add debug mcp tool for testing, fix transport issues

agents/codex.ts

@@ -128,7 +128,7 @@ const messageHandlers: {
         toolName: item.tool,
         input: {
           server: item.server,
-          ...((item as any).args || {}),
+          ...((item as any).arguments || {}),
         },
       });
     }
@@ -167,7 +167,7 @@ const messageHandlers: {
       const reasoningText = item.text.trim();
       // Remove markdown bold markers if present for cleaner output
       const cleanText = reasoningText.replace(/\*\*/g, "");
-      log.info(cleanText);
+      log.box(cleanText, { title: "Codex" });
     }
   },
   error: (event) => {

fixtures/basic.txt

@@ -1 +1 @@
-write a comment to https://github.com/pullfrogai/scratch/pull/29 that tells a joke
+use debug_shell_command tool to run 'git status' and explain the result

mcp/debug.ts

@@ -0,0 +1,45 @@
+import { type } from "arktype";
+import { $ } from "../utils/shell.ts";
+import type { ToolResult } from "./shared.ts";
+import { tool } from "./shared.ts";
+
+export const DebugShellCommand = type({});
+
+export const DebugShellCommandTool = tool({
+  name: "debug_shell_command",
+  description:
+    "debug tool: runs 'git status' and returns the output. use this to test shell command execution in the MCP server.",
+  parameters: DebugShellCommand,
+  execute: async (): Promise<ToolResult> => {
+    try {
+      const result = $("git", ["status"]);
+
+      return {
+        content: [
+          {
+            type: "text",
+            text: JSON.stringify(
+              {
+                success: true,
+                command: "git status",
+                output: result.trim(),
+              },
+              null,
+              2
+            ),
+          },
+        ],
+      };
+    } catch (error) {
+      return {
+        content: [
+          {
+            type: "text",
+            text: `Error: ${error instanceof Error ? error.message : String(error)}`,
+          },
+        ],
+        isError: true,
+      };
+    }
+  },
+});

mcp/server.ts

@@ -9,6 +9,7 @@ import {
   EditCommentTool,
   UpdateWorkingCommentTool,
 } from "./comment.ts";
+import { DebugShellCommandTool } from "./debug.ts";
 import { IssueTool } from "./issue.ts";
 import { PullRequestTool } from "./pr.ts";
 import { PullRequestInfoTool } from "./prInfo.ts";
@@ -35,6 +36,7 @@ addTools(server, [
   GetReviewCommentsTool,
   ListPullRequestReviewsTool,
   GetCheckSuiteLogsTool,
+  DebugShellCommandTool,
 ]);
 
 server.start();

modes.ts

@@ -6,7 +6,7 @@ export interface Mode {
   prompt: string;
 }
 
-const initialCommentInstruction = `Use ${ghPullfrogMcpName}/create_working_comment to create an initial Working Comment with a conversational description of what work you are about to perform.`;
+const initialCommentInstruction = `When the task is associated with an issue/PR, use ${ghPullfrogMcpName}/create_working_comment to create an initial Working Comment with a conversational description of what work you are about to perform.`;
 
 export const modes: Mode[] = [
   {

play.ts

@@ -141,6 +141,8 @@ Examples:
     if (!result.success) {
       process.exit(1);
     }
+
+    process.exit(0);
   } catch (err) {
     log.error((err as Error).message);
     process.exit(1);

utils/shell.ts

@@ -29,8 +29,11 @@ interface ShellOptions {
  */
 export function $(cmd: string, args: string[], options?: ShellOptions): string {
   const encoding = options?.encoding ?? "utf-8";
+
+  // CRITICAL: use "ignore" for stdin instead of "inherit" to avoid breaking MCP transport
+  // when running inside an MCP server, stdin is used for JSON-RPC protocol
   const result = spawnSync(cmd, args, {
-    stdio: ["inherit", "pipe", "pipe"],
+    stdio: ["ignore", "pipe", "pipe"],
     encoding,
     cwd: options?.cwd,
   });
@@ -39,10 +42,19 @@ export function $(cmd: string, args: string[], options?: ShellOptions): string {
   const stderr = result.stderr ?? "";
 
   // Write output to process streams so it behaves like stdio: "inherit"
+  // CRITICAL: when running inside an MCP server, stdout is used for JSON-RPC protocol
+  // so we must write to stderr instead to avoid corrupting the protocol
   // Only log if log option is not explicitly set to false
   if (options?.log !== false) {
+    // if stdout is a TTY, it's safe to write to it; otherwise it's likely a pipe used for JSON-RPC
+    const canWriteToStdout = process.stdout.isTTY === true;
     if (stdout) {
-      process.stdout.write(stdout);
+      if (canWriteToStdout) {
+        process.stdout.write(stdout);
+      } else {
+        // stdout is a pipe (MCP context) - write to stderr instead
+        process.stderr.write(stdout);
+      }
     }
     if (stderr) {
       process.stderr.write(stderr);