What problem does it solve? Running Harbor evaluations against a LobeHub deployment requires coordinating infrastructure, credentials, CLI injection, preflight checks, and failure triage across local and cloud targets, which is error-prone when done manually. ## Core Features & Use Cases - Dual-target execution: Run existing Harbor evaluations against either an isolated local production LobeHub stack on port 3210 or a remote LobeHub Cloud target. - Gated preflight and smoke testing: Validate Compose services, gateways, JWKS keys, and endpoints, then run a model-backed hello-world smoke job before any real Harbor run or resume. - Flexible CLI injection: Inject a checkout build from apps/cli or install the published @lobehub/cli npm package inside Harbor task containers via the LhInstalledAgent. - Use Case: An engineer wants to resume a failed Harbor job against their local checkout; the Skill bootstraps the eval infrastructure, verifies preflight and smoke pass, then hands off to the eval repository's own resume command. ## Quick Start Ask the assistant to run an existing Harbor evaluation against the local LobeHub production build using the checkout CLI, providing the agent ID and eval repository path.
