NemoClaw v0.0.63 is out

[jyaunches]

NemoClaw v0.0.63 is out

NemoClaw v0.0.63 is a stability-focused patch release that tightens the boundary between preserved user state, sandbox recovery, and runtime safety. The release improves how NemoClaw rebuilds stale sandboxes, restores OpenClaw configuration, recovers Docker-driver sandboxes after host reboots, and fails closed when a restored config would require unsafe fallback behavior.

This release also makes the setup path more resilient. Onboarding now diagnoses host DNS blocks earlier, recovers from Ollama auth-proxy port conflicts, reports unexpected Docker access more clearly, and adds an interactive managed-vLLM model picker for local inference workflows.

Under the hood, v0.0.63 continues the migration of onboarding and E2E infrastructure toward typed, testable slices. The result is a patch release with a broad set of recovery fixes, messaging safeguards, Hermes runtime hardening, documentation updates, and CI confidence improvements.

Sandbox, rebuild, and OpenClaw recovery

• #5034 recovers stale sandboxes during rebuild --yes instead of dead-ending when a sandbox is registered locally but absent from a healthy OpenShell gateway. This keeps rebuild retryable and preserves local registry metadata through recovery.
• #5091 recovers Docker-driver sandboxes from container labels after a host reboot. This improves lifecycle continuity when the local registry and Docker state drift.
• #5101 preserves OpenClaw openclaw.json settings across rebuilds. User runtime preferences now survive rebuild paths more consistently.
• #5174 merges restored OpenClaw config with runtime state. This avoids clobbering live state while still carrying forward preserved configuration.
• #5177 fails unsafe OpenClaw config-restore fallbacks. NemoClaw now stops rather than silently applying a restore path that could weaken config safety.
• #5102 detects NemoClaw image drift in upgrade-sandboxes. Operators get clearer upgrade behavior when sandbox images no longer match the expected release surface.

Onboarding, inference, and host diagnostics

• #4201 reports unexpected Docker access during installer checks. This makes preflight failures easier to understand when Docker is reachable in an unanticipated way.
• #5038 adds an interactive managed-vLLM model picker. Users can choose managed vLLM models through the guided inference flow instead of wiring every detail manually.
• #5040 diagnoses Ollama auth-proxy port conflicts and recovers startup. This makes local Ollama-backed onboarding less brittle when ports are already occupied.
• #5041 catches host DNS blocks before provider validation. NemoClaw now identifies an environmental network problem earlier, before it is confused with provider misconfiguration.
• #5109 adds helper-level Vitest coverage for Docker-unreachable gateway-start abort behavior. The regression path is now pinned closer to the unit boundary.
• #5129 makes cron preflight inference.local use trusted env-proxy mode. Scheduled checks now align better with the runtime environment they validate.

Messaging, Hermes, and plugin activation

• #4981 enforces the runtime environment secret guard on direct hermes gateway invocation. Hermes startup now keeps secret handling protections even when launched through the direct gateway path.
• #4986 clarifies Slack token validation and the Slack policy preset in the docs. The messaging setup path is now easier to audit and troubleshoot.
• #5036 adds tests that guard Slack channels.* allowlist synchronization. The allowlist behavior that protects Slack channel access now has regression coverage.
• #5039 warns or blocks a second Slack Socket Mode sandbox on a shared gateway. This prevents confusing duplicate Socket Mode sessions competing for the same messaging surface.
• #5124 forwards shields sub-arguments and surfaces host-only guidance from the plugin. Shield commands now behave more predictably when invoked through plugin routing.
• #5168 detects wrapped Hermes gateway arguments. Hermes launch handling is now more robust when gateway invocation is wrapped by another command layer.

Onboarding internals and E2E infrastructure

• #4507 runs final onboarding phases through an FSM slice. This continues the move toward explicit, testable onboarding state transitions.
• #4530 extracts a live FSM slice runner helper. Shared onboarding execution logic is now easier to test and reuse.
• #4531 covers live FSM slice boundaries. The newly extracted onboarding flow has targeted regression coverage.
• #5052 extends migration inventory to scenario-runner internals. This keeps E2E migration work visible as legacy scripts are retired.
• #5056 adds inference runtime helpers for E2E tests. Scenario tests can now exercise inference runtime behavior with less shell glue.
• #5105 routes the scenario advisor to the Vitest workflow. CI recommendations now line up better with the active E2E execution path.
• #5106 completes the typed-shell-runner cutover from parity to retirement. This reduces legacy shell dependence in E2E coverage.
• #5107 migrates the OpenShell version-pin test to a free-standing Vitest live test. Version-pin coverage now lives in the newer scenario framework.
• #5111 covers rebuild resume snapshot handoff. Rebuild resume behavior now has more direct regression protection.
• #5112 names the resume repair policy. The onboarding recovery path is easier to reason about and maintain.
• #5119, #5126, #5153, and #5155 retire or simplify legacy E2E migration scripts. The test suite keeps moving toward typed, maintainable scenario coverage.
• #5121 extracts policy selection prompts. Onboarding policy selection becomes more modular and easier to test.

Docs and release surface

• #5037 documents the Control UI config endpoint and authentication behavior. Operators now have clearer troubleshooting guidance for the UI configuration path.
• #5157 refreshes the v0.0.62 release docs. Release documentation stays aligned with the previous published patch while this release advances the runtime surface.

Thank you

Thank you to external contributor @abhi-0906 for the Hermes runtime secret-guard fix in #4981 and the Slack documentation clarification in #4986.