From 63c7868a48e8d039d9bd5829a740f2bb48eedce4 Mon Sep 17 00:00:00 2001
From: Uday <jhariyauday60@gmail.com>
Date: Thu, 16 Jul 2026 22:46:56 +0530
Subject: [PATCH] feat: implement secure Docker sandbox for code execution and
 fix upstream LLM API parameter routing

---
 requirements.txt          |   3 +-
 seimei/agents/code_act.py | 117 +++++++++++++++++++++++++++++++-------
 seimei/llm.py             |   2 +-
 3 files changed, 98 insertions(+), 24 deletions(-)

diff --git a/requirements.txt b/requirements.txt
index 7ae79b5..437384d 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -8,4 +8,5 @@ huggingface_hub
 duckduckgo_search
 requests
 ddgs
-PyMuPDF
\ No newline at end of file
+PyMuPDF
+docker
\ No newline at end of file
diff --git a/seimei/agents/code_act.py b/seimei/agents/code_act.py
index 86ee233..7a259c1 100644
--- a/seimei/agents/code_act.py
+++ b/seimei/agents/code_act.py
@@ -1,8 +1,11 @@
 from __future__ import annotations
 
 import asyncio
+import os
 import re
 import subprocess
+import uuid
+import docker
 from pathlib import Path
 from typing import Any, Dict, List, NamedTuple, Optional, Sequence, Tuple
 
@@ -215,30 +218,100 @@ def _run_command(
     python_heredoc: Optional[_PythonHeredoc],
     cwd: Optional[str],
 ) -> subprocess.CompletedProcess[str]:
-    if python_heredoc:
-        args: List[str] = [python_heredoc.executable]
-        if python_heredoc.has_dash:
-            args.append("-")
-        script_input = python_heredoc.script
-        if script_input and not script_input.endswith("\n"):
-            script_input += "\n"
-        return subprocess.run(
-            args,
-            input=script_input,
-            capture_output=True,
-            text=True,
-            timeout=timeout,
-            cwd=cwd,
+    try:
+        client = docker.from_env()
+    except Exception as e:
+        return subprocess.CompletedProcess(
+            args=code, 
+            returncode=1, 
+            stdout="", 
+            stderr=f"Docker is not available or not running: {e}"
         )
-    return subprocess.run(
-        code,
-        shell=True,
-        capture_output=True,
-        text=True,
-        timeout=timeout,
-        cwd=cwd,
-    )
 
+    host_workspace = os.path.abspath(cwd) if cwd and os.path.exists(cwd) else None
+    volumes = {}
+    working_dir = None
+    
+    if host_workspace:
+        volumes[host_workspace] = {'bind': '/workspace', 'mode': 'rw'}
+        working_dir = '/workspace'
+
+    temp_script_name = None
+    if python_heredoc and host_workspace:
+        temp_script_name = f".tmp_agent_{uuid.uuid4().hex}.py"
+        host_script_path = os.path.join(host_workspace, temp_script_name)
+        
+        with open(host_script_path, "w", encoding="utf-8") as f:
+            f.write(python_heredoc.script)
+            
+        container_cmd = [python_heredoc.executable, f"/workspace/{temp_script_name}"]
+    else:
+        container_cmd = ["/bin/sh", "-c", code]
+
+    image_name = "python:3.10-slim"
+    container = None
+    
+    try:
+        try:
+            container = client.containers.run(
+                image_name,
+                command=container_cmd,
+                volumes=volumes,
+                working_dir=working_dir,
+                network_mode="none",
+                mem_limit="256m",
+                nano_cpus=1000000000,
+                detach=True,
+            )
+        except docker.errors.ImageNotFound:
+            client.images.pull(image_name)
+            container = client.containers.run(
+                image_name,
+                command=container_cmd,
+                volumes=volumes,
+                working_dir=working_dir,
+                network_mode="none",
+                mem_limit="256m",
+                nano_cpus=1000000000,
+                detach=True,
+            )
+        
+        result = container.wait(timeout=timeout)
+        returncode = result.get("StatusCode", 0)
+        
+        stdout = container.logs(stdout=True, stderr=False).decode("utf-8", errors="replace")
+        stderr = container.logs(stdout=False, stderr=True).decode("utf-8", errors="replace")
+
+    except Exception as e:
+        if "ReadTimeout" in type(e).__name__ or "Timeout" in type(e).__name__:
+            if container:
+                try:
+                    container.stop(timeout=1)
+                except Exception:
+                    pass
+            raise subprocess.TimeoutExpired(cmd=code, timeout=timeout)
+            
+        returncode = 1
+        stdout = ""
+        stderr = f"Execution failed: {e}"
+        if container:
+            try:
+                container.stop(timeout=1)
+            except Exception:
+                pass
+    finally:
+        if container:
+            try:
+                container.remove(force=True)
+            except Exception:
+                pass
+        if temp_script_name and host_workspace:
+            try:
+                os.remove(os.path.join(host_workspace, temp_script_name))
+            except Exception:
+                pass
+
+    return subprocess.CompletedProcess(args=code, returncode=returncode, stdout=stdout, stderr=stderr)
 
 def _normalize_command(command: str) -> str:
     cmd = (command or "").strip()
diff --git a/seimei/llm.py b/seimei/llm.py
index b6c14e6..6a7f0ef 100644
--- a/seimei/llm.py
+++ b/seimei/llm.py
@@ -615,7 +615,7 @@ async def chat(
         else:
             payload = {
                 "model": self.model,
-                "messages": payload_msgs,
+                "input": payload_msgs,
             }
             payload.update(self._filter_payload(extra_params))
             if self.using_kyotoai: