OpenShell Operations Runbook#

Use this runbook when deploying Intended-generated policy for customer-operated OpenShell / NemoClaw runtime environments.

Preconditions#

• runtime owner and approver identified
• staging environment available
• rollback path validated
• operator has reviewed third-party runtime boundary

Rollout sequence#

1. Compile runtime policy artifact:
   • meritt openshell-compile --input meritt-openshell.json --output meritt-openshell.yaml
2. Review inferred presets, endpoint scope, and binary access.
3. Apply in staging with review-first execution mode.
4. Validate decision-to-audit linkage for first governed runs.
5. Promote to production after approver sign-off.

Production checks#

• verify policy hash/change ticket ID is recorded
• verify audit records include runtime correlation metadata
• verify fail-closed paths deny unapproved high-risk actions
• verify escalation flow for blocked or denied actions

Rollback protocol#

1. Reapply last-known-good policy artifact.
2. Switch runtime-sensitive capabilities to review mode.
3. Pause high-risk connector paths until drift is resolved.
4. Open incident review and attach decision/audit evidence.

Incident escalation#

• trigger when:
  • policy drift detected
  • unknown destination egress appears
  • execution/audit mismatch occurs
  • repeated deny events indicate malformed runtime requests
• include:
  • policy artifact hash
  • runtime environment
  • decision IDs / trace IDs
  • impacted systems and mitigation state

Shared-responsibility reminder#

Intended governs authorization and audit boundaries. Your team remains responsible for upstream runtime deployment, host hardening, credentials, and provider-specific operational controls.