feat: improved UX for tool calls via execute_code (#6205)

crates/goose-cli/src/session/output.rs

@@ -271,6 +271,7 @@ fn render_tool_request(req: &ToolRequest, theme: Theme, debug: bool) {
         Ok(call) => match call.name.to_string().as_str() {
             "developer__text_editor" => render_text_editor_request(call, debug),
             "developer__shell" => render_shell_request(call, debug),
+            "code_execution__execute_code" => render_execute_code_request(call, debug),
             "subagent" => render_subagent_request(call, debug),
             "todo__write" => render_todo_request(call, debug),
             _ => render_default_request(call, debug),
@@ -445,6 +446,61 @@ fn render_shell_request(call: &CallToolRequestParam, debug: bool) {
     println!();
 }
 
+fn render_execute_code_request(call: &CallToolRequestParam, debug: bool) {
+    let tool_graph = call
+        .arguments
+        .as_ref()
+        .and_then(|args| args.get("tool_graph"))
+        .and_then(Value::as_array)
+        .filter(|arr| !arr.is_empty());
+
+    let Some(tool_graph) = tool_graph else {
+        return render_default_request(call, debug);
+    };
+
+    let count = tool_graph.len();
+    let plural = if count == 1 { "" } else { "s" };
+    println!();
+    println!(
+        "─── {} tool call{} | {} ──────────────────────────",
+        style(count).cyan(),
+        plural,
+        style("execute_code").magenta().dim()
+    );
+
+    for (i, node) in tool_graph.iter().filter_map(Value::as_object).enumerate() {
+        let tool = node
+            .get("tool")
+            .and_then(Value::as_str)
+            .unwrap_or("unknown");
+        let desc = node
+            .get("description")
+            .and_then(Value::as_str)
+            .unwrap_or("");
+        let deps: Vec<_> = node
+            .get("depends_on")
+            .and_then(Value::as_array)
+            .into_iter()
+            .flatten()
+            .filter_map(Value::as_u64)
+            .map(|d| (d + 1).to_string())
+            .collect();
+        let deps_str = if deps.is_empty() {
+            String::new()
+        } else {
+            format!(" (uses {})", deps.join(", "))
+        };
+        println!(
+            "  {}. {}: {}{}",
+            style(i + 1).dim(),
+            style(tool).cyan(),
+            style(desc).green(),
+            style(deps_str).dim()
+        );
+    }
+    println!();
+}
+
 fn render_subagent_request(call: &CallToolRequestParam, debug: bool) {
     print_tool_header(call);
 

crates/goose/src/agents/code_execution_extension.rs

@@ -31,10 +31,25 @@ type ToolCallRequest = (
     tokio::sync::oneshot::Sender<Result<String, String>>,
 );
 
+#[derive(Debug, Clone, Serialize, Deserialize, JsonSchema)]
+struct ToolGraphNode {
+    /// Tool name in format "server/tool" (e.g., "developer/shell")
+    tool: String,
+    /// Brief description of what this call does (e.g., "list files in /src")
+    description: String,
+    /// Indices of nodes this depends on (empty if no dependencies)
+    #[serde(default)]
+    depends_on: Vec<usize>,
+}
+
 #[derive(Debug, Serialize, Deserialize, JsonSchema)]
 struct ExecuteCodeParams {
     /// JavaScript code with ES6 imports for MCP tools.
     code: String,
+    /// DAG of tool calls showing execution flow. Each node represents a tool call.
+    /// Use depends_on to show data flow (e.g., node 1 uses output from node 0).
+    #[serde(default)]
+    tool_graph: Vec<ToolGraphNode>,
 }
 
 #[derive(Debug, Serialize, Deserialize, JsonSchema)]
@@ -701,6 +716,7 @@ impl McpClientTrait for CodeExecutionClient {
         Err(Error::TransportClosed)
     }
 
+    #[allow(clippy::too_many_lines)]
     async fn list_tools(
         &self,
         _next_cursor: Option<String>,
@@ -746,6 +762,15 @@ impl McpClientTrait for CodeExecutionClient {
                         - Last expression is the result
                         - No comments in code
 
+                        TOOL_GRAPH: Always provide tool_graph to describe the execution flow for the UI.
+                        Each node has: tool (server/name), description (what it does), depends_on (indices of dependencies).
+                        Example for chained operations:
+                        [
+                          {"tool": "developer/shell", "description": "list files", "depends_on": []},
+                          {"tool": "developer/text_editor", "description": "read README.md", "depends_on": []},
+                          {"tool": "developer/text_editor", "description": "write output.txt", "depends_on": [0, 1]}
+                        ]
+
                         BEFORE CALLING: Use the read_module tool to check required parameters.
                     "#}
                     .to_string(),

crates/goose/src/providers/canonical/build_canonical_models.rs

@@ -289,6 +289,7 @@ impl MappingReport {
     }
 }
 
+#[allow(clippy::too_many_lines)]
 async fn build_canonical_models() -> Result<()> {
     println!("Fetching models from OpenRouter API...");
 

scripts/test_providers.sh

@@ -55,8 +55,10 @@ fi
 if [ "$CODE_EXEC_MODE" = true ]; then
   echo "Mode: code_execution (JS batching)"
   BUILTINS="developer,code_execution"
-  # Match "execute_code | code_execution" or "read_module | code_execution" in output
-  SUCCESS_PATTERN="(execute_code \| code_execution)|(read_module \| code_execution)"
+  # Match code_execution tool usage:
+  # - "execute_code | code_execution" or "read_module | code_execution" (fallback format)
+  # - "tool call | execute_code" or "tool calls | execute_code" (new format with tool_graph)
+  SUCCESS_PATTERN="(execute_code \| code_execution)|(read_module \| code_execution)|(tool calls? \| execute_code)"
   SUCCESS_MSG="code_execution tool called"
   FAILURE_MSG="no code_execution tools called"
 else

ui/desktop/src/components/ToolCallWithResponse.tsx

@@ -17,6 +17,12 @@ import MCPUIResourceRenderer from './MCPUIResourceRenderer';
 import { isUIResource } from '@mcp-ui/client';
 import { CallToolResponse, Content, EmbeddedResource } from '../api';
 
+interface ToolGraphNode {
+  tool: string;
+  description: string;
+  depends_on: number[];
+}
+
 interface ToolCallWithResponseProps {
   isCancelledMessage: boolean;
   toolRequest: ToolRequestMessageContent;
@@ -410,6 +416,20 @@ function ToolCallView({
       case 'computer_control':
         return `poking around...`;
 
+      case 'execute_code': {
+        const toolGraph = args.tool_graph as unknown as ToolGraphNode[] | undefined;
+        if (toolGraph && Array.isArray(toolGraph) && toolGraph.length > 0) {
+          if (toolGraph.length === 1) {
+            return `${toolGraph[0].description}`;
+          }
+          if (toolGraph.length === 2) {
+            return `${toolGraph[0].tool}, ${toolGraph[1].tool}`;
+          }
+          return `${toolGraph.length} tools used`;
+        }
+        return 'executing code';
+      }
+
       default: {
         // Generic fallback for unknown tools: ToolName + CompactArguments
         // This ensures any MCP tool works without explicit handling
@@ -489,12 +509,34 @@ function ToolCallView({
         )
       }
     >
-      {/* Tool Details */}
-      {isToolDetails && (
-        <div className="border-t border-borderSubtle">
-          <ToolDetailsView toolCall={toolCall} isStartExpanded={isExpandToolDetails} />
-        </div>
-      )}
+      {(() => {
+        const toolName = toolCall.name.substring(toolCall.name.lastIndexOf('__') + 2);
+        const toolGraph = toolCall.arguments?.tool_graph as unknown as ToolGraphNode[] | undefined;
+        const code = toolCall.arguments?.code as unknown as string | undefined;
+        const hasToolGraph =
+          toolName === 'execute_code' &&
+          toolGraph &&
+          Array.isArray(toolGraph) &&
+          toolGraph.length > 0;
+
+        if (hasToolGraph) {
+          return (
+            <div className="border-t border-borderSubtle">
+              <ToolGraphView toolGraph={toolGraph} code={code} />
+            </div>
+          );
+        }
+
+        if (isToolDetails) {
+          return (
+            <div className="border-t border-borderSubtle">
+              <ToolDetailsView toolCall={toolCall} isStartExpanded={isExpandToolDetails} />
+            </div>
+          );
+        }
+
+        return null;
+      })()}
 
       {logs && logs.length > 0 && (
         <div className="border-t border-borderSubtle">
@@ -553,6 +595,45 @@ function ToolDetailsView({ toolCall, isStartExpanded }: ToolDetailsViewProps) {
   );
 }
 
+interface ToolGraphViewProps {
+  toolGraph: ToolGraphNode[];
+  code?: string;
+}
+
+function ToolGraphView({ toolGraph, code }: ToolGraphViewProps) {
+  const renderGraph = () => {
+    if (toolGraph.length === 0) return null;
+
+    const lines: string[] = [];
+
+    toolGraph.forEach((node, index) => {
+      const deps =
+        node.depends_on.length > 0 ? ` (uses ${node.depends_on.map((d) => d + 1).join(', ')})` : '';
+      lines.push(`${index + 1}. ${node.tool}: ${node.description}${deps}`);
+    });
+
+    return lines.join('\n');
+  };
+
+  return (
+    <div className="px-4 py-2">
+      <pre className="font-mono text-xs text-textSubtle whitespace-pre-wrap">{renderGraph()}</pre>
+      {code && (
+        <div className="border-t border-borderSubtle -mx-4 mt-2">
+          <ToolCallExpandable
+            label={<span className="pl-4 font-sans text-sm">Code</span>}
+            isStartExpanded={false}
+          >
+            <pre className="font-mono text-xs text-textSubtle whitespace-pre-wrap overflow-x-auto px-4 py-2">
+              {code}
+            </pre>
+          </ToolCallExpandable>
+        </div>
+      )}
+    </div>
+  );
+}
+
 interface ToolResultViewProps {
   result: Content;
   isStartExpanded: boolean;