ground-control_skill

Overview

This skill performs automated post-upgrade verification for OpenClaw to ensure your system remains healthy after updates. It defines a model/cron/channel ground-truth file and runs a five-phase verification flow with optional auto-repair for configuration and cron drift. The tool focuses on deterministic checks and safe, limited remediation to reduce manual troubleshooting.

How this skill works

The skill reads a workspace GROUND_TRUTH file that declares expected models, cron jobs, channels, and non-LLM provider env var rules. It runs five phases: config integrity, API key liveness for non-LLM providers, cron integrity, session smoke tests, and channel liveness; phases 1 and 3 can auto-patch drifted config or cron entries. Sensitive credentials are never logged or enumerated; non-LLM checks only read explicitly named env vars and only call exact allowed HTTPS hostnames.

When to use it

• Immediately after running openclaw update or installing a new OpenClaw release
• When you suspect config drift (model, cron, or channel mismatches)
• Before or after deploying agents that rely on scheduled jobs or third-party provider keys
• As a scheduled health check invoked from automation or via the /verify command

Best practices

• Keep a canonical MODEL_GROUND_TRUTH.md at workspace root and update it with any intentional changes
• Use --dry-run for verification-only runs before enabling auto-fix in production
• Limit non-LLM provider entries to only the env vars you intend to validate and list exact allowed_domain values
• Automate execution of the verification flow in your upgrade SOP to catch regressions early
• Review reports from failed checks promptly and treat auto-fix changes as audit-worthy events

Example use cases

• Run after upgrading OpenClaw to confirm model versions and cron schedules were preserved
• Detect and auto-repair cron job drift that broke scheduled ingestion or maintenance tasks
• Validate that third-party provider keys are still live and that session spawns succeed after an authentication change
• Include in CI/CD upgrade pipelines to prevent silent breakage of agent workflows