fix: 修复4.20模型对话错误和注入提示词以支持openclaw调用工具

app/api/v1/chat.py

@@ -29,6 +29,13 @@ from app.services.grok.utils.response import (
 )
 
 
+try:
+    with open("app_traffic.log", "w", encoding="utf-8") as f:
+        import datetime
+        f.write(f"=== app_traffic.log Cleared on Uvicorn Loaded [{datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}] ===\n\n")
+except Exception:
+    pass
+
 class MessageItem(BaseModel):
     """消息项"""
 
@@ -393,7 +400,7 @@ def validate_request(request: ChatCompletionRequest):
 
         # 字符串内容
         if isinstance(content, str):
-            if not content.strip():
+            if not content.strip() and msg.role != "assistant":
                 raise ValidationException(
                     message="Message content cannot be empty",
                     param=f"messages.{idx}.content",
@@ -514,7 +521,7 @@ def validate_request(request: ChatCompletionRequest):
                         "file.file_data",
                         f"messages.{idx}.content.{block_idx}.file.file_data",
                     )
-        elif content is None and msg.role == "tool":
+        elif content is None and (msg.role == "tool" or msg.tool_calls):
             pass
         elif not isinstance(content, list):
             raise ValidationException(
@@ -1034,6 +1041,20 @@ async def chat_completions(request: ChatCompletionRequest, raw_request: Request)
         except Exception as e:
             return _chat_error_as_success_response(request.model, _video_error_message(e))
     else:
+        try:
+            with open("app_traffic.log", "a", encoding="utf-8") as f:
+                import datetime
+                f.write(f"--- Chat Request [{datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}] ---\n")
+                dump = {
+                    "model": request.model,
+                    "messages": [m.model_dump() for m in request.messages],
+                    "stream": request.stream,
+                    "tools": request.tools if request.tools else None,
+                }
+                f.write(orjson.dumps(dump, option=orjson.OPT_INDENT_2).decode('utf-8') + "\n\n")
+        except Exception:
+            pass
+
         result = await ChatService.completions(
             model=request.model,
             messages=[msg.model_dump() for msg in request.messages],

app/services/grok/services/chat.py

@@ -1,4 +1,4 @@
-"""
+"""
 Grok Chat 服务
 """
 
@@ -130,7 +130,14 @@ class MessageExtractor:
 
             if isinstance(content, str):
                 if content.strip():
-                    parts.append(content)
+                    text_to_add = content
+                    if role == "tool" and len(text_to_add) > 3000:
+                        text_to_add = (
+                            text_to_add[:1500] 
+                            + "\n\n... [Output Truncated for Conciseness] ...\n\n" 
+                            + text_to_add[-1000:]
+                        )
+                    parts.append(text_to_add)
             elif isinstance(content, dict):
                 content = [content]
                 for item in content:
@@ -209,11 +216,11 @@ class MessageExtractor:
                 role_label = role
                 if role == "tool":
                     name = msg.get("name")
-                    call_id = msg.get("tool_call_id")
                     if isinstance(name, str) and name.strip():
-                        role_label = f"tool[{name.strip()}]"
-                    if isinstance(call_id, str) and call_id.strip():
-                        role_label = f"{role_label}#{call_id.strip()}"
+                        role_label = f"--- Action Result: {name.strip()} ---"
+                        full_text = "\n".join(parts)
+                        if "exited with code" in full_text and "code 0" not in full_text:
+                            role_label += " [FAILED]"
                 extracted.append({"role": role_label, "text": "\n".join(parts)})
 
         # 找到最后一条 user 消息
@@ -239,7 +246,26 @@ class MessageExtractor:
         if tools:
             tool_prompt = build_tool_prompt(tools, tool_choice, parallel_tool_calls)
             if tool_prompt:
-                combined = f"{tool_prompt}\n\n{combined}"
+                # 注入 Few-Shot 引导以避免模型惯性，强化 Tool Calling 回语。
+                few_shot = (
+                    "\n\n--- CRITICAL: TOOL CALL REQUIREMENT ---\n"
+                    "Example:\n"
+                    "User: test.txt 里面有什么内容？\n"
+                    "Assistant: <tool_call>{\"name\": \"read\", \"arguments\": {\"path\": \"test.txt\"}}</tool_call>\n"
+                    "User: 【System Notification: Tool Execution Result - read】\nhello world\n"
+                    "Assistant: test.txt 的内容是 hello world。\n"
+                    "--- END EXAMPLE ---\n\n"
+                    "Important Guidelines:\n"
+                    "1. **Task Complete**: If the information you have (including tool execution results) is SUFFICIENT to answer the user, you MUST answer directly in human language. DO NOT call any tools.\n"
+                    "2. **More Tools Needed**: If it is INDISPENSABLE to use a tool, you MUST output ONLY the raw `<tool_call>` block without conversational narrative filler."
+                )
+                combined = f"{tool_prompt}{few_shot}\n\n{combined}"
+                # 尾插强化 (Tail Enforcement)
+                combined += (
+                    "\n\n--- Important Note ---\n"
+                    "- Only respond using the valid structural `<tool_call>` JSON tag if you need data from available tools.\n"
+                    "- Provide only one single execution block and wait for response without dialogue."
+                )
 
         return combined, file_attachments, image_attachments
 

app/services/grok/utils/tool_call.py

@@ -34,9 +34,13 @@ def build_tool_prompt(
         return ""
 
     lines = [
+        "Important Rule: Focus strictly on answering using tool output data and JSON.",
+        "To satisfy the task, leverage available tools in sequence without adding conversational filler.",
+        "Do not formulate standard chatbot dialogue or greet.",
+        "",
         "# Available Tools",
         "",
-        "You have access to the following tools. To call a tool, output a <tool_call> block with a JSON object containing \"name\" and \"arguments\".",
+        "To call a tool, you MUST output a <tool_call> block containing a JSON object with \"name\" and \"arguments\".",
         "",
         "Format:",
         "<tool_call>",
@@ -77,10 +81,14 @@ def build_tool_prompt(
             lines.append(f"IMPORTANT: You MUST call the tool \"{forced_name}\" in your response.")
     else:
         # "auto" or default
-        lines.append("Decide whether to call a tool based on the user's request. If you don't need a tool, respond normally with text only.")
+        lines.append("Decide whether to call a tool based on the user's request. IF YOU CALL A TOOL, DO NOT ADD ANY DIALOGUE. OUTPUT <tool_call> FAST AND ONLY.")
 
     lines.append("")
     lines.append("When you call a tool, you may include text before or after the <tool_call> blocks, but the tool call blocks must be valid JSON.")
+    lines.append("")
+    lines.append("(If the task is complete and no more tools are needed, provide a regular answer. Only use <tool_call> when you actually need to call a tool.)")
+    lines.append("")
+    lines.append("IMPORTANT (Tool Error Handling): If a tool (e.g., `read`) fails with `EISDIR: illegal operation on a directory`, it means the path is a directory. You MUST call `exec` with `dir` (on Windows) or `ls` (on Linux) to list its contents instead.")
 
     return "\n".join(lines)
 
@@ -296,14 +304,8 @@ def format_tool_history(messages: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
             })
 
         elif role == "tool":
-            # Convert tool result to text format
-            tool_name = name or "unknown"
-            call_id = tool_call_id or ""
-            content_str = content if isinstance(content, str) else json.dumps(content, ensure_ascii=False) if content else ""
-            result.append({
-                "role": "user",
-                "content": f"tool ({tool_name}, {call_id}): {content_str}",
-            })
+            # 保留 tool 角色，交给 MessageExtractor 统一压平或由后台处理
+            result.append(msg)
 
         else:
             result.append(msg)

app/services/reverse/app_chat.py

@@ -39,6 +39,18 @@ def _is_transient_network_error(err: Exception) -> bool:
 class AppChatReverse:
     """/rest/app-chat/conversations/new reverse interface."""
 
+    @staticmethod
+    def _resolve_custom_personality() -> Optional[str]:
+        """Resolve optional custom personality from app config."""
+        value = get_config("app.custom_instruction", "")
+        if value is None:
+            return None
+        if not isinstance(value, str):
+            value = str(value)
+        if not value.strip():
+            return None
+        return value
+
     @staticmethod
     def build_payload(
         message: str,
@@ -94,6 +106,13 @@ class AppChatReverse:
         if model_config_override:
             payload["responseMetadata"]["modelConfigOverride"] = model_config_override
 
+        if model == "grok-420":
+            payload["enable420"] = True
+
+        custom_personality = AppChatReverse._resolve_custom_personality()
+        if custom_personality is not None and "Greet the user" not in message[-1000:]:
+            payload["customPersonality"] = custom_personality
+
         return payload
 
     @staticmethod

app/services/reverse/utils/headers.py

@@ -1,8 +1,8 @@
 """Shared header builders for reverse interfaces."""
 
+import re
 import uuid
 import orjson
-import random
 from urllib.parse import urlparse
 from typing import Dict, Optional
 
@@ -10,6 +10,52 @@ from app.core.logger import logger
 from app.core.config import get_config
 from app.services.reverse.utils.statsig import StatsigGenerator
 
+_HEADER_CHAR_REPLACEMENTS = str.maketrans(
+    {
+        "\u2010": "-",  # hyphen
+        "\u2011": "-",  # non-breaking hyphen
+        "\u2012": "-",  # figure dash
+        "\u2013": "-",  # en dash
+        "\u2014": "-",  # em dash
+        "\u2212": "-",  # minus sign
+        "\u2018": "'",  # left single quote
+        "\u2019": "'",  # right single quote
+        "\u201c": '"',  # left double quote
+        "\u201d": '"',  # right double quote
+        "\u00a0": " ",  # nbsp
+        "\u2007": " ",  # figure space
+        "\u202f": " ",  # narrow nbsp
+        "\u200b": "",  # zero width space
+        "\u200c": "",  # zero width non-joiner
+        "\u200d": "",  # zero width joiner
+        "\ufeff": "",  # bom
+    }
+)
+
+
+def _sanitize_header_value(
+    value: Optional[str],
+    *,
+    field_name: str,
+    remove_all_spaces: bool = False,
+) -> str:
+    """Normalize header values and make sure they are latin-1 safe."""
+    raw = "" if value is None else str(value)
+    normalized = raw.translate(_HEADER_CHAR_REPLACEMENTS)
+    if remove_all_spaces:
+        normalized = re.sub(r"\s+", "", normalized)
+    else:
+        normalized = normalized.strip()
+
+    # curl_cffi header encoding defaults to latin-1.
+    normalized = normalized.encode("latin-1", errors="ignore").decode("latin-1")
+
+    if normalized != raw:
+        logger.warning(
+            f"Sanitized header field '{field_name}' (len {len(raw)} -> {len(normalized)})"
+        )
+    return normalized
+
 
 def build_sso_cookie(sso_token: str) -> str:
     """
@@ -23,15 +69,41 @@ def build_sso_cookie(sso_token: str) -> str:
     """
     # Format
     sso_token = sso_token[4:] if sso_token.startswith("sso=") else sso_token
+    sso_token = _sanitize_header_value(
+        sso_token, field_name="sso_token", remove_all_spaces=True
+    )
 
     # SSO Cookie
     cookie = f"sso={sso_token}; sso-rw={sso_token}"
 
     # CF Cookies
-    cf_cookies = get_config("proxy.cf_cookies") or ""
-    if not cf_cookies:
-        cf_clearance = get_config("proxy.cf_clearance")
-        if cf_clearance:
+    cf_cookies = _sanitize_header_value(
+        get_config("proxy.cf_cookies") or "", field_name="proxy.cf_cookies"
+    )
+    cf_clearance = _sanitize_header_value(
+        get_config("proxy.cf_clearance") or "",
+        field_name="proxy.cf_clearance",
+        remove_all_spaces=True,
+    )
+    cf_refresh_enabled = bool(get_config("proxy.enabled"))
+
+    if cf_refresh_enabled:
+        if not cf_cookies and cf_clearance:
+            cf_cookies = f"cf_clearance={cf_clearance}"
+    elif cf_clearance:
+        if cf_cookies:
+            # Replace existing cf_clearance or append if missing.
+            if re.search(r"(?:^|;\\s*)cf_clearance=", cf_cookies):
+                cf_cookies = re.sub(
+                    r"(^|;\\s*)cf_clearance=[^;]*",
+                    r"\\1cf_clearance=" + cf_clearance,
+                    cf_cookies,
+                    count=1,
+                )
+            else:
+                cf_cookies = cf_cookies.rstrip("; ")
+                cf_cookies = f"{cf_cookies}; cf_clearance={cf_clearance}"
+        else:
             cf_cookies = f"cf_clearance={cf_clearance}"
     if cf_cookies:
         if cookie and not cookie.endswith(";"):
@@ -41,6 +113,95 @@ def build_sso_cookie(sso_token: str) -> str:
     return cookie
 
 
+def _extract_major_version(browser: Optional[str], user_agent: Optional[str]) -> Optional[str]:
+    if browser:
+        match = re.search(r"(\d{2,3})", browser)
+        if match:
+            return match.group(1)
+    if user_agent:
+        for pattern in [r"Edg/(\d+)", r"Chrome/(\d+)", r"Chromium/(\d+)"]:
+            match = re.search(pattern, user_agent)
+            if match:
+                return match.group(1)
+    return None
+
+
+def _detect_platform(user_agent: str) -> Optional[str]:
+    ua = user_agent.lower()
+    if "windows" in ua:
+        return "Windows"
+    if "mac os x" in ua or "macintosh" in ua:
+        return "macOS"
+    if "android" in ua:
+        return "Android"
+    if "iphone" in ua or "ipad" in ua:
+        return "iOS"
+    if "linux" in ua:
+        return "Linux"
+    return None
+
+
+def _detect_arch(user_agent: str) -> Optional[str]:
+    ua = user_agent.lower()
+    if "aarch64" in ua or "arm" in ua:
+        return "arm"
+    if "x86_64" in ua or "x64" in ua or "win64" in ua or "intel" in ua:
+        return "x86"
+    return None
+
+
+def _build_client_hints(browser: Optional[str], user_agent: Optional[str]) -> Dict[str, str]:
+    browser = (browser or "").strip().lower()
+    user_agent = user_agent or ""
+    ua = user_agent.lower()
+
+    is_edge = "edge" in browser or "edg" in ua
+    is_brave = "brave" in browser
+    is_chromium = any(key in browser for key in ["chrome", "chromium", "edge", "brave"]) or (
+        "chrome" in ua or "chromium" in ua or "edg" in ua
+    )
+    is_firefox = "firefox" in ua or "firefox" in browser
+    is_safari = ("safari" in ua and "chrome" not in ua and "chromium" not in ua and "edg" not in ua) or "safari" in browser
+
+    if not is_chromium or is_firefox or is_safari:
+        return {}
+
+    version = _extract_major_version(browser, user_agent)
+    if not version:
+        return {}
+
+    if is_edge:
+        brand = "Microsoft Edge"
+    elif "chromium" in browser:
+        brand = "Chromium"
+    elif is_brave:
+        brand = "Brave"
+    else:
+        brand = "Google Chrome"
+
+    sec_ch_ua = (
+        f"\"{brand}\";v=\"{version}\", "
+        f"\"Chromium\";v=\"{version}\", "
+        "\"Not(A:Brand\";v=\"24\""
+    )
+
+    platform = _detect_platform(user_agent)
+    arch = _detect_arch(user_agent)
+    mobile = "?1" if ("mobile" in ua or platform in ("Android", "iOS")) else "?0"
+
+    hints = {
+        "Sec-Ch-Ua": sec_ch_ua,
+        "Sec-Ch-Ua-Mobile": mobile,
+    }
+    if platform:
+        hints["Sec-Ch-Ua-Platform"] = f"\"{platform}\""
+    if arch:
+        hints["Sec-Ch-Ua-Arch"] = arch
+        hints["Sec-Ch-Ua-Bitness"] = "64"
+    hints["Sec-Ch-Ua-Model"] = "" if mobile == "?0" else ""
+    return hints
+
+
 def build_ws_headers(token: Optional[str] = None, origin: Optional[str] = None, extra: Optional[Dict[str, str]] = None) -> Dict[str, str]:
     """
     Build headers for WebSocket requests.
@@ -53,12 +214,24 @@ def build_ws_headers(token: Optional[str] = None, origin: Optional[str] = None,
     Returns:
         Dict[str, str]: The headers dictionary.
     """
+    user_agent = _sanitize_header_value(
+        get_config("proxy.user_agent"), field_name="proxy.user_agent"
+    )
+    if not user_agent:
+        user_agent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/136.0.0.0 Safari/537.36"
+    safe_origin = _sanitize_header_value(origin or "https://grok.com", field_name="origin")
     headers = {
-        "Origin": origin or "https://grok.com",
+        "Origin": safe_origin,
+        "User-Agent": user_agent,
+        "Accept-Language": "zh-CN,zh;q=0.9,en;q=0.8",
         "Cache-Control": "no-cache",
         "Pragma": "no-cache",
     }
 
+    client_hints = _build_client_hints(get_config("proxy.browser"), user_agent)
+    if client_hints:
+        headers.update(client_hints)
+
     if token:
         headers["Cookie"] = build_sso_cookie(token)
 
@@ -81,22 +254,29 @@ def build_headers(cookie_token: str, content_type: Optional[str] = None, origin:
     Returns:
         Dict[str, str]: The headers dictionary.
     """
-    trace_id = uuid.uuid4().hex
-    span_id = uuid.uuid4().hex[:16]
-
+    user_agent = _sanitize_header_value(
+        get_config("proxy.user_agent"), field_name="proxy.user_agent"
+    )
+    if not user_agent:
+        user_agent = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/136.0.0.0 Safari/537.36"
+    safe_origin = _sanitize_header_value(origin or "https://grok.com", field_name="origin")
+    safe_referer = _sanitize_header_value(
+        referer or "https://grok.com/", field_name="referer"
+    )
     headers = {
-        "Baggage": f"sentry-environment=production,sentry-release=c43385ae21231a335971832ae5b5e8bdba69852d,sentry-public_key=b311e0f2690c81f25e2c4cf6d4f7ce1c,sentry-trace_id={trace_id},sentry-org_id=4508179396558848,sentry-sampled=false,sentry-sample_rand={random.random()},sentry-sample_rate=0",
-        "Origin": origin or "https://grok.com",
+        "Accept-Encoding": "gzip, deflate, br, zstd",
+        "Accept-Language": "zh-CN,zh;q=0.9,en;q=0.8",
+        "Baggage": "sentry-environment=production,sentry-release=d6add6fb0460641fd482d767a335ef72b9b6abb8,sentry-public_key=b311e0f2690c81f25e2c4cf6d4f7ce1c",
+        "Origin": safe_origin,
         "Priority": "u=1, i",
-        "Referer": referer or "https://grok.com/",
+        "Referer": safe_referer,
         "Sec-Fetch-Mode": "cors",
-        "Sentry-Trace": f"{trace_id}-{span_id}-0",
-        "Traceparent": f"00-{trace_id}-{span_id}-00",
+        "User-Agent": user_agent,
     }
-    
-    user_agent = get_config("proxy.user_agent")
-    if user_agent:
-        headers["User-Agent"] = user_agent
+
+    client_hints = _build_client_hints(get_config("proxy.browser"), user_agent)
+    if client_hints:
+        headers.update(client_hints)
 
     # Cookie
     headers["Cookie"] = build_sso_cookie(cookie_token)