• **Runtime validation** — Validate `setConfigValue` still creates missing array/object path segments and still rejects `__proto__`, `constructor`, and `prototype` segments.. The refactor appears behavior-preserving and unit-level changes are not required for the mechanical inlining itself, but several touched files are runtime/sandbox/inference paths where targeted validation improves confidence.
• **Runtime validation** — Validate snapshot restore image resolution still returns the registered image for gateway-metadata drivers and the kubectl-discovered image for kubernetes fallback.. The refactor appears behavior-preserving and unit-level changes are not required for the mechanical inlining itself, but several touched files are runtime/sandbox/inference paths where targeted validation improves confidence.
• **Runtime validation** — Validate messaging manifest compilation still trims config env values and reports secret input availability without exposing secret values.. The refactor appears behavior-preserving and unit-level changes are not required for the mechanical inlining itself, but several touched files are runtime/sandbox/inference paths where targeted validation improves confidence.
• **Runtime validation** — Validate tar listing still rejects listings over `TAR_LISTING_MAX_OUTPUT_BYTES` and removes the temporary listing directory after success and failure.. The refactor appears behavior-preserving and unit-level changes are not required for the mechanical inlining itself, but several touched files are runtime/sandbox/inference paths where targeted validation improves confidence.
• **Runtime validation** — Validate sandbox session parsing still matches only exact `openshell-<sandbox>` SSH host names and not prefix matches.. The refactor appears behavior-preserving and unit-level changes are not required for the mechanical inlining itself, but several touched files are runtime/sandbox/inference paths where targeted validation improves confidence.
• **Acceptance clause:** Tests and type-checks relevant to touched files pass. — add test evidence or identify existing coverage. The PR body reports `npx prek run --all-files` and `npm test` passing, but this review did not execute tests or evaluate external CI status.
• **Acceptance clause:** Add a TypeScript-aware filter, or equivalent review script, that only reports variables with exactly one local reference. — add test evidence or identify existing coverage. The PR body claims candidates were generated locally and manually filtered, but no helper script is committed and the exact filter cannot be independently verified from repository contents.
• **Acceptance clause:** Manually review candidates before editing. — add test evidence or identify existing coverage. The PR body states candidates were manually filtered, but manual review process evidence is outside the repository diff.

E2E Recommendation Advisor

Base: origin/main
Head: HEAD
Confidence: high

Required E2E

• None. No merge-blocking E2E is recommended: every shown change is a behavior-preserving local-variable inline/removal in sensitive runtime files, with no logic, control-flow, IO, validation, command, policy, or configuration semantics changed. Optional jobs are listed only for extra confidence in adjacent high-risk domains.

Optional E2E

• snapshot-commands-e2e (medium): Optional regression confidence for snapshot create/list/restore and credential absence because snapshot, migration-state, and tar-listing files were touched, even though the diff is behavior-preserving.
• inference-routing-e2e (medium): Optional confidence for inference route error classification and credential isolation after no-op refactors in local inference diagnostics and onboard probe helpers.
• messaging-providers-e2e (high): Optional coverage for messaging provider credential placeholder/L7-proxy flows because manifest input/env normalization code was touched.
• sandbox-operations-e2e (medium): Optional lifecycle confidence for sandbox operations and destroy cleanup paths after a no-op refactor in sandbox destroy/session-related code.
• shields-config-e2e (medium): Optional check for shields state/timer lifecycle because shields timer and destroy-artifact cleanup code was touched.
• network-policy-e2e (medium): Optional confidence for policy parsing/enforcement because blueprint policy validation code was touched, despite the shown change being an equivalent refactor.

New E2E recommendations

• None.

E2E Scenario Advisor

Base: origin/main
Head: HEAD
Confidence: medium

Required scenario E2E

• ubuntu-repo-cloud-openclaw-custom-policies: Best single Ubuntu scenario for the changed OpenClaw source paths that affect onboarding, custom policy/blueprint handling, inference routing, sandbox snapshot create/list/restore, state/tar listing, and normal sandbox teardown behavior.
  • Dispatch: gh workflow run e2e-scenarios.yaml --ref <pr-head-ref> --field scenarios=ubuntu-repo-cloud-openclaw-custom-policies
• ubuntu-rebuild-openclaw: Exercises the migration-state snapshot bundle and rebuild lifecycle path touched by nemoclaw/src/commands/migration-state.ts, plus post-rebuild inference and policy preservation checks.
  • Dispatch: gh workflow run e2e-scenarios.yaml --ref <pr-head-ref> --field scenarios=ubuntu-rebuild-openclaw
• ubuntu-repo-cloud-openclaw-telegram: Minimal routed messaging scenario to cover the manifest compiler input/env normalization path changed in src/lib/messaging/compiler/manifest-compiler.ts, including common messaging provider attachment and no-secret-leak checks.
  • Dispatch: gh workflow run e2e-scenarios.yaml --ref <pr-head-ref> --field scenarios=ubuntu-repo-cloud-openclaw-telegram

Optional scenario E2E

• gpu-repo-local-ollama-openclaw: Optional special-runner coverage for the local Ollama provider path in src/lib/inference/local.ts; this is the scenario suite's dedicated local-Ollama/GPU path.
  • Dispatch: gh workflow run e2e-scenarios.yaml --ref <pr-head-ref> --field scenarios=gpu-repo-local-ollama-openclaw
• ubuntu-repo-openai-compatible-openclaw: Optional adjacent provider-onboarding coverage for src/lib/inference/onboard-probes.ts changes affecting OpenAI-compatible endpoint probe parsing beyond the primary NVIDIA cloud onboarding path.
  • Dispatch: gh workflow run e2e-scenarios.yaml --ref <pr-head-ref> --field scenarios=ubuntu-repo-openai-compatible-openclaw

Relevant changed files

• nemoclaw/src/blueprint/runner.ts
• nemoclaw/src/commands/migration-state.ts
• src/lib/actions/sandbox/destroy.ts
• src/lib/actions/sandbox/snapshot.ts
• src/lib/inference/local.ts
• src/lib/inference/onboard-probes.ts
• src/lib/messaging/compiler/manifest-compiler.ts
• src/lib/shields/timer-control.ts
• src/lib/state/sandbox-session.ts
• src/lib/state/tar-listing.ts