Skip to content
78 changes: 76 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -39,9 +39,14 @@ Run `rtz` with no arguments for the full-screen UI.

| Key | Action |
|---|---|
| `:` | command bar — `:pods`, `:deploy`, `:sts`, `:ds`, `:svc`, `:ing`, `:nodes`, `:all`, `:risk`, `:signals`, `:changes`, `:logs`, `:traces`, `:ns <name>`, `:q` |
| `:` | browse every screen: the menu lists them grouped (resources, reliability, observability, capacity and lifecycle, add-ons) and narrows as you type — `↑↓` select, `tab` complete, `enter` open. A typo gets a "did you mean" |
| `:` | capacity and lifecycle — `:capacity`, `:pending`, `:storage`, `:maintain`, `:upgrade` |
| `:` | add-on workspaces — `:karpenter`, `:ca`, `:traefik`, `:istio`, `:argocd`, `:cni`, `:ebs` |
| `:` | resources and the rest — `:pods`, `:deploy`, `:sts`, `:ds`, `:svc`, `:ing`, `:nodes`, `:all`, `:risk`, `:signals`, `:changes`, `:logs`, `:traces`, `:ns <name>`, `:q` |
| `1`–`5` | switch group — Resources, Reliability, Observability, Capacity, Add-ons — back to the screen you last used there |
| `tab` / `shift+tab` | next / previous screen in the group; the two tab lines at the top show both |
| `/` | filter the visible rows as you type |
| `enter` / `d` | describe the selected workload |
| `enter` / `d` | describe the selected workload; on capacity, Maintain, upgrade and add-on screens, open the row's findings with their verify commands |
| `p` | pause or resume auto-refresh |
| `ctrl+r` | refresh now |
| `?` | keys and commands |
Expand Down Expand Up @@ -117,6 +122,57 @@ otherwise reads exactly like one built on fresh data. Extended-support costs are

`upgrade fleet` reports clusters it could not analyse rather than folding them into a pass.

`upgrade check` also lists the add-ons that must move with Kubernetes, how many patches the
cluster is behind, and an **Applications** section: per namespace, what the app team must
change before the upgrade and what is only affected by it.

## Capacity

```bash
rtz capacity # headroom per axis, node pools, pending pods
rtz capacity --section blocked # every pending pod and why (pools|nodes|blocked|storage|network|batch|special)
rtz capacity trend --days 30 # growth and days until each pool is full
rtz capacity cost # node cost per pool and the levers that change it
rtz capacity namespaces # requests, quota and pending pods per namespace
rtz capacity fleet # every cluster, soonest-full first
```

Usage gaps (nodes with no samples), truncated rows, uncollected quotas and unpriced nodes
are stated as warnings — an unread figure prints `?` or `unpriced`, never `0`.

## Maintain

```bash
rtz maintain items --within 90 # support deadlines across the org, soonest first
rtz maintain items --fail-on-days 30 # CI: exit 4 when a deadline is 30 days away or less
rtz maintain packages --ecosystem maven --q log4j
rtz maintain dependents data/StatefulSet/postgres # who connects to it
rtz maintain impact data/StatefulSet/postgres --target 17
```

Platform items (Kubernetes, add-ons, system workloads) are hidden unless `--platform` —
`rtz upgrade check` owns those. A list the backend could not finish reading exits 1.

## Add-on workspaces

```bash
rtz addons nodepools [name] # Karpenter (alias: karpenter)
rtz addons cluster-autoscaler [group] # alias: ca
rtz addons ingress [ns/name] # Traefik routes (alias: traefik)
rtz addons mesh [ns/name] # Istio (alias: istio)
rtz addons gitops [ns/name] # Argo CD (alias: argocd)
rtz addons vpc-cni [node] # Amazon VPC CNI
rtz addons ebs-csi [node] # Amazon EBS CSI
rtz addons istio --fail-on high # CI gate on add-on findings
```

Without an argument each lists its objects with the worst finding; with one it prints that
object's findings with the command that verifies each. An add-on that is not installed, not
synced or has its rules off says so instead of printing an empty table.

`-o json` / `-o yaml` on every command above prints the full server document, not the subset
the tables show.

## Pre-merge gate

```bash
Expand Down Expand Up @@ -165,6 +221,24 @@ computed from missing evidence is the most dangerous output this tool could prod
- `rca explain` states whether the answer came from a model, a cache replay, or a
deterministic gate — the prose alone does not reveal which.

## MCP server (Claude Code and other agents)

`rtz mcp` serves the Runtimez MCP server over stdio, so an MCP client can ask Runtimez
instead of guessing: what is running, what changed, ranked risk findings, the upgrade
readiness ledger with the vendor's verbatim rule text, and a pre-merge manifest check. It
is a bridge — every message is forwarded to the hosted endpoint (`POST /eac/mcp`) with this
CLI's credential, so the tool catalog is exactly the server's and every tool is read-only.

```bash
claude mcp add runtimez -- rtz mcp # uses the current context
claude mcp add runtimez -- rtz mcp --context prod --cluster prod-eu
```

Remote clients (Claude Code `--transport http`, Cursor, claude.ai connectors) can skip the
CLI and point at `https://<api>/eac/mcp` with `Authorization: Bearer rk_...` (an API key
with the `api:read` scope). A `--cluster` (or the context's cluster) is injected as the
default `cluster` argument when the client did not pass one.

## Everyday commands

```bash
Expand Down
Loading
Loading