chore: bump version to 4.26.0-beta.9 (#8895)

* chore: add release preparation workflow

* fix: address release workflow review feedback

AGENTS.md

@@ -51,6 +51,12 @@ ruff check .
 6. For path handling, use `pathlib.Path` instead of string paths, and use `astrbot.core.utils.path_utils` to get the AstrBot data and temp directory.
 7. When backend API routes, request/response schemas, or OpenAPI definitions change, regenerate the frontend API client by running `cd dashboard && pnpm generate:api`.
 
+### KISS and First Principles
+
+Follow the KISS principle and reason from first principles during development. Start by identifying the real problem, required behavior, and smallest useful change before adding code. Do not pile on features, configuration switches, abstractions, dependencies, or compatibility layers unless they directly solve the current problem and have clear evidence of need.
+
+Prefer the simplest implementation that is correct, maintainable, and consistent with the existing codebase. If a broader design seems attractive, reduce it to the essential behavior needed now and leave optional expansion for a later, explicit requirement.
+
 ### No Unnecessary Helpers
 
 Prioritize inline implementation over abstraction. Avoid over-engineering and do not create helper functions unless absolutely necessary.
@@ -94,7 +100,34 @@ def calculate_metrics(user_id: int, force_refresh: bool = False) -> dict:
 
 ## Release versions
 
-1. Replace current version name to specific version name.
-2. Write changelog in `changelogs/`, you can refer to the full commit messages between the latest tag to the latest commit.
-3. Make and push a commit into master branch with message format like: `chore: bump version to 4.25.0`
-4. Create a tag and push the tag. For example: `git tag v4.25.0 && git push origin v4.25.0`
+Use a short-lived `release/*` branch for each release. The release branch is the stabilization area for version bumps, changelog updates, release-blocking fixes, and final validation only. Do not add unrelated features or broad refactors to a release branch.
+
+Prepare a release from a clean worktree with:
+
+```bash
+uv run python scripts/prepare_release.py 4.25.0
+```
+
+The script updates `pyproject.toml`, creates `changelogs/v4.25.0.md`, runs the required Python checks, and prints the remaining steps. Use these flags when needed:
+
+```bash
+uv run python scripts/prepare_release.py 4.25.0 --generate-api-client
+uv run python scripts/prepare_release.py 4.25.0 --dashboard-build
+uv run python scripts/prepare_release.py 4.25.0 --commit --push
+```
+
+Open a PR from `release/4.25.0` to `master`. The PR title must use the conventional commit format, for example `chore: bump version to 4.25.0`. After the release PR is merged, create and push the tag from the updated `master` branch so the tag points to the exact code that was merged:
+
+```bash
+git checkout master
+git pull --ff-only origin master
+git tag v4.25.0
+git push origin v4.25.0
+```
+
+For one-off release candidate branches, delete the release branch after the tag is pushed and verified. For maintained release lines, use a branch such as `release/4.25` and keep it until that line reaches EOL.
+
+```bash
+git branch -d release/4.25.0
+git push origin --delete release/4.25.0
+```

astrbot/core/agent/mcp_client.py

@@ -9,6 +9,7 @@ from datetime import timedelta
 from pathlib import Path, PureWindowsPath
 from typing import Any, Generic
 
+import httpx
 from tenacity import (
     before_sleep_log,
     retry,
@@ -102,12 +103,22 @@ except (ModuleNotFoundError, ImportError):
         "Warning: Missing 'mcp' dependency, MCP services will be unavailable."
     )
 
+streamable_http_client_legacy = None
+streamable_http_client = None
+
 try:
-    from mcp.client.streamable_http import streamablehttp_client
-except (ModuleNotFoundError, ImportError):
-    logger.warning(
-        "Warning: Missing 'mcp' dependency or MCP library version too old, Streamable HTTP connection unavailable.",
+    from mcp.client.streamable_http import (
+        streamablehttp_client as streamable_http_client_legacy,
     )
+except (ModuleNotFoundError, ImportError):
+    try:
+        from mcp.client.streamable_http import (
+            streamable_http_client as streamable_http_client,
+        )
+    except (ModuleNotFoundError, ImportError):
+        logger.warning(
+            "Warning: Missing 'mcp' dependency or MCP library version too old, Streamable HTTP connection unavailable.",
+        )
 
 
 def _prepare_config(config: dict) -> dict:
@@ -459,17 +470,38 @@ class MCPClient:
                     ),
                 )
             else:
-                timeout = timedelta(seconds=cfg.get("timeout", 30))
-                sse_read_timeout = timedelta(
-                    seconds=cfg.get("sse_read_timeout", 60 * 5),
-                )
-                self._streams_context = streamablehttp_client(
-                    url=cfg["url"],
-                    headers=cfg.get("headers", {}),
-                    timeout=timeout,
-                    sse_read_timeout=sse_read_timeout,
-                    terminate_on_close=cfg.get("terminate_on_close", True),
-                )
+                timeout_seconds = cfg.get("timeout", 30)
+                sse_read_timeout_seconds = cfg.get("sse_read_timeout", 60 * 5)
+                if streamable_http_client_legacy:
+                    timeout = timedelta(seconds=timeout_seconds)
+                    sse_read_timeout = timedelta(seconds=sse_read_timeout_seconds)
+                    self._streams_context = streamable_http_client_legacy(
+                        url=cfg["url"],
+                        headers=cfg.get("headers", {}),
+                        timeout=timeout,
+                        sse_read_timeout=sse_read_timeout,
+                        terminate_on_close=cfg.get("terminate_on_close", True),
+                    )
+                elif streamable_http_client:
+                    http_client = await self.exit_stack.enter_async_context(
+                        httpx.AsyncClient(
+                            headers=cfg.get("headers", {}),
+                            timeout=httpx.Timeout(
+                                timeout_seconds,
+                                read=sse_read_timeout_seconds,
+                            ),
+                            follow_redirects=True,
+                        ),
+                    )
+                    self._streams_context = streamable_http_client(
+                        url=cfg["url"],
+                        http_client=http_client,
+                        terminate_on_close=cfg.get("terminate_on_close", True),
+                    )
+                else:
+                    raise RuntimeError(
+                        "Streamable HTTP transport is not available in the installed MCP library version."
+                    )
                 read_s, write_s, _ = await self.exit_stack.enter_async_context(
                     self._streams_context,
                 )

astrbot/core/agent/runners/tool_loop_agent_runner.py

@@ -224,6 +224,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
         custom_compressor: ContextCompressor | None = None,
         tool_schema_mode: str | None = "full",
         fallback_providers: list[Provider] | None = None,
+        request_max_retries: int | None = None,
         tool_result_overflow_dir: str | None = None,
         read_tool: FunctionTool | None = None,
         **kwargs: T.Any,
@@ -237,6 +238,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
         self.truncate_turns = truncate_turns
         self.custom_token_counter = custom_token_counter
         self.custom_compressor = custom_compressor
+        self.request_max_retries = request_max_retries
         self.tool_result_overflow_dir = tool_result_overflow_dir
         self.read_tool = read_tool
         self._tool_result_token_counter = EstimateTokenCounter()
@@ -463,6 +465,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
             "session_id": self.req.session_id,
             "extra_user_content_parts": self.req.extra_user_content_parts,  # list[ContentPart]
             "abort_signal": self._abort_signal,
+            "request_max_retries": self.request_max_retries,
         }
         if include_model:
             # For primary provider we keep explicit model selection if provided.
@@ -1305,6 +1308,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
                     extra_user_content_parts=self.req.extra_user_content_parts,
                     # tool_choice="required",
                     abort_signal=self._abort_signal,
+                    request_max_retries=self.request_max_retries,
                 )
                 if requery_resp:
                     llm_resp = requery_resp
@@ -1331,6 +1335,7 @@ class ToolLoopAgentRunner(BaseAgentRunner[TContext]):
                         extra_user_content_parts=self.req.extra_user_content_parts,
                         # tool_choice="required",
                         abort_signal=self._abort_signal,
+                        request_max_retries=self.request_max_retries,
                     )
                     if repair_resp:
                         llm_resp = repair_resp

astrbot/core/astr_main_agent.py

@@ -1629,6 +1629,7 @@ async def build_main_agent(
         enforce_max_turns=config.max_context_length,
         tool_schema_mode=config.tool_schema_mode,
         fallback_providers=fallback_providers,
+        request_max_retries=config.provider_settings.get("request_max_retries", 5),
         tool_result_overflow_dir=(
             get_astrbot_system_tmp_path()
             if req.func_tool and req.func_tool.get_tool("astrbot_file_read_tool")

astrbot/core/config/default.py

@@ -1,11 +1,39 @@
 """如需修改配置，请在 `data/cmd_config.json` 中修改或者在管理面板中可视化修改。"""
 
 import os
+import re
+from importlib.metadata import PackageNotFoundError
+from importlib.metadata import version as package_version
+from pathlib import Path
 
 from astrbot.core.computer.booters.cua_defaults import CUA_DEFAULT_CONFIG
 from astrbot.core.utils.astrbot_path import get_astrbot_data_path
+from astrbot.core.utils.toml_parser import read_pyproject_project_version
+
+try:
+    import tomllib
+except ModuleNotFoundError:
+    # <= Python 3.10 compatibility
+    tomllib = None
+
+try:
+    pyproject_path = Path(__file__).resolve().parents[3] / "pyproject.toml"
+    if tomllib is None:
+        VERSION = read_pyproject_project_version(pyproject_path)
+    else:
+        with pyproject_path.open("rb") as f:
+            VERSION = tomllib.load(f)["project"]["version"]
+except (FileNotFoundError, IndexError, KeyError, TypeError, ValueError):
+    try:
+        VERSION = package_version("astrbot")  # PEP 440 version style, e.g. 1.2.3a4
+        match = re.match(r"^(\d+(?:\.\d+)*)(a|b|rc)(\d+)$", VERSION)
+        if match:
+            release, prerelease, number = match.groups()
+            prerelease = {"a": "alpha", "b": "beta", "rc": "rc"}[prerelease]
+            VERSION = f"{release}-{prerelease}.{number}"
+    except PackageNotFoundError:
+        VERSION = "0.0.0"
 
-VERSION = "4.26.0-beta.8"
 DB_PATH = os.path.join(get_astrbot_data_path(), "data_v4.db")
 PERSONAL_WECHAT_CONFIG_METADATA = {
     "weixin_oc_base_url": {
@@ -101,6 +129,7 @@ DEFAULT_CONFIG = {
         "enable": True,
         "default_provider_id": "",
         "fallback_chat_models": [],
+        "request_max_retries": 5,
         "default_image_caption_provider_id": "",
         "image_caption_prompt": "Please describe the image using Chinese.",
         "provider_pool": ["*"],  # "*" 表示使用所有可用的提供者
@@ -2808,6 +2837,9 @@ CONFIG_METADATA_2 = {
                         "type": "list",
                         "items": {"type": "string"},
                     },
+                    "request_max_retries": {
+                        "type": "int",
+                    },
                     "wake_prefix": {
                         "type": "string",
                     },
@@ -3167,6 +3199,11 @@ CONFIG_METADATA_3 = {
                         "_special": "select_providers",
                         "hint": "主聊天模型请求失败时，按顺序切换到这些模型。",
                     },
+                    "provider_settings.request_max_retries": {
+                        "description": "请求最大重试次数",
+                        "type": "int",
+                        "hint": "单次模型请求遇到可重试错误时的最大尝试次数。",
+                    },
                     "provider_settings.default_image_caption_provider_id": {
                         "description": "默认图片转述模型",
                         "type": "string",

astrbot/core/provider/provider.py

@@ -106,6 +106,7 @@ class Provider(AbstractProvider):
         model: str | None = None,
         extra_user_content_parts: list[ContentPart] | None = None,
         tool_choice: Literal["auto", "required"] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ) -> LLMResponse:
         """获得 LLM 的文本对话结果。会使用当前的模型进行对话。
@@ -120,6 +121,7 @@ class Provider(AbstractProvider):
             contexts: 上下文，和 prompt 二选一使用
             tool_calls_result: 回传给 LLM 的工具调用结果。参考: https://platform.openai.com/docs/guides/function-calling
             extra_user_content_parts: 额外的内容块列表，用于在用户消息后添加额外的文本块（如系统提醒、指令等）
+            request_max_retries: 可重试请求错误的最大尝试次数，包含首次请求。
             kwargs: 其他参数
 
         Notes:
@@ -142,6 +144,7 @@ class Provider(AbstractProvider):
         tool_calls_result: ToolCallsResult | list[ToolCallsResult] | None = None,
         model: str | None = None,
         tool_choice: Literal["auto", "required"] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ) -> AsyncGenerator[LLMResponse, None]:
         """获得 LLM 的流式文本对话结果。会使用当前的模型进行对话。在生成的最后会返回一次完整的结果。
@@ -155,6 +158,7 @@ class Provider(AbstractProvider):
             tool_choice: 工具调用策略，`auto` 表示由模型自行决定，`required` 表示要求模型必须调用工具
             contexts: 上下文，和 prompt 二选一使用
             tool_calls_result: 回传给 LLM 的工具调用结果。参考: https://platform.openai.com/docs/guides/function-calling
+            request_max_retries: 可重试请求错误的最大尝试次数，包含首次请求。
             kwargs: 其他参数
 
         Notes:

astrbot/core/provider/sources/anthropic_source.py

@@ -27,6 +27,7 @@ from astrbot.core.utils.network_utils import (
 )
 
 from ..register import register_provider_adapter
+from .request_retry import retry_provider_request, retry_provider_request_context
 
 
 @register_provider_adapter(
@@ -353,7 +354,13 @@ class ProviderAnthropic(Provider):
         logger.warning(f"未知的 tool_choice 值: {tool_choice}，已回退为 'auto'")
         return {"type": "auto"}
 
-    async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
+    async def _query(
+        self,
+        payloads: dict,
+        tools: ToolSet | None,
+        *,
+        request_max_retries: int | None = None,
+    ) -> LLMResponse:
         if tools:
             if tool_list := tools.get_func_desc_anthropic_style():
                 payloads["tools"] = tool_list
@@ -368,8 +375,12 @@ class ProviderAnthropic(Provider):
         self._apply_thinking_config(payloads)
 
         try:
-            completion = await self.client.messages.create(
-                **payloads, stream=False, extra_body=extra_body
+            completion = await retry_provider_request(
+                "Anthropic",
+                lambda: self.client.messages.create(
+                    **payloads, stream=False, extra_body=extra_body
+                ),
+                max_attempts=request_max_retries,
             )
         except httpx.RequestError as e:
             proxy = self.provider_config.get("proxy", "")
@@ -438,6 +449,8 @@ class ProviderAnthropic(Provider):
         self,
         payloads: dict,
         tools: ToolSet | None,
+        *,
+        request_max_retries: int | None = None,
     ) -> AsyncGenerator[LLMResponse, None]:
         if tools:
             if tool_list := tools.get_func_desc_anthropic_style():
@@ -461,8 +474,10 @@ class ProviderAnthropic(Provider):
             payloads["max_tokens"] = 65536
         self._apply_thinking_config(payloads)
 
-        async with self.client.messages.stream(
-            **payloads, extra_body=extra_body
+        async with retry_provider_request_context(
+            "Anthropic",
+            lambda: self.client.messages.stream(**payloads, extra_body=extra_body),
+            max_attempts=request_max_retries,
         ) as stream:
             assert isinstance(stream, anthropic.AsyncMessageStream)
             async for event in stream:
@@ -601,6 +616,7 @@ class ProviderAnthropic(Provider):
         model=None,
         extra_user_content_parts=None,
         tool_choice: Literal["auto", "any", "tool", "none"] | dict[str, str] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ) -> LLMResponse:
         if contexts is None:
@@ -650,7 +666,11 @@ class ProviderAnthropic(Provider):
 
         llm_response = None
         try:
-            llm_response = await self._query(payloads, func_tool)
+            llm_response = await self._query(
+                payloads,
+                func_tool,
+                request_max_retries=request_max_retries,
+            )
         except Exception as e:
             raise e
 
@@ -669,6 +689,7 @@ class ProviderAnthropic(Provider):
         model=None,
         extra_user_content_parts=None,
         tool_choice: Literal["auto", "any", "tool", "none"] | dict[str, str] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ):
         if contexts is None:
@@ -715,7 +736,11 @@ class ProviderAnthropic(Provider):
                 else system_prompt
             )
 
-        async for llm_response in self._query_stream(payloads, func_tool):
+        async for llm_response in self._query_stream(
+            payloads,
+            func_tool,
+            request_max_retries=request_max_retries,
+        ):
             yield llm_response
 
     def _detect_image_mime_type(self, data: bytes) -> str:
@@ -827,7 +852,10 @@ class ProviderAnthropic(Provider):
 
     async def get_models(self) -> list[str]:
         models_str = []
-        models = await self.client.models.list()
+        models = await retry_provider_request(
+            "Anthropic",
+            lambda: self.client.models.list(),
+        )
         models = sorted(models.data, key=lambda x: x.id)
         for model in models:
             models_str.append(model.id)

astrbot/core/provider/sources/gemini_source.py

@@ -26,6 +26,7 @@ from astrbot.core.utils.media_utils import (
 from astrbot.core.utils.network_utils import is_connection_error, log_connection_failure
 
 from ..register import register_provider_adapter
+from .request_retry import retry_provider_request
 
 
 class SuppressNonTextPartsWarning(logging.Filter):
@@ -577,7 +578,13 @@ class ProviderGoogleGenAI(Provider):
             )
         return chain_result
 
-    async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
+    async def _query(
+        self,
+        payloads: dict,
+        tools: ToolSet | None,
+        *,
+        request_max_retries: int | None = None,
+    ) -> LLMResponse:
         """非流式请求 Gemini API"""
         system_instruction = next(
             (msg["content"] for msg in payloads["messages"] if msg["role"] == "system"),
@@ -604,10 +611,14 @@ class ProviderGoogleGenAI(Provider):
                     modalities,
                     temperature,
                 )
-                result = await self.client.models.generate_content(
-                    model=model,
-                    contents=cast(types.ContentListUnion, conversation),
-                    config=config,
+                result = await retry_provider_request(
+                    "Gemini",
+                    lambda: self.client.models.generate_content(
+                        model=model,
+                        contents=cast(types.ContentListUnion, conversation),
+                        config=config,
+                    ),
+                    max_attempts=request_max_retries,
                 )
                 logger.debug(f"genai result: {result}")
 
@@ -672,6 +683,8 @@ class ProviderGoogleGenAI(Provider):
         self,
         payloads: dict,
         tools: ToolSet | None,
+        *,
+        request_max_retries: int | None = None,
     ) -> AsyncGenerator[LLMResponse, None]:
         """流式请求 Gemini API"""
         system_instruction = next(
@@ -690,10 +703,14 @@ class ProviderGoogleGenAI(Provider):
                     payloads.get("tool_choice", "auto"),
                     system_instruction,
                 )
-                result = await self.client.models.generate_content_stream(
-                    model=model,
-                    contents=cast(types.ContentListUnion, conversation),
-                    config=config,
+                result = await retry_provider_request(
+                    "Gemini",
+                    lambda: self.client.models.generate_content_stream(
+                        model=model,
+                        contents=cast(types.ContentListUnion, conversation),
+                        config=config,
+                    ),
+                    max_attempts=request_max_retries,
                 )
                 break
             except APIError as e:
@@ -809,6 +826,7 @@ class ProviderGoogleGenAI(Provider):
         model=None,
         extra_user_content_parts=None,
         tool_choice: Literal["auto", "required"] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ) -> LLMResponse:
         if contexts is None:
@@ -850,7 +868,11 @@ class ProviderGoogleGenAI(Provider):
 
         for _ in range(retry):
             try:
-                return await self._query(payloads, func_tool)
+                return await self._query(
+                    payloads,
+                    func_tool,
+                    request_max_retries=request_max_retries,
+                )
             except APIError as e:
                 if await self._handle_api_error(e, keys):
                     continue
@@ -871,6 +893,7 @@ class ProviderGoogleGenAI(Provider):
         model=None,
         extra_user_content_parts=None,
         tool_choice: Literal["auto", "required"] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ) -> AsyncGenerator[LLMResponse, None]:
         if contexts is None:
@@ -912,7 +935,11 @@ class ProviderGoogleGenAI(Provider):
 
         for _ in range(retry):
             try:
-                async for response in self._query_stream(payloads, func_tool):
+                async for response in self._query_stream(
+                    payloads,
+                    func_tool,
+                    request_max_retries=request_max_retries,
+                ):
                     yield response
                 break
             except APIError as e:
@@ -922,7 +949,10 @@ class ProviderGoogleGenAI(Provider):
 
     async def get_models(self):
         try:
-            models = await self.client.models.list()
+            models = await retry_provider_request(
+                "Gemini",
+                lambda: self.client.models.list(),
+            )
             return [
                 m.name.replace("models/", "")
                 for m in models

astrbot/core/provider/sources/openai_source.py

@@ -41,6 +41,7 @@ from astrbot.core.utils.network_utils import (
 from astrbot.core.utils.string_utils import normalize_and_dedupe_strings
 
 from ..register import register_provider_adapter
+from .request_retry import retry_provider_request
 
 
 @register_provider_adapter(
@@ -420,7 +421,10 @@ class ProviderOpenAIOfficial(Provider):
     async def get_models(self):
         try:
             models_str = []
-            models = await self.client.models.list()
+            models = await retry_provider_request(
+                "OpenAI",
+                lambda: self.client.models.list(),
+            )
             models = sorted(models.data, key=lambda x: x.id)
             for model in models:
                 models_str.append(model.id)
@@ -465,7 +469,13 @@ class ProviderOpenAIOfficial(Provider):
 
         payloads["messages"] = cleaned
 
-    async def _query(self, payloads: dict, tools: ToolSet | None) -> LLMResponse:
+    async def _query(
+        self,
+        payloads: dict,
+        tools: ToolSet | None,
+        *,
+        request_max_retries: int | None = None,
+    ) -> LLMResponse:
         if tools:
             model = payloads.get("model", "").lower()
             omit_empty_param_field = "gemini" in model
@@ -496,10 +506,14 @@ class ProviderOpenAIOfficial(Provider):
 
         self._sanitize_assistant_messages(payloads)
 
-        completion = await self.client.chat.completions.create(
-            **payloads,
-            stream=False,
-            extra_body=extra_body,
+        completion = await retry_provider_request(
+            "OpenAI",
+            lambda: self.client.chat.completions.create(
+                **payloads,
+                stream=False,
+                extra_body=extra_body,
+            ),
+            max_attempts=request_max_retries,
         )
 
         if not isinstance(completion, ChatCompletion):
@@ -517,6 +531,8 @@ class ProviderOpenAIOfficial(Provider):
         self,
         payloads: dict,
         tools: ToolSet | None,
+        *,
+        request_max_retries: int | None = None,
     ) -> AsyncGenerator[LLMResponse, None]:
         """流式查询API，逐步返回结果"""
         if tools:
@@ -548,11 +564,15 @@ class ProviderOpenAIOfficial(Provider):
 
         self._sanitize_assistant_messages(payloads)
 
-        stream = await self.client.chat.completions.create(
-            **payloads,
-            stream=True,
-            extra_body=extra_body,
-            stream_options={"include_usage": True},
+        stream = await retry_provider_request(
+            "OpenAI",
+            lambda: self.client.chat.completions.create(
+                **payloads,
+                stream=True,
+                extra_body=extra_body,
+                stream_options={"include_usage": True},
+            ),
+            max_attempts=request_max_retries,
         )
 
         llm_response = LLMResponse("assistant", is_chunk=True)
@@ -1104,6 +1124,7 @@ class ProviderOpenAIOfficial(Provider):
         model=None,
         extra_user_content_parts=None,
         tool_choice: Literal["auto", "required"] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ) -> LLMResponse:
         payloads, context_query = await self._prepare_chat_payload(
@@ -1131,7 +1152,11 @@ class ProviderOpenAIOfficial(Provider):
         for retry_cnt in range(max_retries):
             try:
                 self.client.api_key = chosen_key
-                llm_response = await self._query(payloads, func_tool)
+                llm_response = await self._query(
+                    payloads,
+                    func_tool,
+                    request_max_retries=request_max_retries,
+                )
                 break
             except Exception as e:
                 last_exception = e
@@ -1176,6 +1201,7 @@ class ProviderOpenAIOfficial(Provider):
         tool_calls_result=None,
         model=None,
         tool_choice: Literal["auto", "required"] = "auto",
+        request_max_retries: int | None = None,
         **kwargs,
     ) -> AsyncGenerator[LLMResponse, None]:
         """流式对话，与服务商交互并逐步返回结果"""
@@ -1202,7 +1228,11 @@ class ProviderOpenAIOfficial(Provider):
         for retry_cnt in range(max_retries):
             try:
                 self.client.api_key = chosen_key
-                async for response in self._query_stream(payloads, func_tool):
+                async for response in self._query_stream(
+                    payloads,
+                    func_tool,
+                    request_max_retries=request_max_retries,
+                ):
                     yield response
                 break
             except Exception as e:

astrbot/core/provider/sources/request_retry.py

@@ -0,0 +1,163 @@
+from collections.abc import AsyncIterator, Awaitable, Callable
+from contextlib import AbstractAsyncContextManager, asynccontextmanager
+from typing import TypeVar
+
+from tenacity import (
+    AsyncRetrying,
+    RetryCallState,
+    retry_if_exception,
+    stop_after_attempt,
+    wait_exponential,
+)
+
+from astrbot import logger
+from astrbot.core.utils.config_number import coerce_int_config
+from astrbot.core.utils.network_utils import is_connection_error
+
+T = TypeVar("T")
+
+REQUEST_RETRY_ATTEMPTS = 5  # default value
+REQUEST_RETRY_WAIT_MIN_S = 0.2
+REQUEST_RETRY_WAIT_MAX_S = 30
+REQUEST_RETRY_STATUS_CODES = {408, 409, 429, 500, 502, 503, 504, 529}
+
+
+def _get_status_code(error: BaseException) -> int | None:
+    for attr in ("status_code", "status", "code"):
+        value = getattr(error, attr, None)
+        if isinstance(value, int):
+            return value
+
+    response = getattr(error, "response", None)
+    if response is not None:
+        status_code = getattr(response, "status_code", None)
+        if isinstance(status_code, int):
+            return status_code
+
+    return None
+
+
+def _is_retryable_provider_request_error(
+    error: BaseException,
+    *,
+    retry_rate_limits: bool,
+) -> bool:
+    if is_connection_error(error):
+        return True
+
+    error_type_name = type(error).__name__
+    if error_type_name in {"APIConnectionError", "APITimeoutError"}:
+        return True
+
+    status_code = _get_status_code(error)
+    if status_code is None:
+        return False
+
+    if status_code == 429 and not retry_rate_limits:
+        return False
+
+    return status_code in REQUEST_RETRY_STATUS_CODES or 500 <= status_code <= 599
+
+
+def _log_retry(
+    provider_label: str,
+    retry_state: RetryCallState,
+    max_attempts: int,
+) -> None:
+    error = retry_state.outcome.exception() if retry_state.outcome else None
+    logger.warning(
+        f"[{provider_label}] Request failed with retryable error; "
+        f"retrying ({retry_state.attempt_number + 1}/{max_attempts}): "
+        f"{error}"
+    )
+
+
+def _build_retrying(
+    provider_label: str,
+    *,
+    retry_rate_limits: bool,
+    max_attempts: int | None = None,
+) -> AsyncRetrying:
+    max_attempts = coerce_int_config(
+        max_attempts if max_attempts is not None else REQUEST_RETRY_ATTEMPTS,
+        default=REQUEST_RETRY_ATTEMPTS,
+        min_value=1,
+        field_name="request_max_retries",
+        source=provider_label,
+    )
+
+    return AsyncRetrying(
+        retry=retry_if_exception(
+            lambda error: _is_retryable_provider_request_error(
+                error,
+                retry_rate_limits=retry_rate_limits,
+            )
+        ),
+        stop=stop_after_attempt(max_attempts),
+        wait=wait_exponential(
+            multiplier=1,
+            min=REQUEST_RETRY_WAIT_MIN_S,
+            max=REQUEST_RETRY_WAIT_MAX_S,
+        ),
+        before_sleep=lambda retry_state: _log_retry(
+            provider_label,
+            retry_state,
+            max_attempts,
+        ),
+        reraise=True,
+    )
+
+
+async def retry_provider_request(
+    provider_label: str,
+    request_factory: Callable[[], Awaitable[T]],
+    *,
+    retry_rate_limits: bool = True,
+    max_attempts: int | None = None,
+) -> T:
+    retrying = _build_retrying(
+        provider_label,
+        retry_rate_limits=retry_rate_limits,
+        max_attempts=max_attempts,
+    )
+
+    async for attempt in retrying:
+        with attempt:
+            return await request_factory()
+
+    raise RuntimeError("Provider request retry loop exited unexpectedly.")
+
+
+@asynccontextmanager
+async def retry_provider_request_context(
+    provider_label: str,
+    context_manager_factory: Callable[[], AbstractAsyncContextManager[T]],
+    *,
+    retry_rate_limits: bool = True,
+    max_attempts: int | None = None,
+) -> AsyncIterator[T]:
+    manager: AbstractAsyncContextManager[T] | None = None
+
+    async def _enter_context() -> T:
+        nonlocal manager
+        manager = context_manager_factory()
+        return await manager.__aenter__()
+
+    value = await retry_provider_request(
+        provider_label,
+        _enter_context,
+        retry_rate_limits=retry_rate_limits,
+        max_attempts=max_attempts,
+    )
+
+    if manager is None:
+        raise RuntimeError("Provider request context was not created.")
+
+    try:
+        yield value
+    except BaseException as error:
+        if await manager.__aexit__(type(error), error, error.__traceback__):
+            return
+        raise
+    else:
+        await manager.__aexit__(None, None, None)

astrbot/core/utils/toml_parser.py

@@ -0,0 +1,184 @@
+"""Small TOML readers for bootstrapping paths without parser dependencies."""
+
+from pathlib import Path
+
+
+def _read_quoted_value(value: str, field_name: str) -> tuple[str, str]:
+    """Read one quoted TOML string value and return its tail.
+
+    Args:
+        value: Raw value text that starts with a quoted string.
+        field_name: Field name used in error messages.
+
+    Returns:
+        A tuple containing the unquoted string and the remaining text.
+
+    Raises:
+        ValueError: The value is not a supported quoted string.
+    """
+    value = value.strip()
+    if len(value) < 2 or value[0] not in ("'", '"'):
+        raise ValueError(f"Unsupported {field_name} value")
+
+    quote = value[0]
+    end_index = value.find(quote, 1)
+    if end_index == -1:
+        raise ValueError(f"Unterminated {field_name} string")
+
+    result = value[1:end_index]
+    if not result:
+        raise ValueError(f"Empty {field_name} value")
+    return result, value[end_index + 1 :].strip()
+
+
+def _read_dependency_array(raw_value: str) -> list[str]:
+    """Read a simple inline TOML string array.
+
+    Args:
+        raw_value: Raw dependency array text, including the surrounding brackets.
+
+    Returns:
+        Parsed dependency strings.
+
+    Raises:
+        ValueError: The array is missing brackets or contains unsupported entries.
+    """
+    value = raw_value.strip()
+    if not value.startswith("["):
+        raise ValueError("Unsupported project.dependencies value")
+
+    dependencies = []
+    value = value[1:].strip()
+    while value:
+        if value.startswith("]"):
+            tail = value[1:].strip()
+            if tail and not tail.startswith("#"):
+                raise ValueError("Unsupported content after project.dependencies")
+            return dependencies
+
+        dependency, tail = _read_quoted_value(value, "project.dependencies entry")
+        dependencies.append(dependency)
+
+        if tail.startswith(","):
+            value = tail[1:].strip()
+            continue
+        if tail.startswith("]"):
+            value = tail
+            continue
+        if tail:
+            raise ValueError("Unsupported content after project.dependencies entry")
+        raise ValueError("Unterminated project.dependencies array")
+
+    raise ValueError("Unterminated project.dependencies array")
+
+
+def read_pyproject_project_version(pyproject_path: Path) -> str:
+    """Read the project version from a pyproject.toml file.
+
+    Args:
+        pyproject_path: Path to the pyproject.toml file.
+
+    Returns:
+        The value of the project.version field.
+
+    Raises:
+        FileNotFoundError: The pyproject.toml file does not exist.
+        ValueError: The project.version field is missing or unsupported.
+    """
+    in_project_section = False
+    for raw_line in pyproject_path.read_text(encoding="utf-8").splitlines():
+        line = raw_line.strip()
+        if not line or line.startswith("#"):
+            continue
+
+        if line.startswith("[") and line.endswith("]"):
+            in_project_section = line == "[project]"
+            continue
+
+        if not in_project_section:
+            continue
+
+        key, separator, raw_value = line.partition("=")
+        if key.strip() != "version":
+            continue
+        if not separator:
+            raise ValueError("Missing value separator for project.version")
+
+        version, tail = _read_quoted_value(raw_value, "project.version")
+        if tail and not tail.startswith("#"):
+            raise ValueError("Unsupported content after project.version")
+        return version
+
+    raise ValueError("Missing project.version")
+
+
+def read_pyproject_project_dependencies(pyproject_path: Path) -> list[str]:
+    """Read project dependencies from a pyproject.toml file.
+
+    Args:
+        pyproject_path: Path to the pyproject.toml file.
+
+    Returns:
+        The values in the project.dependencies array.
+
+    Raises:
+        FileNotFoundError: The pyproject.toml file does not exist.
+        ValueError: The project.dependencies field is missing or unsupported.
+    """
+    dependencies = []
+    in_project_section = False
+    in_dependencies_array = False
+
+    for raw_line in pyproject_path.read_text(encoding="utf-8").splitlines():
+        line = raw_line.strip()
+        if not line or line.startswith("#"):
+            continue
+
+        if in_dependencies_array:
+            if line.startswith("]"):
+                tail = line[1:].strip()
+                if tail and not tail.startswith("#"):
+                    raise ValueError("Unsupported content after project.dependencies")
+                return dependencies
+
+            dependency, tail = _read_quoted_value(
+                line,
+                "project.dependencies entry",
+            )
+            if tail.startswith(","):
+                tail = tail[1:].strip()
+            if tail.startswith("]"):
+                tail = tail[1:].strip()
+                dependencies.append(dependency)
+                if tail and not tail.startswith("#"):
+                    raise ValueError("Unsupported content after project.dependencies")
+                return dependencies
+            if tail and not tail.startswith("#"):
+                raise ValueError("Unsupported content after project.dependencies entry")
+
+            dependencies.append(dependency)
+            continue
+
+        if line.startswith("[") and line.endswith("]"):
+            in_project_section = line == "[project]"
+            continue
+
+        if not in_project_section:
+            continue
+
+        key, separator, raw_value = line.partition("=")
+        if key.strip() != "dependencies":
+            continue
+        if not separator:
+            raise ValueError("Unsupported project.dependencies value")
+        raw_value = raw_value.strip()
+        if raw_value == "[" or raw_value.startswith("[ #"):
+            in_dependencies_array = True
+            continue
+        if raw_value.startswith("["):
+            return _read_dependency_array(raw_value)
+        raise ValueError("Unsupported project.dependencies value")
+
+    if in_dependencies_array:
+        raise ValueError("Unterminated project.dependencies array")
+    raise ValueError("Missing project.dependencies")

changelogs/v4.26.0-beta.9.md

@@ -0,0 +1,54 @@
+- [更新日志(简体中文)](#chinese)
+- [Changelog(English)](#english)
+
+<a id="chinese"></a>
+
+## What's Changed
+
+### 重点更新
+
+- 为 OpenAI、Gemini、Anthropic 等模型请求加入可配置的重试机制，并新增请求最大重试次数配置，提升临时网络错误与 5xx 服务端错误下的稳定性。([#8893](https://github.com/AstrBotDevs/AstrBot/pull/8893))
+- 新增托管 Core 包下载能力，并加强 Core 与 Dashboard 包下载归档校验。([#8888](https://github.com/AstrBotDevs/AstrBot/pull/8888))
+- 支持在请求中加载 workspace skills，并加固 workspace skill 发现流程。([#8884](https://github.com/AstrBotDevs/AstrBot/pull/8884))
+
+### 修复
+
+- 修复 OpenAPI 文件上传能力，恢复 `/api/v1/file` OpenAPI 暴露、文件范围 API Key 与相关文档/客户端产物。
+- 修复新版 MCP 中 Streamable HTTP client 重命名导致的兼容问题，并保持 `mcp` 依赖小于 2。
+- 加固人格工具边界，确保人格限定的工具范围在主 Agent 请求中正确生效。([#8786](https://github.com/AstrBotDevs/AstrBot/pull/8786))
+- 加强 Future Task 所有者校验，避免越权访问定时任务。([#8881](https://github.com/AstrBotDevs/AstrBot/pull/8881))
+- 在受限本地文件系统工具中拒绝 hardlink 文件，避免通过工作区 hardlink 别名读写允许目录外的文件。
+
+### 发布流程
+
+- 新增 `scripts/prepare_release.py`，统一 release 分支、版本号、changelog 与校验流程。([#8891](https://github.com/AstrBotDevs/AstrBot/pull/8891))
+
+### 文档
+
+- 明确 OpenAPI Chat 中 `username` 字段的身份含义。([#8880](https://github.com/AstrBotDevs/AstrBot/pull/8880))
+
+<a id="english"></a>
+
+## What's Changed (EN)
+
+### Highlights
+
+- Added configurable retry handling for OpenAI, Gemini, Anthropic, and related provider requests, including a maximum request retry setting to improve stability for transient network failures and 5xx server errors. ([#8893](https://github.com/AstrBotDevs/AstrBot/pull/8893))
+- Added hosted Core package downloads and strengthened archive validation for hosted Core and Dashboard packages. ([#8888](https://github.com/AstrBotDevs/AstrBot/pull/8888))
+- Added workspace skills support in requests and hardened workspace skill discovery. ([#8884](https://github.com/AstrBotDevs/AstrBot/pull/8884))
+
+### Bug Fixes
+
+- Restored OpenAPI file uploads by exposing `/api/v1/file`, enabling file-scoped API keys, and regenerating docs/client artifacts.
+- Fixed compatibility with the renamed MCP Streamable HTTP client while keeping the `mcp` dependency below 2.
+- Hardened persona tool boundaries so persona-restricted tool scopes are enforced correctly in main Agent requests. ([#8786](https://github.com/AstrBotDevs/AstrBot/pull/8786))
+- Enforced Future Task owner checks to prevent unauthorized scheduled-task access. ([#8881](https://github.com/AstrBotDevs/AstrBot/pull/8881))
+- Rejected hardlinked files in restricted local filesystem tools to prevent workspace hardlink aliases from reading or overwriting files outside allowed directories.
+
+### Release Process
+
+- Added `scripts/prepare_release.py` to standardize release branches, version bumps, changelog generation, and validation. ([#8891](https://github.com/AstrBotDevs/AstrBot/pull/8891))
+
+### Docs
+
+- Clarified the identity semantics of the `username` field in OpenAPI Chat. ([#8880](https://github.com/AstrBotDevs/AstrBot/pull/8880))

dashboard/src/i18n/locales/en-US/features/config-metadata.json

@@ -45,6 +45,10 @@
           "description": "Fallback chat model IDs",
           "hint": "When the primary chat model request fails, fallback to these chat models in order."
         },
+        "request_max_retries": {
+          "description": "Request Max Retries",
+          "hint": "Maximum attempts for a single model request when retryable errors occur."
+        },
         "default_image_caption_provider_id": {
           "description": "Default Image Caption Model",
           "hint": "Leave empty to disable; useful for non-multimodal models"

dashboard/src/i18n/locales/ru-RU/features/config-metadata.json

@@ -45,6 +45,10 @@
                     "description": "Резервные модели чата (ID)",
                     "hint": "Если текущая модель недоступна, запрос будет перенаправлен на эти модели по порядку."
                 },
+                "request_max_retries": {
+                    "description": "Максимум повторов запроса",
+                    "hint": "Максимальное число попыток для одного запроса модели при повторяемых ошибках."
+                },
                 "default_image_caption_provider_id": {
                     "description": "Модель описания изображений",
                     "hint": "Оставьте пустым для отключения; полезно для моделей без поддержки мультимодальности"

dashboard/src/i18n/locales/zh-CN/features/config-metadata.json

@@ -45,6 +45,10 @@
           "description": "回退对话模型列表",
           "hint": "主对话模型请求失败时，按顺序切换到这些对话模型。"
         },
+        "request_max_retries": {
+          "description": "请求最大重试次数",
+          "hint": "单次模型请求遇到可重试错误时的最大尝试次数。"
+        },
         "default_image_caption_provider_id": {
           "description": "默认图片转述模型",
           "hint": "留空代表不使用,可用于非多模态模型"

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "AstrBot"
-version = "4.26.0-beta.8"
+version = "4.26.0-beta.9"
 description = "Easy-to-use multi-platform LLM chatbot and development framework"
 readme = "README.md"
 license = { text = "AGPL-3.0-or-later" }
@@ -29,7 +29,7 @@ dependencies = [
   "google-genai>=1.56.0",
   "httpx[socks]>=0.28.1",
   "lark-oapi>=1.4.15",
-  "mcp>=1.8.0",
+  "mcp>=1.8.0,<2",
   "openai>=1.78.0",
   "ormsgpack>=1.9.1",
   "pillow>=11.2.1",

requirements.txt

@@ -18,7 +18,7 @@ filelock>=3.18.0
 google-genai>=1.56.0
 httpx[socks]>=0.28.1
 lark-oapi>=1.4.15
-mcp>=1.8.0
+mcp>=1.8.0,<2
 openai>=1.78.0
 ormsgpack>=1.9.1
 pillow>=11.2.1

scripts/prepare_release.py

@@ -0,0 +1,431 @@
+#!/usr/bin/env python3
+"""Prepare an AstrBot release branch and release metadata."""
+
+from __future__ import annotations
+
+import argparse
+import re
+import subprocess
+import sys
+from pathlib import Path
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+VERSION_PATTERN = re.compile(r"^\d+\.\d+\.\d+(?:[-+._a-zA-Z0-9]+)?$")
+
+
+class ReleaseError(RuntimeError):
+    """Error raised when a release preparation step cannot continue."""
+
+
+def run_command(
+    args: list[str],
+    *,
+    cwd: Path = REPO_ROOT,
+    capture_output: bool = False,
+) -> str:
+    """Run a command and return captured stdout when requested.
+
+    Args:
+        args: Command and arguments to run.
+        cwd: Working directory for the command.
+        capture_output: Whether to capture and return stdout instead of streaming it.
+
+    Returns:
+        Captured stdout without surrounding whitespace when capture_output is true;
+        otherwise an empty string.
+
+    Raises:
+        ReleaseError: The command is missing or exits with a non-zero status.
+    """
+    printable = " ".join(args)
+    print(f"$ {printable}")
+    try:
+        if capture_output:
+            result = subprocess.run(
+                args,
+                cwd=cwd,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+            return result.stdout.strip()
+
+        subprocess.run(args, cwd=cwd, check=True)
+        return ""
+    except FileNotFoundError as exc:
+        raise ReleaseError(f"Command not found: {args[0]}") from exc
+    except subprocess.CalledProcessError as exc:
+        if capture_output and exc.stderr:
+            print(exc.stderr.strip(), file=sys.stderr)
+        raise ReleaseError(f"Command failed ({exc.returncode}): {printable}") from exc
+
+
+def git(args: list[str], *, capture_output: bool = False) -> str:
+    """Run a git command in the repository root.
+
+    Args:
+        args: Arguments to pass after `git`.
+        capture_output: Whether to capture and return stdout.
+
+    Returns:
+        Captured stdout when capture_output is true; otherwise an empty string.
+
+    Raises:
+        ReleaseError: Git exits with a non-zero status.
+    """
+    return run_command(["git", *args], capture_output=capture_output)
+
+
+def ensure_clean_worktree() -> None:
+    """Ensure the release starts from a clean worktree.
+
+    Raises:
+        ReleaseError: The repository contains tracked or untracked changes.
+    """
+    status = git(["status", "--porcelain"], capture_output=True)
+    if status:
+        raise ReleaseError(
+            "Working tree must be clean before preparing a release.\n"
+            "Commit, stash, or remove these changes first:\n"
+            f"{status}"
+        )
+
+
+def validate_version(version: str) -> str:
+    """Validate a release version string.
+
+    Args:
+        version: Version string without the leading tag prefix.
+
+    Returns:
+        The validated version string.
+
+    Raises:
+        ReleaseError: The version is empty, starts with `v`, or has an unsupported
+            shape.
+    """
+    if version.startswith("v"):
+        raise ReleaseError(
+            "Pass the version without the tag prefix, for example 4.25.0"
+        )
+    if not VERSION_PATTERN.fullmatch(version):
+        raise ReleaseError(
+            "Unsupported version format. Expected a value like 4.25.0 or 4.26.0-beta.8"
+        )
+    return version
+
+
+def latest_tag() -> str:
+    """Return the most recent reachable tag, if one exists.
+
+    Returns:
+        The latest tag name, or an empty string when the repository has no tags.
+    """
+    try:
+        return git(["describe", "--tags", "--abbrev=0"], capture_output=True)
+    except ReleaseError:
+        return ""
+
+
+def release_commits(tag: str) -> list[str]:
+    """Read commit subjects for the release range.
+
+    Args:
+        tag: Latest tag to use as the lower bound. When empty, all reachable
+            commits are considered.
+
+    Returns:
+        Commit subjects formatted for changelog draft entries.
+
+    Raises:
+        ReleaseError: Git log fails.
+    """
+    log_range = f"{tag}..HEAD" if tag else "HEAD"
+    output = git(
+        ["log", "--reverse", "--pretty=format:%s (%h)", log_range],
+        capture_output=True,
+    )
+    return [line for line in output.splitlines() if line.strip()]
+
+
+def update_pyproject_version(version: str) -> Path:
+    """Update `[project].version` in pyproject.toml.
+
+    Args:
+        version: Release version to write.
+
+    Returns:
+        Path to the modified pyproject.toml file.
+
+    Raises:
+        ReleaseError: The project version field cannot be found or parsed.
+    """
+    pyproject_path = REPO_ROOT / "pyproject.toml"
+    lines = pyproject_path.read_text(encoding="utf-8").splitlines(keepends=True)
+    in_project_section = False
+
+    for index, line in enumerate(lines):
+        stripped = line.strip()
+        if stripped.startswith("[") and stripped.endswith("]"):
+            in_project_section = stripped == "[project]"
+            continue
+        if not in_project_section:
+            continue
+
+        key, separator, _raw_value = stripped.partition("=")
+        if key.strip() != "version":
+            continue
+        if not separator:
+            raise ReleaseError("Unsupported pyproject.toml project.version format")
+
+        match = re.match(
+            r"^(\s*version\s*=\s*)([\"'])(.*?)(\2)(\s*(?:#.*)?)(\n?)$",
+            line,
+        )
+        if not match:
+            raise ReleaseError("Unsupported pyproject.toml project.version format")
+
+        prefix, quote, _current, _closing_quote, suffix, newline = match.groups()
+        lines[index] = f"{prefix}{quote}{version}{quote}{suffix}{newline}"
+        pyproject_path.write_text("".join(lines), encoding="utf-8")
+        return pyproject_path
+
+    raise ReleaseError("Missing [project].version in pyproject.toml")
+
+
+def write_changelog(version: str, commits: list[str]) -> Path:
+    """Write a changelog draft for the release.
+
+    Args:
+        version: Release version without the leading `v`.
+        commits: Commit subject lines to include as the first changelog draft.
+
+    Returns:
+        Path to the created changelog file.
+
+    Raises:
+        ReleaseError: The changelog file already exists.
+    """
+    changelog_path = REPO_ROOT / "changelogs" / f"v{version}.md"
+    if changelog_path.exists():
+        raise ReleaseError(f"Changelog already exists: {changelog_path}")
+
+    changelog_path.parent.mkdir(parents=True, exist_ok=True)
+    entries = [f"- {commit}" for commit in commits] or ["- "]
+    changelog_path.write_text(
+        "\n".join(
+            [
+                "## What's Changed",
+                "",
+                "<!-- Review, group, and polish these entries before publishing. -->",
+                "",
+                *entries,
+                "",
+            ]
+        ),
+        encoding="utf-8",
+    )
+    return changelog_path
+
+
+def create_release_branch(version: str, base_branch: str, remote: str) -> str:
+    """Create a release branch from the updated base branch.
+
+    Args:
+        version: Release version without the leading `v`.
+        base_branch: Base branch to release from.
+        remote: Remote name used for fetching and fast-forward pulls.
+
+    Returns:
+        Created release branch name.
+
+    Raises:
+        ReleaseError: The branch already exists or Git cannot create it.
+    """
+    branch = f"release/{version}"
+    git(["checkout", base_branch])
+    git(["pull", "--ff-only", remote, base_branch])
+    git(["fetch", "--tags", remote])
+
+    local_branch = git(["branch", "--list", branch], capture_output=True)
+    if local_branch:
+        raise ReleaseError(f"Local branch already exists: {branch}")
+
+    remote_branch = git(["ls-remote", "--heads", remote, branch], capture_output=True)
+    if remote_branch:
+        raise ReleaseError(f"Remote branch already exists: {remote}/{branch}")
+
+    git(["switch", "-c", branch])
+    return branch
+
+
+def run_validation(args: argparse.Namespace) -> None:
+    """Run release validation commands selected by CLI flags.
+
+    Args:
+        args: Parsed CLI arguments.
+
+    Raises:
+        ReleaseError: A validation command fails.
+    """
+    if args.generate_api_client:
+        run_command(["pnpm", "generate:api"], cwd=REPO_ROOT / "dashboard")
+
+    if not args.skip_checks:
+        run_command(["uv", "run", "ruff", "format", "--check", "."])
+        run_command(["uv", "run", "ruff", "check", "."])
+
+    if args.dashboard_build:
+        run_command(["pnpm", "install"], cwd=REPO_ROOT / "dashboard")
+        run_command(["pnpm", "build"], cwd=REPO_ROOT / "dashboard")
+
+
+def commit_and_maybe_push(
+    version: str,
+    branch: str,
+    changelog_path: Path,
+    args: argparse.Namespace,
+) -> None:
+    """Commit release preparation changes and optionally push the branch.
+
+    Args:
+        version: Release version without the leading `v`.
+        branch: Release branch name.
+        changelog_path: Changelog file created for this release.
+        args: Parsed CLI arguments.
+
+    Raises:
+        ReleaseError: Git add, commit, or push fails.
+    """
+    git(["add", "pyproject.toml", str(changelog_path.relative_to(REPO_ROOT))])
+    if args.generate_api_client:
+        git(["add", "dashboard/src/api/generated"])
+
+    git(["commit", "-m", f"chore: bump version to {version}"])
+    if args.push:
+        git(["push", "-u", args.remote, branch])
+
+
+def print_next_steps(
+    version: str,
+    branch: str,
+    changelog_path: Path,
+    args: argparse.Namespace,
+) -> None:
+    """Print the manual steps that remain after preparation.
+
+    Args:
+        version: Release version without the leading `v`.
+        branch: Release branch name.
+        changelog_path: Changelog file created for this release.
+        args: Parsed CLI arguments.
+    """
+    changelog_rel = changelog_path.relative_to(REPO_ROOT)
+    print("\nRelease preparation complete.")
+    print(f"Branch: {branch}")
+    print(f"Changelog: {changelog_rel}")
+
+    if args.commit:
+        if not args.push:
+            print(f"Next: git push -u {args.remote} {branch}")
+    else:
+        print("Next:")
+        print(f"1. Review and polish {changelog_rel}")
+        print(f"2. git add pyproject.toml {changelog_rel}")
+        print(f'3. git commit -m "chore: bump version to {version}"')
+        print(f"4. git push -u {args.remote} {branch}")
+
+    print(f"Open a PR from {branch} to {args.base_branch}.")
+    print(
+        "After the PR is merged, tag from the updated base branch with "
+        f"`git tag v{version}` and `git push {args.remote} v{version}`."
+    )
+
+
+def parse_args(argv: list[str]) -> argparse.Namespace:
+    """Parse command-line arguments.
+
+    Args:
+        argv: Raw command-line arguments excluding the executable name.
+
+    Returns:
+        Parsed CLI arguments.
+
+    Raises:
+        ReleaseError: Push is requested without commit.
+    """
+    parser = argparse.ArgumentParser(
+        description="Prepare an AstrBot release branch, version bump, and changelog.",
+    )
+    parser.add_argument("version", help="Release version without the leading v")
+    parser.add_argument("--base-branch", default="master", help="Release base branch")
+    parser.add_argument("--remote", default="origin", help="Git remote name")
+    parser.add_argument(
+        "--generate-api-client",
+        action="store_true",
+        help="Run dashboard API client generation before validation",
+    )
+    parser.add_argument(
+        "--dashboard-build",
+        action="store_true",
+        help="Run dashboard install and build validation",
+    )
+    parser.add_argument(
+        "--skip-checks",
+        action="store_true",
+        help="Skip ruff format and ruff check",
+    )
+    parser.add_argument(
+        "--commit",
+        action="store_true",
+        help="Commit the generated release preparation changes",
+    )
+    parser.add_argument(
+        "--push",
+        action="store_true",
+        help="Push the release branch after committing; requires --commit",
+    )
+    args = parser.parse_args(argv)
+    if args.push and not args.commit:
+        raise ReleaseError("--push requires --commit")
+    return args
+
+
+def main(argv: list[str] | None = None) -> int:
+    """Run the release preparation workflow.
+
+    Args:
+        argv: Optional command-line arguments for tests or programmatic calls.
+
+    Returns:
+        Process exit code.
+    """
+    try:
+        args = parse_args(sys.argv[1:] if argv is None else argv)
+        version = validate_version(args.version)
+        ensure_clean_worktree()
+
+        branch = create_release_branch(version, args.base_branch, args.remote)
+        tag = latest_tag()
+        if tag:
+            print(f"Latest tag: {tag}")
+        else:
+            print("No existing tags found; changelog will use all reachable commits.")
+
+        commits = release_commits(tag)
+        update_pyproject_version(version)
+        changelog_path = write_changelog(version, commits)
+        run_validation(args)
+
+        if args.commit:
+            commit_and_maybe_push(version, branch, changelog_path, args)
+
+        print_next_steps(version, branch, changelog_path, args)
+        return 0
+    except ReleaseError as exc:
+        print(f"prepare-release: {exc}", file=sys.stderr)
+        return 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

tests/test_anthropic_kimi_code_provider.py

@@ -1,9 +1,12 @@
 import builtins
+from types import SimpleNamespace
 
+import httpx
 import pytest
 
 import astrbot.core.provider.sources.anthropic_source as anthropic_source
 import astrbot.core.provider.sources.kimi_code_source as kimi_code_source
+import astrbot.core.provider.sources.request_retry as request_retry
 from astrbot.core.exceptions import EmptyModelOutputError
 from astrbot.core.provider.entities import LLMResponse
 
@@ -171,6 +174,36 @@ def test_create_http_client_falls_back_to_global_httpx_module(monkeypatch):
     assert captured["httpx_module"] is anthropic_source.httpx
 
 
+@pytest.mark.asyncio
+async def test_anthropic_get_models_retries_transient_request_error(monkeypatch):
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MIN_S", 0)
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MAX_S", 0)
+
+    class FakeModels:
+        def __init__(self):
+            self.calls = 0
+
+        async def list(self):
+            self.calls += 1
+            if self.calls == 1:
+                raise httpx.ConnectError("temporary connection failure")
+            return SimpleNamespace(
+                data=[
+                    SimpleNamespace(id="claude-b"),
+                    SimpleNamespace(id="claude-a"),
+                ]
+            )
+
+    models = FakeModels()
+    provider = anthropic_source.ProviderAnthropic.__new__(
+        anthropic_source.ProviderAnthropic
+    )
+    provider.client = SimpleNamespace(models=models)
+
+    assert await provider.get_models() == ["claude-a", "claude-b"]
+    assert models.calls == 2
+
+
 @pytest.mark.asyncio
 async def test_text_chat_wraps_string_system_prompt_as_list(monkeypatch):
     monkeypatch.setattr(anthropic_source, "AsyncAnthropic", _FakeAsyncAnthropic)
@@ -187,7 +220,7 @@ async def test_text_chat_wraps_string_system_prompt_as_list(monkeypatch):
 
     captured_payloads: dict[str, object] = {}
 
-    async def fake_query(payloads, tools):
+    async def fake_query(payloads, tools, *, request_max_retries=None):
         captured_payloads.update(payloads)
         return LLMResponse(role="assistant", completion_text="ok")
 
@@ -214,7 +247,7 @@ async def test_text_chat_passes_through_list_system_prompt(monkeypatch):
 
     captured_payloads: dict[str, object] = {}
 
-    async def fake_query(payloads, tools):
+    async def fake_query(payloads, tools, *, request_max_retries=None):
         captured_payloads.update(payloads)
         return LLMResponse(role="assistant", completion_text="ok")
 

tests/test_gemini_source.py

@@ -1,6 +1,10 @@
+from types import SimpleNamespace
+
+import httpx
 import pytest
 
 from astrbot.core.exceptions import EmptyModelOutputError
+import astrbot.core.provider.sources.request_retry as request_retry
 from astrbot.core.provider.entities import LLMResponse
 from astrbot.core.provider.sources.gemini_source import ProviderGoogleGenAI
 
@@ -27,3 +31,35 @@ def test_gemini_reasoning_only_output_is_allowed():
         response_id="resp_reasoning",
         finish_reason="STOP",
     )
+
+
+@pytest.mark.asyncio
+async def test_gemini_get_models_retries_transient_request_error(monkeypatch):
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MIN_S", 0)
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MAX_S", 0)
+
+    class FakeModels:
+        def __init__(self):
+            self.calls = 0
+
+        async def list(self):
+            self.calls += 1
+            if self.calls == 1:
+                raise httpx.ConnectError("temporary connection failure")
+            return [
+                SimpleNamespace(
+                    name="models/gemini-a",
+                    supported_actions=["generateContent"],
+                ),
+                SimpleNamespace(
+                    name="models/gemini-b",
+                    supported_actions=["embedContent"],
+                ),
+            ]
+
+    models = FakeModels()
+    provider = ProviderGoogleGenAI.__new__(ProviderGoogleGenAI)
+    provider.client = SimpleNamespace(models=models)
+
+    assert await provider.get_models() == ["gemini-a"]
+    assert models.calls == 2

tests/test_httpx_socks_dependency.py

@@ -2,7 +2,8 @@ import re
 from pathlib import Path
 
 import pytest
-import tomllib
+
+from astrbot.core.utils.toml_parser import read_pyproject_project_dependencies
 
 PROJECT_ROOT = Path(__file__).resolve().parents[1]
 REQUIREMENTS_PATH = PROJECT_ROOT / "requirements.txt"
@@ -28,9 +29,7 @@ def _read_requirements() -> list[str]:
 
 
 def _read_pyproject_dependencies() -> list[str]:
-    with PYPROJECT_PATH.open("rb") as file:
-        pyproject = tomllib.load(file)
-    return pyproject["project"]["dependencies"]
+    return read_pyproject_project_dependencies(PYPROJECT_PATH)
 
 
 def test_requirements_include_httpx_socks_dependency() -> None:

tests/test_openai_source.py

@@ -3,13 +3,16 @@ import builtins
 from io import BytesIO
 from types import SimpleNamespace
 
+import httpx
 import pytest
 from openai.types.chat.chat_completion import ChatCompletion
 from openai.types.chat.chat_completion_chunk import ChatCompletionChunk
 from PIL import Image as PILImage
 
 import astrbot.core.provider.sources.openai_source as openai_source_module
+import astrbot.core.provider.sources.request_retry as request_retry
 from astrbot.core.exceptions import EmptyModelOutputError
+from astrbot.core.provider.entities import LLMResponse
 from astrbot.core.provider.sources.groq_source import ProviderGroq
 from astrbot.core.provider.sources.openai_source import ProviderOpenAIOfficial
 from astrbot.core.utils.media_utils import ResolvedMediaData, file_uri_to_path
@@ -117,6 +120,57 @@ def test_create_http_client_falls_back_to_global_httpx_module(monkeypatch):
     assert captured["httpx_module"] is openai_source_module.httpx
 
 
+@pytest.mark.asyncio
+async def test_get_models_retries_transient_request_error(monkeypatch):
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MIN_S", 0)
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MAX_S", 0)
+
+    class FakeModels:
+        def __init__(self):
+            self.calls = 0
+
+        async def list(self):
+            self.calls += 1
+            if self.calls == 1:
+                raise httpx.ConnectError("temporary connection failure")
+            return SimpleNamespace(
+                data=[
+                    SimpleNamespace(id="gpt-b"),
+                    SimpleNamespace(id="gpt-a"),
+                ]
+            )
+
+    models = FakeModels()
+    provider = ProviderOpenAIOfficial.__new__(ProviderOpenAIOfficial)
+    provider.client = SimpleNamespace(models=models)
+
+    assert await provider.get_models() == ["gpt-a", "gpt-b"]
+    assert models.calls == 2
+
+
+@pytest.mark.asyncio
+async def test_text_chat_passes_request_max_retries_to_query():
+    captured: dict[str, object] = {}
+
+    provider = ProviderOpenAIOfficial.__new__(ProviderOpenAIOfficial)
+    provider.api_keys = ["test-key"]
+    provider.client = SimpleNamespace(api_key=None)
+
+    async def fake_prepare_chat_payload(*args, **kwargs):
+        return {"messages": [], "model": "gpt-4o-mini"}, []
+
+    async def fake_query(payloads, func_tool, *, request_max_retries=None):
+        captured["request_max_retries"] = request_max_retries
+        return LLMResponse(role="assistant", completion_text="ok")
+
+    provider._prepare_chat_payload = fake_prepare_chat_payload
+    provider._query = fake_query
+
+    await provider.text_chat(prompt="hello", request_max_retries=2)
+
+    assert captured["request_max_retries"] == 2
+
+
 @pytest.mark.asyncio
 async def test_handle_api_error_content_moderated_removes_images():
     provider = _make_provider(

tests/test_request_retry.py

@@ -0,0 +1,27 @@
+import httpx
+import pytest
+
+import astrbot.core.provider.sources.request_retry as request_retry
+from astrbot.core.provider.sources.request_retry import retry_provider_request
+
+
+@pytest.mark.asyncio
+async def test_retry_provider_request_uses_configured_max_retries(monkeypatch):
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MIN_S", 0)
+    monkeypatch.setattr(request_retry, "REQUEST_RETRY_WAIT_MAX_S", 0)
+
+    calls = 0
+
+    async def request():
+        nonlocal calls
+        calls += 1
+        raise httpx.ConnectError("temporary connection failure")
+
+    with pytest.raises(httpx.ConnectError):
+        await retry_provider_request(
+            "Test",
+            request,
+            max_attempts=2,
+        )
+
+    assert calls == 2

tests/test_toml_parser.py

@@ -0,0 +1,184 @@
+from pathlib import Path
+
+import pytest
+
+from astrbot.core.utils.toml_parser import (
+    read_pyproject_project_dependencies,
+    read_pyproject_project_version,
+)
+
+
+def test_read_pyproject_project_version_reads_project_section(tmp_path: Path) -> None:
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text(
+        "\n".join(
+            [
+                'version = "ignored"',
+                "[project]",
+                'name = "AstrBot"',
+                'version = "1.2.3-beta.4" # release version',
+                "[tool.example]",
+                'version = "ignored-again"',
+            ]
+        ),
+        encoding="utf-8",
+    )
+
+    assert read_pyproject_project_version(pyproject_path) == "1.2.3-beta.4"
+
+
+@pytest.mark.parametrize(
+    ("version_line", "expected"),
+    [
+        ('version = "1.2.3"', "1.2.3"),
+        ("version='1.2.3-beta.4'", "1.2.3-beta.4"),
+        ('   version  =  "1.2.3-rc.1"   ', "1.2.3-rc.1"),
+    ],
+)
+def test_read_pyproject_project_version_accepts_simple_variants(
+    tmp_path: Path,
+    version_line: str,
+    expected: str,
+) -> None:
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text(
+        "\n".join(
+            [
+                "[project]",
+                'name = "AstrBot"',
+                version_line,
+            ]
+        ),
+        encoding="utf-8",
+    )
+
+    assert read_pyproject_project_version(pyproject_path) == expected
+
+
+@pytest.mark.parametrize(
+    ("version_line", "message"),
+    [
+        ("version", "Missing value separator for project.version"),
+        ('version = "1.2.3', "Unterminated project.version string"),
+        ('version = "1.2.3" extra', "Unsupported content after project.version"),
+        ('version = ""', "Empty project.version value"),
+    ],
+)
+def test_read_pyproject_project_version_rejects_invalid_values(
+    tmp_path: Path,
+    version_line: str,
+    message: str,
+) -> None:
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text(
+        "\n".join(
+            [
+                "[project]",
+                'name = "AstrBot"',
+                version_line,
+            ]
+        ),
+        encoding="utf-8",
+    )
+
+    with pytest.raises(ValueError, match=message):
+        read_pyproject_project_version(pyproject_path)
+
+
+def test_read_pyproject_project_dependencies_reads_multiline_array(
+    tmp_path: Path,
+) -> None:
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text(
+        "\n".join(
+            [
+                "[project]",
+                "dependencies = [",
+                '  "aiohttp>=3.11.18",',
+                "  \"audioop-lts ; python_full_version >= '3.13'\", # marker",
+                "] # end dependencies",
+            ]
+        ),
+        encoding="utf-8",
+    )
+
+    assert read_pyproject_project_dependencies(pyproject_path) == [
+        "aiohttp>=3.11.18",
+        "audioop-lts ; python_full_version >= '3.13'",
+    ]
+
+
+@pytest.mark.parametrize(
+    ("dependency_line", "expected"),
+    [
+        ("dependencies = []", []),
+        ('dependencies = ["aiohttp>=3.11.18"]', ["aiohttp>=3.11.18"]),
+        (
+            'dependencies = ["psutil>=5.8.0,<7.2.0", "httpx[socks]>=0.28.1"]',
+            ["psutil>=5.8.0,<7.2.0", "httpx[socks]>=0.28.1"],
+        ),
+    ],
+)
+def test_read_pyproject_project_dependencies_accepts_inline_arrays(
+    tmp_path: Path,
+    dependency_line: str,
+    expected: list[str],
+) -> None:
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text(
+        "\n".join(
+            [
+                "[project]",
+                dependency_line,
+            ]
+        ),
+        encoding="utf-8",
+    )
+
+    assert read_pyproject_project_dependencies(pyproject_path) == expected
+
+
+@pytest.mark.parametrize(
+    ("project_lines", "message"),
+    [
+        (["[project]", 'name = "AstrBot"'], "Missing project.dependencies"),
+        (
+            ["[project]", "dependencies = ["],
+            "Unterminated project.dependencies array",
+        ),
+        (
+            ["[project]", 'dependencies = "aiohttp>=3.11.18"'],
+            "Unsupported project.dependencies value",
+        ),
+        (
+            ["[project]", "dependencies = [", "  aiohttp>=3.11.18,", "]"],
+            "Unsupported project.dependencies entry value",
+        ),
+        (
+            ["[project]", "dependencies = [", '  "aiohttp>=3.11.18" extra', "]"],
+            "Unsupported content after project.dependencies entry",
+        ),
+        (
+            ["[project]", "dependencies = [", '  ""', "]"],
+            "Empty project.dependencies entry value",
+        ),
+    ],
+)
+def test_read_pyproject_project_dependencies_rejects_invalid_values(
+    tmp_path: Path,
+    project_lines: list[str],
+    message: str,
+) -> None:
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text("\n".join(project_lines), encoding="utf-8")
+
+    with pytest.raises(ValueError, match=message):
+        read_pyproject_project_dependencies(pyproject_path)
+
+
+def test_read_pyproject_project_version_raises_when_missing(tmp_path: Path) -> None:
+    pyproject_path = tmp_path / "pyproject.toml"
+    pyproject_path.write_text('[project]\nname = "AstrBot"\n', encoding="utf-8")
+
+    with pytest.raises(ValueError, match="Missing project.version"):
+        read_pyproject_project_version(pyproject_path)