feat(tab): AI agent usage meter in the tab bar

docs/docs/config.mdx

@@ -108,6 +108,12 @@ wsh editconfig
 | tab:preset <DeprecatedBadge />       | string   | a "bg@" preset to automatically apply to new tabs. e.g. `bg@green`. should match the preset key. deprecated in favor of `tab:background`                                                                                                                     |
 | tab:background <VersionBadge version="v0.14.4" />  | string   | a "bg@" preset to automatically apply to new tabs. e.g. `bg@green`. should match the preset key                                                                                                                                                               |
 | library:taborder                     | string   | order of the Library block's tabs, `"snippets,notes"` (default) or `"notes,snippets"`; the first tab opens by default. Set it by dragging one tab onto the other, or right-click the tab bar → Swap Tab Order |
+| agentusage:agent                     | string   | AI agent whose usage the tab bar meter tracks, `"claude"` (default) or `"codex"` (set to an empty string to turn the meter off; it then shows only a gauge icon). Usage is read from the agent's local session logs (`~/.claude/projects`, `~/.codex/sessions`) |
+| agentusage:conn                      | string   | connection the agent runs on, e.g. `user@host`; empty means this machine. Remote hosts need wsh installed |
+| agentusage:sessionhours              | int      | length in hours of the plan's rolling usage session (default 5, as on Claude Pro/Max); the session opens with the first message, floored to the hour |
+| agentusage:sessionlimit              | int      | your own token budget for one session; 0 shows raw usage without a bar |
+| agentusage:dailylimit                | int      | your own token budget for today (input + output + cache-write tokens); 0 shows raw usage without a bar. The plan's real limits are not in the logs |
+| agentusage:weeklylimit               | int      | your own token budget for the last 7 days; 0 shows raw usage without a bar |
 | tab:confirmclose                     | bool     | if set to true, a confirmation dialog will be shown before closing a tab (defaults to false)                                                                                                                                                                   |
 | tab:confirmcloserunning              | bool     | when true (default), closing a tab whose terminals are running a command shows a warning listing them, with Cancel as the default button. Detected via shell integration, for `term:tmux` terminals by asking tmux what each pane runs, and for plain SSH shells by the processes tagged with the terminal's `LC_WAVETERM_BLOCK` (needs the host's sshd to accept `LC_*`, the common default, and a `/proc` filesystem) |
 | widget:showhelp                      | bool     | whether to show help/tips widgets in right sidebar                                                                                                                                                                                                            |

frontend/app/store/wshclientapi.ts

@@ -24,6 +24,12 @@ export class RpcApiType {
         return client.wshRpcCall("activity", data, opts);
     }
 
+    // command "agentusage" [call]
+    AgentUsageCommand(client: WshClient, data: CommandAgentUsageData, opts?: RpcOpts): Promise<AgentUsageData> {
+        if (this.mockClient) return this.mockClient.mockWshRpcCall(client, "agentusage", data, opts);
+        return client.wshRpcCall("agentusage", data, opts);
+    }
+
     // command "aisendmessage" [call]
     AiSendMessageCommand(client: WshClient, data: AiMessageData, opts?: RpcOpts): Promise<void> {
         if (this.mockClient) return this.mockClient.mockWshRpcCall(client, "aisendmessage", data, opts);
@@ -738,6 +744,12 @@ export class RpcApiType {
         return client.wshRpcCall("recordtevent", data, opts);
     }
 
+    // command "remoteagentusage" [call]
+    RemoteAgentUsageCommand(client: WshClient, data: CommandAgentUsageData, opts?: RpcOpts): Promise<AgentUsageData> {
+        if (this.mockClient) return this.mockClient.mockWshRpcCall(client, "remoteagentusage", data, opts);
+        return client.wshRpcCall("remoteagentusage", data, opts);
+    }
+
     // command "remotedisconnectfromjobmanager" [call]
     RemoteDisconnectFromJobManagerCommand(client: WshClient, data: CommandRemoteDisconnectFromJobManagerData, opts?: RpcOpts): Promise<void> {
         if (this.mockClient) return this.mockClient.mockWshRpcCall(client, "remotedisconnectfromjobmanager", data, opts);