PR Pipeline — Authorize → Validate → Review → Merge / auth (pull_request) Successful in 9s
PR Pipeline — Authorize → Validate → Review → Merge / validate (pull_request) Successful in 6s
PR Pipeline — Authorize → Validate → Review → Merge / ai-review (pull_request) Successful in 5s
PR Pipeline — Authorize → Validate → Review → Merge / lint (pull_request) Failing after 28s
PR Pipeline — Authorize → Validate → Review → Merge / gate (pull_request) Skipped
The rebuilt VM 101 (192.168.68.8) kept only one SSH key in root's authorized_keys, so the health check's root probe returns Permission denied and misreports the healthy host as UNREACHABLE. The llama-server runs as llmuser, so that user can see the :8080 pid via ss -tlnp. Add per-host user to GPU_HOSTS (default root, llmuser for .8) and pass it through check_gpu_ports() into all ssh() calls. Closes the gpu-unreachable:192.168.68.8 leg while leaving the root SSH security decision for the captain.