refactor: disable AI provider settings and remove first-run onboarding UI logic

Author: pierreneter

README.md

@@ -120,13 +120,21 @@ k10s --version        # which build is this
 
 ### Every resource, grouped the way you think about them
 
-16 kinds across **Workloads · Network · Config · Storage · Cluster · Custom
-Resources**, each with a live row count for the current namespace. Your CRDs
-are discovered automatically - no configuration.
-
-Pods · Deployments · StatefulSets · DaemonSets · Jobs · CronJobs · Services ·
-Ingresses · ConfigMaps · Secrets · PVCs · Nodes · Namespaces · Events · CRDs ·
-Custom Resources
+30 kinds across **Workloads · Network · Config · Storage · RBAC · Cluster ·
+Custom Resources**, each with a live row count for the current namespace.
+Your CRDs are discovered automatically - no configuration.
+
+Groups fold: `space`, `left` or a click on the header, remembered across
+restarts, with Config/Storage/RBAC folded to begin with. A folded group also
+asks your cluster for nothing - the sidebar is what decides which kinds get
+counted, and counting is one `limit=1` request per *visible* kind, six at a
+time, with refusals remembered.
+
+Pods · Deployments · ReplicaSets · StatefulSets · DaemonSets · Jobs ·
+CronJobs · HPAs · Services · Endpoints · Ingresses · NetworkPolicies ·
+ConfigMaps · Secrets · ResourceQuotas · LimitRanges · PDBs · PVCs · PVs ·
+StorageClasses · ServiceAccounts · Roles · RoleBindings · ClusterRoles ·
+ClusterRoleBindings · Nodes · Namespaces · Events · CRDs · Custom Resources
 
 ### The whole day-2 toolkit, on single keys
 
@@ -194,8 +202,8 @@ path, so [adding a theme is one struct](docs/themes.md).
 - **An honest loading state** instead of "no resources found" while a watch
   is still syncing.
 - **Namespace and context pickers that never ask you to type a name.**
-- **First-run onboarding** asks what you call kubectl (`kubectl`/`k8s`/`k`/
-  your own alias) and uses it in every hint from then on.
+- **No setup screen.** First run opens the cluster; every setting has a
+  working default and `/settings` is one keystroke away when you want it.
 - **A grow-able prompt** (`ctrl+z`) - because a long kubectl line or an AI
   question does not fit in a one-row field that scrolls sideways.
 - **A terminal-too-small notice** instead of a garbled layout.
@@ -244,22 +252,51 @@ The short version - the [full reference is here](docs/keybindings.md).
 
 ## Commands
 
-Two prefixes, and each one only ever shows its own set. **`/` does
-something. `:` narrows what is on screen.** `enter` runs the highlighted
-suggestion immediately - no second trip through the prompt.
+Two prefixes, and each one only ever shows its own set. **`/` is k10s itself
+(theme, settings, updates). `:` is the cluster, in the k9s vocabulary** - a
+resource view, the namespace, the context, or something acting on what is on
+screen. `enter` runs the highlighted suggestion immediately - no second trip
+through the prompt.
 
 ```
-/ns   /context   /theme   /settings   /update   /version   /help
-:search <term>   :filter <term>   :scale <n>   :mouse
+/theme   /settings   /update   /version   /help        k10s itself
+
+:po  :deploy  :rs  :sts  :ds  :job  :cj  :hpa      workloads
+:svc  :ep  :ing  :netpol                           network
+:cm  :sec  :quota  :limits  :pdb                   config
+:pvc  :pv  :sc                                     storage
+:sa  :role  :rb  :crole  :crb                      rbac
+:no  :ns  :ev  :crd  :cr                           cluster + custom
+
+the short form is what the popup lists; every kind also answers to its
+plural and singular - :pods, :pod, :deployments, :persistentvolumes ...
+and `:aliases` prints the lot
+
+:po kube-system                 that kind, in that namespace (`all` works)
+:svc api                        an argument that is not a namespace filters
+:ns   :ctx                      namespace / context picker
+:ns <name>   :ctx <name>        switch either outright
+:aliases  :search <term>  :filter <term>  :scale <n>  :mouse  :q
 ```
 
-None of `/ns`, `/context` or `/theme` takes a name - each opens a chooser
-showing what actually exists. [Full command reference](docs/commands.md).
+**`/` is k10s, `:` is the cluster** - namespace and context are `:ns` and
+`:ctx`, not `/ns` and `/context`. `:aliases` prints the whole `:` vocabulary.
+
+Anything that is not a `/` or `:` command **runs as a shell command** and its
+output opens in the main panel - `date`, `kubectl get pods -o wide`,
+`helm list`. Run through your own `$SHELL`, capped at 30s, with no terminal
+attached (use `s` on a pod for an interactive shell).
+[Full command reference](docs/commands.md).
 
 ## Config
 
-Theme, context, namespace, AI settings and the update check persist to
-`~/.k10s/config.yaml` (override with `K10S_CONFIG`), saved on every change.
+k10s opens on **kubeconfig's current-context** — the cluster `kubectl` would
+talk to. `:ctx <name>` switches for the session without touching kubeconfig;
+`kubectl config use-context` is what changes where it opens next time.
+
+Theme, namespace, folded sidebar groups, AI settings and the update check
+persist to `~/.k10s/config.yaml` (override with `K10S_CONFIG`), saved on
+every change.
 See [config.md](docs/config.md).
 
 > **Note:** the AI API key is stored as plain text in that file (mode 0600)

docs/architecture.md

@@ -18,13 +18,17 @@ internal/ui/
   model.go              Model state, Update: key/mouse dispatch, async commands
   view.go               View: header, list, table/text, actions, prompt
   actions.go            the action table (id, hotkey, label, risky)
-  commands.go           slash commands + /help text
+  commands.go           "/" and ":" commands, kind aliases, /help text
+  shellcmd.go           plain text at the prompt → a shell command
   msgs.go               async result messages
   connect.go            startup: pendingSource + the background connect
   update.go             /update + /version: the check, the confirm, the install
-  pickers.go/_view.go   onboarding + theme picker
+  pickers.go/_view.go   theme picker
   palette.go/_view.go   global search palette, namespace picker
-  nspicker.go           namespace picker logic
+  ctxpicker.go          context chooser, in the main panel
+  settings.go/_view.go  the settings dialog (CLI name, AI, update check)
+  logview.go            the log viewer
+  shell.go              the in-panel interactive shell (exec)
   loading.go            spinner / indeterminate progress
 ```
 
@@ -112,9 +116,9 @@ does the work off-thread and posts a message back (`internal/ui/msgs.go`):
 
 | Message                          | Produced by                                  |
 |----------------------------------|----------------------------------------------|
-| `textResultMsg`                  | describe / YAML / logs snapshot / top / AI   |
+| `textResultMsg`                  | describe / YAML / logs snapshot / top / AI / a shell command |
 | `actionResultMsg`                | delete, restart, scale, cordon, drain, apply |
-| `ctxSwitchMsg`                   | `/context` — rebuilds the whole client       |
+| `ctxSwitchMsg`                   | `:ctx` — rebuilds the whole client           |
 | `logStartMsg` / `logLineMsg`     | `LogsFollow` streaming                       |
 | `editFetchedMsg` / `editExitMsg` | the `$EDITOR` round trip                     |
 | `portForwardMsg`                 | port-forward start                           |
@@ -127,9 +131,9 @@ chain synchronously without a real event loop.
 
 Every clickable element is a named zone: `res:<i>`, `row:<i>`, `act:<id>`,
 `zoom`, `close`, `theme`, `nsfield`, `prompt`, `aimode`, `tablesearch`,
-`sug:<i>`, `pal:<i>`, `nsp:<i>`, `thm:<i>`, `ob:<i>`, `cf:ok`/`cf:no`,
-`cfg:*`. `handleMouse` checks overlay zones first (confirm → config → theme →
-onboarding → palette → namespace → suggestions), then chrome, then lists.
+`sug:<i>`, `pal:<i>`, `grp:<i>`, `thm:<i>`, `cf:ok`/`cf:no`,
+`set:<i>`. `handleMouse` checks overlay zones first (confirm → settings →
+theme → palette → suggestions), then chrome, then lists.
 
 Two fallbacks make panes feel solid rather than only their contents:
 - the wheel scrolls **and focuses** whichever pane the pointer is over
@@ -145,7 +149,7 @@ selection — the only way to copy text out of a mouse-capturing TUI.
 1. `ctrl+c` always quits
 2. confirm modal (enter/y confirm, esc/n cancel)
 3. search palette / namespace picker
-4. first-run onboarding, theme picker
+4. theme picker
 5. AI settings modal (incl. inline edit sub-state)
 6. prompt focused → suggestions navigation, then textinput
 7. resource list focused → type-to-filter

docs/backends.md

@@ -31,7 +31,7 @@ a `domain.Source` that serves what kubeconfig already knows — the kind list,
 the context list — and answers `errConnecting` to everything else. The main
 panel shows a spinner meanwhile, so an API server behind a downed VPN, or an
 exec credential plugin that stalls, leaves a working UI on screen instead of
-a blank terminal. Picking a context in `/context` during that wait retargets
+a blank terminal. Picking a context in `:ctx` during that wait retargets
 the connection rather than switching from a backend that isn't there yet;
 each attempt carries a generation, so a slow first one landing later is
 dropped instead of overwriting the newer one.
@@ -43,7 +43,7 @@ dropped instead of overwriting the newer one.
 | `client.go`          | kubeconfig loading, clientset/dynamic/discovery/metrics |
 | `store.go`           | the Store, lazy informer registry, metrics polling      |
 | `listers.go`         | lazy per-kind lister accessors                          |
-| `kinds.go`           | the 16 kinds, their columns and allowed actions         |
+| `kinds.go`           | the 30 kinds, their columns and allowed actions         |
 | `rows.go`            | object → row formatting, sorting, namespace filtering   |
 | `counts.go`          | cheap background counts for unopened kinds              |
 | `describe.go`        | kubectl's own describers, generic fallback for CRs      |
@@ -69,6 +69,15 @@ Table data comes from client-go informers, which keep a local cache warm off
 a watch. Informers are started **lazily, per kind** — see
 [performance.md](performance.md).
 
+Adding a kind means touching five places, all of them switch-shaped so a
+missing one fails loudly: `kinds.go` (entry, columns, allowed actions),
+`store.go` (informer registration + its GroupVersionResource, which is what
+YAML/edit/delete/badge counts all resolve through), `listers.go` (the lazy
+lister accessor), `rows.go` (object → row, plus the `RowCount` case), and
+`describe.go` (the GroupKind kubectl's describer is registered under). Then
+mirror it in `internal/mock` — `TestBackendsServeTheSameKinds` fails if you
+don't — and give it aliases in `internal/ui/commands.go`.
+
 `Describe` uses kubectl's own describers (`kubectl/pkg/describe`) so output
 matches `kubectl describe`, with a generic unstructured describer for CRDs
 and custom resources. `YAML` goes through the dynamic client and strips
@@ -95,8 +104,9 @@ and custom resources. `YAML` goes through the dynamic client and strips
   backwards cursor, so "older" means re-reading with a larger tail.
   `LogsFollow` then streams new lines over a channel; the UI tags each with a
   generation number so a stale stream can't append into a view you've left.
-  Workloads resolve to one of their pods via the workload's label selector,
-  preferring a running one. Kinds with no logs return `domain.ErrNoLogs`,
+  Workloads (Deployment, StatefulSet, DaemonSet, Job, ReplicaSet) resolve to
+  one of their pods via the workload's label selector, preferring a running
+  one. Kinds with no logs return `domain.ErrNoLogs`,
   which the UI treats as "show describe instead", not as a failure.
 - **Exec** — `ShellSession` opens an interactive exec and hands back a
   writer for keystrokes plus a channel of raw output bytes, so the UI can
@@ -116,7 +126,8 @@ reads its cache.
 
 ## Offline demo — `internal/mock`
 
-Static, in-memory fake cluster: 16 kinds, three nodes (one NotReady), and
+Static, in-memory fake cluster: the same 30 kinds, three nodes (one
+NotReady), and
 rows seeded with deliberate trouble — `billing-worker` in CrashLoopBackOff
 ×17, a `payment-api` pod Pending on insufficient CPU, a Terminating pod — so
 status colors and events have something real to point at.
@@ -127,7 +138,7 @@ It backs:
 - k10s itself when no cluster is reachable.
 
 A resource's base `Rows` are implicitly namespace `default`; `Extra` holds
-rows tagged with another namespace, which is what makes `/ns` switching show
+rows tagged with another namespace, which is what makes `:ns` switching show
 genuinely different data. `argocd` / `cert-manager` only carry Custom
 Resource instances.