Rewrite RPC documentation with accurate types

packages/coding-agent/README.md

@@ -786,7 +786,19 @@ Execute shell commands directly and add output to the LLM context by prefixing w
 - **History**: Commands are added to editor history (navigate with Up/Down arrows)
 - **Visual feedback**: Editor border turns green in bash mode; cancelled commands show yellow warning
 
-Output is automatically added to the conversation context, allowing the LLM to see command results without manual copy-paste.
+**How bash output reaches the LLM:**
+
+When you execute a bash command, the output is stored as a `BashExecutionMessage` in the conversation state. This message is transformed and sent to the LLM as part of your **next prompt**. The LLM sees it as a user message formatted like:
+
+```
+Ran `ls -la`
+\`\`\`
+total 48
+drwxr-xr-x ...
+\`\`\`
+```
+
+This means you can run multiple bash commands before sending a prompt, and all outputs will be included in the context together.
 
 ### Keyboard Shortcuts
 

packages/coding-agent/src/modes/rpc/rpc-mode.ts

@@ -86,9 +86,8 @@ export async function runRpcMode(session: AgentSession): Promise<never> {
 			// =================================================================
 
 			case "get_state": {
-				const model = session.model;
 				const state: RpcSessionState = {
-					model: model ? { provider: model.provider, id: model.id, contextWindow: model.contextWindow } : null,
+					model: session.model,
 					thinkingLevel: session.thinkingLevel,
 					isStreaming: session.isStreaming,
 					queueMode: session.queueMode,
@@ -112,7 +111,7 @@ export async function runRpcMode(session: AgentSession): Promise<never> {
 					return error(id, "set_model", `Model not found: ${command.provider}/${command.modelId}`);
 				}
 				await session.setModel(model);
-				return success(id, "set_model", { provider: model.provider, id: model.id });
+				return success(id, "set_model", model);
 			}
 
 			case "cycle_model": {
@@ -120,23 +119,12 @@ export async function runRpcMode(session: AgentSession): Promise<never> {
 				if (!result) {
 					return success(id, "cycle_model", null);
 				}
-				return success(id, "cycle_model", {
-					model: { provider: result.model.provider, id: result.model.id },
-					thinkingLevel: result.thinkingLevel,
-					isScoped: result.isScoped,
-				});
+				return success(id, "cycle_model", result);
 			}
 
 			case "get_available_models": {
 				const models = await session.getAvailableModels();
-				return success(id, "get_available_models", {
-					models: models.map((m) => ({
-						provider: m.provider,
-						id: m.id,
-						contextWindow: m.contextWindow,
-						reasoning: !!m.reasoning,
-					})),
-				});
+				return success(id, "get_available_models", { models });
 			}
 
 			// =================================================================

packages/coding-agent/src/modes/rpc/rpc-types.ts

@@ -6,6 +6,7 @@
  */
 
 import type { AppMessage, Attachment, ThinkingLevel } from "@mariozechner/pi-agent-core";
+import type { Model } from "@mariozechner/pi-ai";
 import type { CompactionResult, SessionStats } from "../../core/agent-session.js";
 import type { BashResult } from "../../core/bash-executor.js";
 
@@ -59,7 +60,7 @@ export type RpcCommand =
 // ============================================================================
 
 export interface RpcSessionState {
-	model: { provider: string; id: string; contextWindow: number } | null;
+	model: Model<any> | null;
 	thinkingLevel: ThinkingLevel;
 	isStreaming: boolean;
 	queueMode: "all" | "one-at-a-time";
@@ -91,21 +92,21 @@ export type RpcResponse =
 			type: "response";
 			command: "set_model";
 			success: true;
-			data: { provider: string; id: string };
+			data: Model<any>;
 	  }
 	| {
 			id?: string;
 			type: "response";
 			command: "cycle_model";
 			success: true;
-			data: { model: { provider: string; id: string }; thinkingLevel: ThinkingLevel; isScoped: boolean } | null;
+			data: { model: Model<any>; thinkingLevel: ThinkingLevel; isScoped: boolean } | null;
 	  }
 	| {
 			id?: string;
 			type: "response";
 			command: "get_available_models";
 			success: true;
-			data: { models: Array<{ provider: string; id: string; contextWindow: number; reasoning: boolean }> };
+			data: { models: Model<any>[] };
 	  }
 
 	// Thinking