Charan Jagwani <cjagwani@nvidia.com>