coding-proxy 0.5.1a4__tar.gz → 0.5.1a5__tar.gz

{coding_proxy-0.5.1a4 → coding_proxy-0.5.1a5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coding-proxy
-Version: 0.5.1a4
+Version: 0.5.1a5
 Summary: A High-Availability, Transparent, and Smart Multi-Vendor Proxy for Claude Code. Support Claude Plans, GitHub Copilot, Google Antigravity, ZAI/GLM, MiniMax, Qwen, Xiaomi, Kimi, Doubao...
 Project-URL: Source Code, https://github.com/ThreeFish-AI/coding-proxy
 Project-URL: User Guide, https://github.com/ThreeFish-AI/coding-proxy/blob/master/docs/user-guide.md

{coding_proxy-0.5.1a4 → coding_proxy-0.5.1a5}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "coding-proxy"
-version = "0.5.1a4"
+version = "0.5.1a5"
 description = "A High-Availability, Transparent, and Smart Multi-Vendor Proxy for Claude Code. Support Claude Plans, GitHub Copilot, Google Antigravity, ZAI/GLM, MiniMax, Qwen, Xiaomi, Kimi, Doubao..."
 readme = "README.md"
 requires-python = ">=3.12"

{coding_proxy-0.5.1a4 → coding_proxy-0.5.1a5}/src/coding/proxy/logging/db.py

@@ -335,6 +335,20 @@ class TokenLogger:
         )
         await self._db.commit()
 
+    async def update_empty_session_title(self, session_key: str, title: str) -> None:
+        """为标题为空的 session 补写标题（幂等,仅覆盖空标题行）.
+
+        使用 ``AND title = ''`` 条件确保不覆盖已有标题,
+        即使行已存在但标题为空也会被更新。
+        """
+        if not self._db or not title or not session_key:
+            return
+        await self._db.execute(
+            "UPDATE session_meta SET title = ? WHERE session_key = ? AND title = ''",
+            (title, session_key),
+        )
+        await self._db.commit()
+
     async def get_session_titles(self, session_keys: list[str]) -> dict[str, str]:
         """批量查询 session 标题."""
         if not self._db or not session_keys:

{coding_proxy-0.5.1a4 → coding_proxy-0.5.1a5}/src/coding/proxy/routing/executor.py

@@ -50,11 +50,13 @@ from ..compat.canonical import (
     CompatibilityStatus,
     build_canonical_request,
 )
-from ..model.compat import CanonicalRequest
+from ..model.compat import CanonicalMessagePart, CanonicalRequest
 
 logger = logging.getLogger(__name__)
 
 _SESSION_TITLE_MAX_LEN = 600
+# 回退标题截取长度 — 工具结果等非用户直接输入的摘要上限。
+_FALLBACK_TITLE_MAX_LEN = 80
 
 # Claude Code 注入的"噪声"标签 — 系统级上下文,不应进入 Session 标题。
 # 这些标签由 CC harness 在首个 user 消息 content 中拼接,高度同质,
@@ -63,7 +65,8 @@ _NOISE_TAG_PATTERN = re.compile(
     r"<(?P<tag>system-reminder|user-preferences|"
     r"local-command-stdout|local-command-stderr|"
     r"bash-input|bash-stdout|bash-stderr|"
-    r"ide_selection|stdin|system_instruction|session)\b[^>]*>"
+    r"ide_selection|stdin|system_instruction|session|"
+    r"artifactMetadata|thinking)\b[^>]*>"
     r".*?</(?P=tag)>",
     flags=re.DOTALL | re.IGNORECASE,
 )
@@ -109,13 +112,19 @@ def _sanitize_user_text(raw: str) -> str:
     return re.sub(r"\s+", " ", cleaned).strip()
 
 
-def _extract_session_title(request: CanonicalRequest) -> str:
-    """从规范化请求中提取首个用户消息文本作为 session 标题。
+# ── Session 标题提取: 多层级回退策略 ──────────────────────────────
+#
+# Level 1: user TEXT → 噪声剥离 → 首条非空文本 (原有逻辑)
+# Level 2: user TOOL_RESULT → text 截取 → "[Tool output] <snippet>"
+# Level 3: user IMAGE → 计数 → "[1 Image]" / "[N Images]"
+# Level 4: 请求元数据 → tool_names / model → "[Tool call] Bash, Read"
+#          / "[Session] claude-opus-4-8"
+# ─────────────────────────────────────────────────────────────────
 
-    跳过 Claude Code 注入的系统级 XML 块(system-reminder、user-preferences 等),
-    确保标题反映用户真实输入而非高同质化的系统模板。
-    """
-    for part in request.messages:
+
+def _extract_title_from_user_text(messages: list[CanonicalMessagePart]) -> str:
+    """Level 1: 从 user TEXT 部分提取经噪声剥离后的首条非空文本."""
+    for part in messages:
         if part.role != "user" or part.type != CanonicalPartType.TEXT:
             continue
         cleaned = _sanitize_user_text(part.text)
@@ -124,6 +133,59 @@ def _extract_session_title(request: CanonicalRequest) -> str:
     return ""
 
 
+def _extract_title_from_tool_results(messages: list[CanonicalMessagePart]) -> str:
+    """Level 2: 从 user TOOL_RESULT 部分截取文本摘要."""
+    for part in messages:
+        if part.role != "user" or part.type != CanonicalPartType.TOOL_RESULT:
+            continue
+        if not part.text:
+            continue
+        cleaned = _sanitize_user_text(part.text)
+        if cleaned:
+            snippet = cleaned[:_FALLBACK_TITLE_MAX_LEN]
+            return f"[Tool output] {snippet}"
+    return ""
+
+
+def _extract_title_from_images(messages: list[CanonicalMessagePart]) -> str:
+    """Level 3: 统计 user IMAGE 部分数量,生成图片描述标题."""
+    count = sum(
+        1 for p in messages if p.role == "user" and p.type == CanonicalPartType.IMAGE
+    )
+    if count == 0:
+        return ""
+    return f"[{count} Image{'s' if count > 1 else ''}]"
+
+
+def _extract_title_from_metadata(request: CanonicalRequest) -> str:
+    """Level 4: 从请求元数据 (tool_names / model) 合成兜底标题."""
+    if request.tool_names:
+        names = ", ".join(request.tool_names[:3])
+        return f"[Tool call] {names}"
+    if request.model:
+        return f"[Session] {request.model}"
+    return ""
+
+
+def _extract_session_title(request: CanonicalRequest) -> str:
+    """从规范化请求中提取 session 标题 — 多层级回退策略。
+
+    依次尝试: user TEXT 噪声剥离 → TOOL_RESULT 摘要 → IMAGE 计数 → 元数据兜底。
+    任意级别命中即返回,确保 Dashboard 尽可能展示有辨识度的标题。
+    """
+    messages = request.messages
+    for extractor in (
+        _extract_title_from_user_text,
+        _extract_title_from_tool_results,
+        _extract_title_from_images,
+    ):
+        title = extractor(messages)
+        if title:
+            return title[:_SESSION_TITLE_MAX_LEN]
+    # Level 4 依赖 request 元数据,签名不同
+    return _extract_title_from_metadata(request)[:_SESSION_TITLE_MAX_LEN]
+
+
 def _build_semantic_rejection_diagnostic(body: dict[str, Any]) -> str:
     """构建语义拒绝的请求体诊断上下文.
 
@@ -663,6 +725,13 @@ class _RouteExecutor:
                 await self._recorder.set_session_title(
                     canonical_request.session_key, title
                 )
+        else:
+            # 延迟标题补写: 若 session 尚无标题,尝试从当前请求中提取并回写。
+            title = _extract_session_title(canonical_request)
+            if title:
+                await self._recorder.update_empty_session_title(
+                    canonical_request.session_key, title
+                )
         incompatible_reasons: list[str] = []
         effective_tiers = self._resolve_effective_tiers(canonical_request.session_key)
         last_idx = len(effective_tiers) - 1
@@ -842,6 +911,13 @@ class _RouteExecutor:
                 await self._recorder.set_session_title(
                     canonical_request.session_key, title
                 )
+        else:
+            # 延迟标题补写: 若 session 尚无标题,尝试从当前请求中提取并回写。
+            title = _extract_session_title(canonical_request)
+            if title:
+                await self._recorder.update_empty_session_title(
+                    canonical_request.session_key, title
+                )
         incompatible_reasons: list[str] = []
         effective_tiers = self._resolve_effective_tiers(canonical_request.session_key)
         last_idx = len(effective_tiers) - 1

{coding_proxy-0.5.1a4 → coding_proxy-0.5.1a5}/src/coding/proxy/routing/usage_recorder.py

@@ -33,6 +33,11 @@ class UsageRecorder:
         if self._token_logger:
             await self._token_logger.set_session_title(session_key, title)
 
+    async def update_empty_session_title(self, session_key: str, title: str) -> None:
+        """为标题为空的 session 补写标题（委托给 TokenLogger）."""
+        if self._token_logger:
+            await self._token_logger.update_empty_session_title(session_key, title)
+
     # ── 用量信息构建 ──────────────────────────────────────
 
     @staticmethod

{coding_proxy-0.5.1a4 → coding_proxy-0.5.1a5}/tests/test_router_executor.py

@@ -20,10 +20,15 @@ from coding.proxy.compat.canonical import (
     build_canonical_request,
 )
 from coding.proxy.routing.executor import (
+    _FALLBACK_TITLE_MAX_LEN,
     _SESSION_TITLE_MAX_LEN,
     _VENDOR_PROTOCOL_LABEL_MAP,
     _build_semantic_rejection_diagnostic,
     _extract_session_title,
+    _extract_title_from_images,
+    _extract_title_from_metadata,
+    _extract_title_from_tool_results,
+    _extract_title_from_user_text,
     _has_tool_results,
     _is_likely_request_format_error,
     _log_vendor_response_error,
@@ -2258,6 +2263,20 @@ class TestSanitizeUserText:
         raw = "<session>\nline1\nline2\n</session>真实标题"
         assert _sanitize_user_text(raw) == "真实标题"
 
+    def test_strips_artifact_metadata_tag(self):
+        """``<artifactMetadata>`` 标签应被完整剥离."""
+        raw = "<artifactMetadata>artifact context</artifactMetadata>用户文本"
+        assert _sanitize_user_text(raw) == "用户文本"
+
+    def test_strips_thinking_tag(self):
+        """``<thinking>`` 标签应被完整剥离."""
+        raw = "<thinking>内部推理过程</thinking>用户实际提问"
+        assert _sanitize_user_text(raw) == "用户实际提问"
+
+    def test_strips_thinking_tag_multiline(self):
+        raw = "<thinking>\nline1\nline2\n</thinking>清理后文本"
+        assert _sanitize_user_text(raw) == "清理后文本"
+
 
 class TestExtractSessionTitle:
     """``_extract_session_title`` — 端到端从 CanonicalRequest 抽取标题."""
@@ -2304,14 +2323,16 @@ class TestExtractSessionTitle:
         req = self._build_request([{"role": "user", "content": raw}])
         assert _extract_session_title(req) == "/commit feat: 新增标题清洗"
 
-    def test_returns_empty_when_only_noise(self):
+    def test_returns_metadata_fallback_when_only_noise(self):
+        """纯噪声文本回退到 Level 4 元数据兜底(使用 model 名称)."""
         raw = "<system-reminder>纯噪声</system-reminder>"
         req = self._build_request([{"role": "user", "content": raw}])
-        assert _extract_session_title(req) == ""
+        assert _extract_session_title(req) == "[Session] test"
 
-    def test_returns_empty_for_no_user_messages(self):
+    def test_returns_metadata_fallback_for_no_user_messages(self):
+        """无 user 消息时回退到 Level 4 元数据兜底."""
         req = self._build_request([{"role": "assistant", "content": "你好"}])
-        assert _extract_session_title(req) == ""
+        assert _extract_session_title(req) == "[Session] test"
 
     def test_skips_noise_only_part_to_find_real_input(self):
         """首个 user text part 全噪声时,fallback 到下一个非空 user part."""
@@ -2338,3 +2359,266 @@ class TestExtractSessionTitle:
         ]
         req = self._build_request(messages)
         assert _extract_session_title(req) == "新的用户问题"
+
+
+# ═══════════════════════════════════════════════════════════════════
+# 多层级回退标题提取测试
+# ═══════════════════════════════════════════════════════════════════
+
+
+class TestExtractTitleFromUserText:
+    """Level 1 辅助函数 ``_extract_title_from_user_text``."""
+
+    def test_returns_first_non_empty_user_text(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT, role="user", text="用户输入"
+            ),
+        ]
+        assert _extract_title_from_user_text(msgs) == "用户输入"
+
+    def test_skips_assistant_text(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT, role="assistant", text="助手回复"
+            ),
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT, role="user", text="用户问题"
+            ),
+        ]
+        assert _extract_title_from_user_text(msgs) == "用户问题"
+
+    def test_returns_empty_for_noise_only(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TEXT,
+                role="user",
+                text="<system-reminder>纯噪声</system-reminder>",
+            ),
+        ]
+        assert _extract_title_from_user_text(msgs) == ""
+
+
+class TestExtractTitleFromToolResults:
+    """Level 2 辅助函数 ``_extract_title_from_tool_results``."""
+
+    def test_extracts_tool_result_text(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TOOL_RESULT,
+                role="user",
+                text="file contents here",
+            ),
+        ]
+        title = _extract_title_from_tool_results(msgs)
+        assert title == "[Tool output] file contents here"
+
+    def test_skips_empty_tool_result(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TOOL_RESULT, role="user", text=""
+            ),
+        ]
+        assert _extract_title_from_tool_results(msgs) == ""
+
+    def test_truncates_long_tool_result(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        long_text = "A" * 200
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TOOL_RESULT, role="user", text=long_text
+            ),
+        ]
+        title = _extract_title_from_tool_results(msgs)
+        assert title.startswith("[Tool output] ")
+        assert len(title) <= len("[Tool output] ") + _FALLBACK_TITLE_MAX_LEN
+
+    def test_sanitizes_noise_in_tool_result(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TOOL_RESULT,
+                role="user",
+                text="<system-reminder>noise</system-reminder>clean output",
+            ),
+        ]
+        title = _extract_title_from_tool_results(msgs)
+        assert title == "[Tool output] clean output"
+
+    def test_returns_empty_when_all_noise(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(
+                type=CanonicalPartType.TOOL_RESULT,
+                role="user",
+                text="<system-reminder>纯噪声</system-reminder>",
+            ),
+        ]
+        assert _extract_title_from_tool_results(msgs) == ""
+
+
+class TestExtractTitleFromImages:
+    """Level 3 辅助函数 ``_extract_title_from_images``."""
+
+    def test_single_image(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(type=CanonicalPartType.IMAGE, role="user"),
+        ]
+        assert _extract_title_from_images(msgs) == "[1 Image]"
+
+    def test_multiple_images(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(type=CanonicalPartType.IMAGE, role="user"),
+            CanonicalMessagePart(type=CanonicalPartType.IMAGE, role="user"),
+            CanonicalMessagePart(type=CanonicalPartType.IMAGE, role="user"),
+        ]
+        assert _extract_title_from_images(msgs) == "[3 Images]"
+
+    def test_no_images(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(type=CanonicalPartType.TEXT, role="user", text="文本"),
+        ]
+        assert _extract_title_from_images(msgs) == ""
+
+    def test_skips_assistant_images(self):
+        from coding.proxy.model.compat import CanonicalMessagePart, CanonicalPartType
+
+        msgs = [
+            CanonicalMessagePart(type=CanonicalPartType.IMAGE, role="assistant"),
+        ]
+        assert _extract_title_from_images(msgs) == ""
+
+
+class TestExtractTitleFromMetadata:
+    """Level 4 辅助函数 ``_extract_title_from_metadata``."""
+
+    @staticmethod
+    def _build_request_with_meta(tool_names: list[str] | None = None, model: str = ""):
+        body: dict = {"model": model, "messages": []}
+        if tool_names:
+            body["tools"] = [{"name": n} for n in tool_names]
+        return build_canonical_request(body, {})
+
+    def test_uses_tool_names(self):
+        req = self._build_request_with_meta(
+            tool_names=["Bash", "Read", "Edit"], model="claude-opus-4-8"
+        )
+        assert _extract_title_from_metadata(req) == "[Tool call] Bash, Read, Edit"
+
+    def test_limits_to_three_tool_names(self):
+        req = self._build_request_with_meta(
+            tool_names=["Bash", "Read", "Edit", "Write", "Grep"], model="test"
+        )
+        assert _extract_title_from_metadata(req) == "[Tool call] Bash, Read, Edit"
+
+    def test_uses_model_when_no_tools(self):
+        req = self._build_request_with_meta(tool_names=[], model="claude-sonnet-4-6")
+        assert _extract_title_from_metadata(req) == "[Session] claude-sonnet-4-6"
+
+    def test_returns_empty_when_nothing(self):
+        req = self._build_request_with_meta(tool_names=[], model="")
+        assert _extract_title_from_metadata(req) == ""
+
+
+class TestExtractSessionTitleFallback:
+    """``_extract_session_title`` 多层级回退集成测试."""
+
+    @staticmethod
+    def _build_request(messages: list[dict], **extra):
+        body: dict = {"model": "test-model", "messages": messages, **extra}
+        return build_canonical_request(body, {})
+
+    def test_level1_takes_priority(self):
+        """Level 1 命中时不回退到 Level 2."""
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "用户真实问题"},
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "tu_1",
+                        "content": "工具输出",
+                    },
+                ],
+            }
+        ]
+        req = self._build_request(messages)
+        assert _extract_session_title(req) == "用户真实问题"
+
+    def test_level2_when_no_text(self):
+        """无 user TEXT 时,回退到 Level 2 TOOL_RESULT."""
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "tool_result",
+                        "tool_use_id": "tu_1",
+                        "content": [{"type": "text", "text": "文件内容摘要"}],
+                    },
+                ],
+            }
+        ]
+        req = self._build_request(messages)
+        assert _extract_session_title(req) == "[Tool output] 文件内容摘要"
+
+    def test_level3_when_only_images(self):
+        """无 TEXT 和 TOOL_RESULT 时,回退到 Level 3 IMAGE."""
+        messages = [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "image",
+                        "source": {
+                            "type": "base64",
+                            "media_type": "image/png",
+                            "data": "abc",
+                        },
+                    },
+                ],
+            }
+        ]
+        req = self._build_request(messages)
+        assert _extract_session_title(req) == "[1 Image]"
+
+    def test_level4_uses_tool_names(self):
+        """所有消息级别均无内容时,回退到 Level 4 元数据."""
+        req = self._build_request([], tools=[{"name": "Bash"}, {"name": "Read"}])
+        assert _extract_session_title(req) == "[Tool call] Bash, Read"
+
+    def test_level4_uses_model_name(self):
+        """无 tools 时,Level 4 使用 model 名称."""
+        req = self._build_request([])
+        assert _extract_session_title(req) == "[Session] test-model"
+
+    def test_fallback_cascade_full(self):
+        """Level 1 全噪声 → Level 2 全噪声 → Level 3 无图 → Level 4 模型名."""
+        messages = [
+            {
+                "role": "user",
+                "content": "<system-reminder>纯噪声</system-reminder>",
+            },
+        ]
+        req = self._build_request(messages)
+        assert _extract_session_title(req) == "[Session] test-model"

{coding_proxy-0.5.1a4 → coding_proxy-0.5.1a5}/uv.lock

@@ -74,7 +74,7 @@ wheels = [
 
 [[package]]
 name = "coding-proxy"
-version = "0.5.1a4"
+version = "0.5.1a5"
 source = { editable = "." }
 dependencies = [
     { name = "aiosqlite" },