Harden WorkBuddy risk controls

.env.example

@@ -97,6 +97,9 @@ CODEBUDDY2API_MAX_CONCURRENT=64
 # Per-account in-flight limit; zero keeps account capacity unlimited.
 # 单账号在途请求上限；设为 0 保持账号容量不受限。
 # CODEBUDDY2API_MAX_INFLIGHT_PER_ACCOUNT=0
+# Optional system prompt prefix; empty adds no gateway default, prepend uses the routed profile prompt.
+# 可选系统提示词前缀；empty 不附加网关默认提示词，prepend 按后端插入精简官方提示词。
+# CODEBUDDY2API_PROMPT_MODE=empty
 # Optional session/attempt tracing; legacy keeps existing upstream header behavior.
 # 可选的会话/尝试追踪；legacy 保持现有上游请求头行为。
 # CODEBUDDY2API_REQUEST_CONTEXT_MODE=legacy

app/client_profiles.py

@@ -5,9 +5,12 @@
 
 from .site_routing import domain_for_auth, profile_for_auth, profile_product, profile_region
 
-CLI_VERSION = "2.149.0"
-WORKBUDDY_VERSION = "5.5.2"
-WORKBUDDY_CLI_VERSION = "2.137.1"
+CLI_VERSION = "2.165.0"
+WORKBUDDY_VERSION = "5.9.2"
+WORKBUDDY_CLI_VERSION = CLI_VERSION
+INTL_CLI_VERSION = "2.161.0"
+INTL_WORKBUDDY_VERSION = "5.8.4"
+INTL_WORKBUDDY_CLI_VERSION = INTL_CLI_VERSION
 CLI_USER_AGENT = f"CLI/{CLI_VERSION} CodeBuddy/{CLI_VERSION}"
 
 SDK_HEADERS = {
@@ -26,14 +29,23 @@
 }
 
 
+def _versions(profile):
+    if profile_region(profile) == "intl":
+        return INTL_CLI_VERSION, INTL_WORKBUDDY_VERSION, INTL_WORKBUDDY_CLI_VERSION
+    return CLI_VERSION, WORKBUDDY_VERSION, WORKBUDDY_CLI_VERSION
+
+
 def identity_headers(profile: str) -> dict:
     """Use verified package versions without reading desktop settings or device identities."""
+    cli_version, workbuddy_version, workbuddy_cli_version = _versions(profile)
     if profile_product(profile) == "cli":
-        return {"User-Agent": CLI_USER_AGENT, "X-IDE-Type": "CLI", "X-IDE-Name": "CLI", "X-IDE-Version": CLI_VERSION}
+        return {"User-Agent": f"CLI/{cli_version} CodeBuddy/{cli_version}",
+                "X-IDE-Type": "CLI", "X-IDE-Name": "CLI", "X-IDE-Version": cli_version}
     name = "WorkBuddy AI" if profile_region(profile) == "intl" else "WorkBuddy"
     return {
-        "User-Agent": f"WorkBuddy/{WORKBUDDY_VERSION} {name}/{WORKBUDDY_VERSION} CLI/{WORKBUDDY_CLI_VERSION}",
-        "X-IDE-Type": "WorkBuddy", "X-IDE-Name": "WorkBuddy", "X-IDE-Version": WORKBUDDY_VERSION,
+        "User-Agent": f"WorkBuddy/{workbuddy_version} {name}/{workbuddy_version} CLI/{workbuddy_cli_version}",
+        "X-IDE-Type": "WorkBuddy", "X-IDE-Name": "WorkBuddy", "X-IDE-Version": workbuddy_version,
+        "X-Private-Data": "true",
     }
 
 
@@ -168,6 +180,7 @@ def canonical_uid(account) -> str:
 
 
 def catalog_cache_key(profile: str, identity: str | None = None) -> str:
-    revision = CLI_VERSION if profile_product(profile) == "cli" else f"{WORKBUDDY_VERSION}:{WORKBUDDY_CLI_VERSION}"
+    cli_version, workbuddy_version, workbuddy_cli_version = _versions(profile)
+    revision = cli_version if profile_product(profile) == "cli" else f"{workbuddy_version}:{workbuddy_cli_version}"
     key = f"{profile}:{revision}"
     return f"{key}:account:{identity}" if identity is not None else key

app/desensitize.py

@@ -109,11 +109,16 @@
 )
 
 # Limit substitutions to known templates without globally stripping whitespace.
-def _template_pattern(text: str) -> re.Pattern:
+def _template_source(text: str) -> str:
     words = [(_ZWSP + "*").join(re.escape(char) for char in word) for word in text.split(" ")]
-    return re.compile(r"(?<!\w)" + r"[\s\u200b]+".join(words) + _ZWSP + r"*(?!\w)", re.IGNORECASE)
+    return r"[\s\u200b]+".join(words) + _ZWSP + r"*"
+
 
+def _template_pattern(text: str) -> re.Pattern:
+    return re.compile(r"(?<!\w)" + _template_source(text) + r"(?!\w)", re.IGNORECASE)
 
+
+_DSH_WORKBUDDY_IDENTITY = "You are WorkBuddy, a powerful AI assistant"
 _WORKBUDDY_IDENTITY = "You are CodeBuddy, Tencent's official CLI"
 _CLAUDE_IDENTITY = "You are Claude Code, Anthropic's official CLI for Claude"
 _TEMPLATE_REPLACEMENTS = tuple(
@@ -123,10 +128,32 @@ def _template_pattern(text: str) -> re.Pattern:
         ("You are a Claude agent, built on Anthropic's Claude Agent SDK", _WORKBUDDY_IDENTITY),
         ("Main branch (you will usually use this for PRs)",
          "Main branch (you will usually use this for PR)"),
-        ("You are an AI agent powered by DeepSeek Harness.", "You are a helpful AI assistant."),
+        ("You are an AI agent powered by DeepSeek Harness.", _DSH_WORKBUDDY_IDENTITY + "."),
     )
 )
 _WORKBUDDY_IDENTITY_PATTERN = _template_pattern(_WORKBUDDY_IDENTITY)
+_CHANNEL_IDENTITY_PATTERN = _template_pattern("You are an AI agent powered by DeepSeek Harness")
+_DSH_CODING_PATTERN = re.compile(
+    r"(?<!\w)" + _template_source("You are a coding agent powered by the")
+    + r"[\s\u200b]+(?P<model>[^\r\n]{1,200}?)[\s\u200b]+"
+    + _template_source("model") + r"(?!\w)",
+    re.IGNORECASE,
+)
+_DSH_MINIMAL_PATTERN = _template_pattern("You are a helpful software engineer assistant")
+_DSH_DIRECT_PATTERN = _template_pattern("You are DeepSeek Harness")
+_DSH_WEB_GUI_PATTERN = _template_pattern("DeepSeek Harness Web GUI")
+_CHANNEL_IDENTITY_FIXED = _DSH_WORKBUDDY_IDENTITY
+
+WORKBUDDY_SYSTEM_PROMPT = (
+    "You are WorkBuddy, a powerful AI assistant.\n\n"
+    "Your main goal is to follow the USER's instructions at each message, denoted by the <user_query> tag."
+)
+CODEBUDDY_SYSTEM_PROMPT = (
+    "You are CodeBuddy Code.\n\n"
+    "You are an interactive CLI tool that helps users with software engineering tasks. "
+    "Use the instructions below and the tools available to you to assist the user. "
+    "Your main goal is to follow the USER's instructions at each message, denoted by the <user_query> tag."
+)
 _GIT_STATUS_CONTEXT = re.compile(
     r"^\s*(?:gitStatus:\s*|#\s*gitStatus\s*\n)"
     r"This is the git status at the start of the conversation\.", re.IGNORECASE | re.MULTILINE,
@@ -226,6 +253,22 @@ def desensitize_text(text: str) -> str:
     return _PATTERN.sub(lambda m: _zero_width_split(m.group(0)), text)
 
 
+def _replace_dsh_coding_persona(match: re.Match) -> str:
+    model = match.group("model").strip()
+    return f"{_DSH_WORKBUDDY_IDENTITY}. This conversation is powered by the {model} model"
+
+
+def desensitize_channel_text(text: str) -> str:
+    """Convert official DSH role prompts without changing unrelated brand references."""
+    if not text:
+        return text
+    text = _DSH_CODING_PATTERN.sub(_replace_dsh_coding_persona, text)
+    text = _CHANNEL_IDENTITY_PATTERN.sub(_CHANNEL_IDENTITY_FIXED, text)
+    text = _DSH_MINIMAL_PATTERN.sub(_CHANNEL_IDENTITY_FIXED, text)
+    text = _DSH_DIRECT_PATTERN.sub(_CHANNEL_IDENTITY_FIXED, text)
+    return _DSH_WEB_GUI_PATTERN.sub("WorkBuddy Web GUI", text)
+
+
 def _iter_text_blocks(content):
     """Yield text containers and keys from string or block-based Chat content."""
     if isinstance(content, str):
@@ -400,6 +443,46 @@ def desensitize_body(body: dict, roles: tuple[str, ...] = ("system",),
     return nb if changed else body
 
 
+def desensitize_channel_body(body: dict) -> dict:
+    """Remove the known channel identity from instruction and assistant text."""
+    if not body.get("messages"):
+        return body
+    messages = []
+    for message in body["messages"]:
+        if not isinstance(message, dict) or message.get("role") not in ("system", "developer", "assistant"):
+            messages.append(message)
+            continue
+        adapted = dict(message)
+        content = message.get("content")
+        if isinstance(content, str):
+            adapted["content"] = desensitize_channel_text(content)
+        elif isinstance(content, list):
+            adapted["content"] = [
+                {**block, "text": desensitize_channel_text(block["text"])}
+                if isinstance(block, dict) and block.get("type") == "text" and isinstance(block.get("text"), str)
+                else block
+                for block in content
+            ]
+        messages.append(adapted)
+    return {**body, "messages": messages}
+
+
+def apply_system_prompt_mode(body: dict, profile: str, mode: str) -> dict:
+    """Apply an optional profile prompt without mutating canonical messages."""
+    if mode == "empty":
+        return body
+    if mode != "prepend":
+        raise ValueError("prompt_mode 仅支持 empty 或 prepend")
+    prompt = WORKBUDDY_SYSTEM_PROMPT if profile in ("cn-work", "intl-work") else CODEBUDDY_SYSTEM_PROMPT
+    messages = body.get("messages")
+    if not isinstance(messages, list):
+        return body
+    if messages and isinstance(messages[0], dict) and messages[0].get("role") == "system" \
+            and _content_to_text(messages[0].get("content")) == prompt:
+        return body
+    return {**body, "messages": [{"role": "system", "content": prompt}, *messages]}
+
+
 # ---------------------------------------------------------------------------
 # Standalone template adaptation checks
 # ---------------------------------------------------------------------------

app/growth_tasks.py

@@ -76,17 +76,19 @@ def _growth_request(client, token, uid, domain, method, path, *, body=None, plat
             if len(raw) + len(chunk) > MAX_RESPONSE_BYTES:
                 raise Failure("protocol", response.status_code)
             raw.extend(chunk)
-        if response.status_code != 200:
-            raise Failure("http", response.status_code)
         try:
             payload = json.loads(raw)
         except (ValueError, RecursionError):
+            if response.status_code != 200:
+                raise Failure("http", response.status_code) from None
             raise Failure("protocol", response.status_code) from None
         code = payload.get("code") if isinstance(payload, dict) else None
+        message = str(payload.get("msg") or "")[:200] if isinstance(payload, dict) else ""
+        if response.status_code != 200:
+            raise Failure("http", response.status_code, code if type(code) is int else None, message)
         if code != 0:
             raise Failure("business", response.status_code,
-                          code if type(code) is int else None,
-                          str((payload or {}).get("msg") or "")[:200])
+                          code if type(code) is int else None, message)
         data = payload.get("data")
         return data if data is not None else {}
 
@@ -184,7 +186,10 @@ def sweep_account(token, uid="", domain=""):
                 claimed.append(code)
                 credit += got_credit
                 energy += got_energy
-        except Failure:
+        except Failure as error:
             errors += 1
+            if error.details.get("code") == 11140:
+                return {"ok": False, "error": error.details, "accepted": accepted, "claimed": claimed,
+                        "credit": credit, "energy": energy, "errors": errors}
     return {"ok": errors == 0, "accepted": accepted, "claimed": claimed, "credit": credit,
             "energy": energy, "errors": errors}

app/request_context.py

@@ -61,10 +61,32 @@ def _content_parts(content):
 @dataclass(frozen=True)
 class Attempt:
     id: str
+    request_id: str
+    trace_id: str
     index: int
     span: str
 
 
+_WORKBUDDY_PROFILES = frozenset({"cn-work", "intl-work"})
+
+
+def _uuid_text(value):
+    return str(uuid.UUID(hex=value)) if isinstance(value, str) and len(value) == 32 else str(value)
+
+
+def build_attempt_headers(headers, profile, *, root, message_id, request_id, trace_id, span, parent):
+    """Build one upstream attempt with profile-specific ID relationships."""
+    if profile in _WORKBUDDY_PROFILES:
+        message_id, request_id = _uuid_text(message_id), _uuid_text(request_id)
+    else:
+        request_id, trace_id = message_id, root
+    return {**headers, "X-Request-ID": request_id, "X-Conversation-Message-ID": message_id,
+            "X-Conversation-Request-ID": root, "X-Root-Request-ID": root, "X-Trace-ID": trace_id,
+            "traceparent": f"00-{trace_id}-{span}-01", "b3": f"{trace_id}-{span}-1-{parent}",
+            "X-B3-TraceId": trace_id, "X-B3-SpanId": span, "X-B3-ParentSpanId": parent,
+            "X-B3-Sampled": "1"}
+
+
 class RequestContext:
     def __init__(self, protocol, mode="legacy", headers=(), *, capability_guard=True):
         self.request_id = uuid.uuid4().hex
@@ -118,18 +140,19 @@ def conversation_id(self, profile, account):
     def start_attempt(self):
         with self._lock:
             self._attempt_index += 1
-            self.attempt = Attempt(uuid.uuid4().hex, self._attempt_index, secrets.token_hex(8))
+            self.attempt = Attempt(id=uuid.uuid4().hex, request_id=uuid.uuid4().hex,
+                                   trace_id=uuid.uuid4().hex, index=self._attempt_index,
+                                   span=secrets.token_hex(8))
             return self.attempt
 
-    def attempt_headers(self, headers, attempt):
+    def attempt_headers(self, headers, attempt, profile=None):
         if not self.scoped:
             return dict(headers)
-        root, span = self.request_id, attempt.span
-        return {**headers, "X-Request-ID": attempt.id, "X-Conversation-Message-ID": attempt.id,
-                "X-Conversation-Request-ID": root, "X-Root-Request-ID": root, "X-Trace-ID": root,
-                "traceparent": f"00-{root}-{span}-01", "b3": f"{root}-{span}-1-{self._root_span}",
-                "X-B3-TraceId": root, "X-B3-SpanId": span, "X-B3-ParentSpanId": self._root_span,
-                "X-B3-Sampled": "1"}
+        return build_attempt_headers(
+            headers, profile, root=self.request_id, message_id=attempt.id,
+            request_id=attempt.request_id, trace_id=attempt.trace_id,
+            span=attempt.span, parent=self._root_span,
+        )
 
     def attempt_metadata(self):
         data = {"request_id": self.request_id, "session_source": self.session_source}

app/settings.py

@@ -105,6 +105,8 @@ def _item(default, type_, label, *, mode="hot", env=None, minimum=None, maximum=
     "admin_allowed_origins": _item("", "string", "管理页额外信任来源", env="CODEBUDDY2API_ADMIN_ORIGINS",
                                    allow_empty=True, max_length=2000, validator=normalize_allowed_origins),
     "desensitize": _item(False, "boolean", "提示词脱敏"),
+    "prompt_mode": _item("empty", "string", "系统提示词模式",
+                         env="CODEBUDDY2API_PROMPT_MODE", choices=["empty", "prepend"]),
     "no_compact": _item(False, "boolean", "保留提示词全文"),
     "keep_tool_metadata": _item(False, "boolean", "保留工具描述", env="CODEBUDDY2API_KEEP_TOOL_METADATA"),
     "skip_check": _item(False, "boolean", "跳过启动预检", mode="restart"),