claude-skills

Claude Code skills (operational runbooks) for deploying and verifying Together T4 (GNS control-plane + s3-proxy data-plane) and CS3 (s3-cache-proxy) on a k3s cluster backed by Rook-Ceph object storage.

Each skill is a self-contained SKILL.md (with YAML frontmatter name + description) that Claude Code can discover and invoke. Drop a skill directory into ~/.claude/skills/ (user-global) or <repo>/.claude/skills/ (project-scoped).

Skills

Skill	What it does
deploy-t4-cs3-over-ceph	Clean-install / reinstall GNS + s3-proxy + CS3 over a Rook-Ceph RGW (helm, secrets, Postgres wipe+reseed, ECR pull secrets, both data paths).
gns-seed-ceph-store	Seed/repoint the GNS Postgres site→store routing to a live Ceph RGW endpoint.
s3-proxy-ceph-e2e	End-to-end S3 round-trip (create/put/get/list) through the s3-proxy LB; diagnose auth (keyExpired/AccessDenied) from proxy logs.
create-ums-sigv4-key	Mint a SigV4 S3 access key on UMS via its gRPC Key API (includes a runnable Go client in client/).

The skills cross-reference each other via [[skill-name]] links.

Placeholders — fill in for your environment

These runbooks are sanitized. Replace the placeholders with your own values (e.g. via env vars or a local, untracked notes file):

Placeholder	Meaning
<TRANSPORTER_IP> / <TRANSPORTER_WORKER_IP>	k3s control-plane / worker node IPs
<TRANSPORTER_HOST> / <TRANSPORTER_WORKER_HOST>	k3s node hostnames
<CEPH_IP> / <CEPH_HOST>	Rook-Ceph node IP / hostname
<CEPH_RGW_NODEPORT>	Ceph RGW endpoint reachable cross-cluster (e.g. http://<CEPH_IP>:30080)
<T4_ECR_ACCOUNT> / <CS3_ECR_ACCOUNT>	AWS account IDs hosting the T4 / CS3 ECR repos
<UMS_ENDPOINT>	UMS gRPC endpoint (host:50051)
<USER> / <CEPH_USER>	SSH users for the compute / storage hosts
<GNS_PG_PASSWORD>	GNS Postgres password
<UMS_SERVICE_KEY>	UMS service key (ums_v1_…) — never commit a real one

Never commit real credentials. The Go client reads UMS_ENDPOINT, UMS_SERVICE_KEY, and EXISTING_KEY from the environment.

License

Internal operational documentation. No warranty.