feat: M4 文风 + M5 生成/多provider/Skill + Kimi Code 订阅接入 + 本地联调修复

M4（文风）: style-auditor 双轨（提取指纹/漂移第四审）+ jobs 长任务框架(zombie reaper) + 回炉 refine + GET /style read-back。
M5（生成+扩展）: worldbuilder/character-gen（入库 continuity 409 gate + partition_writes 白名单 + schema→JSONB 形变）；
  网关多 provider 回退链/熔断/能力降级（Anthropic/Gemini 适配器）；Skill registry + 表权限沙箱 + 规则；
  前端 角色生成器/世界观/Codex/规则页/技能库/⌘K 命令面板。
K1（Kimi Code 订阅接入）: OAuth device-flow（kimi-code）+ 静态 Console key（kimi-code-key）两路径；
  coding 端点 KimiCLI 伪造头（实测 UA allow-list 门禁，缺则 403）+ JSON 模式结构化（thinking ⊥ tool_choice）。
本地联调修复: CORS 中间件；assemble 注入 premise+「写第N章」指令（修空 prompt 400）；
  GET /outline·/draft read-back + 大纲/工作台/审稿页重载；写页 client/server 常量边界 + notFound 健壮化；
  字数 toLocaleString locale 水合；审稿页终稿从已存草稿 seed（修 accept 422）。
门禁: backend ruff/mypy(157)/alembic 无漂移/pytest 451 · frontend lint/tsc/vitest/build。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

packages/llm_gateway/ww_llm_gateway/adapters/gemini.py

@@ -0,0 +1,127 @@
+"""Google Gemini 适配器（ARCH §4.2/§4.4）。
+
+- 能力：原生结构化输出（`response_mime_type=application/json` + `response_schema`）、思考。
+  前缀缓存走 Gemini 隐式/显式缓存，机制差异大，原型保守标 `prefix_cache=False`。
+- 客户端注入（`GeminiClient` Protocol，`google.genai.Client().aio` 满足）——测试注入替身，
+  绝不联网；真实 SDK（`google-genai`）仅在 apps/api 构造客户端时需要，本模块不硬 import。
+- 瞬时故障翻译为 `TransientProviderError`，交网关退避/回退。
+"""
+
+from __future__ import annotations
+
+from collections.abc import AsyncIterator
+from typing import Any, Protocol
+
+from ..errors import TransientProviderError
+from ..types import LlmRequest
+from .base import Capabilities, ProviderResult, ProviderUsage, StreamChunk
+
+_TRANSIENT_NAMES = frozenset(
+    {
+        "ResourceExhausted",
+        "ServiceUnavailable",
+        "DeadlineExceeded",
+        "InternalServerError",
+        "ServerError",
+        "APIError",
+    }
+)
+
+
+class GeminiModels(Protocol):
+    async def generate_content(self, **kwargs: Any) -> Any: ...
+
+    def generate_content_stream(self, **kwargs: Any) -> Any: ...
+
+
+class GeminiAio(Protocol):
+    models: GeminiModels
+
+
+class GeminiClient(Protocol):
+    """`google.genai.Client()` 满足此 Protocol（用其 `.aio.models`）。"""
+
+    aio: GeminiAio
+
+
+def _is_transient(exc: Exception) -> bool:
+    name = type(exc).__name__
+    if name in _TRANSIENT_NAMES:
+        return True
+    status = getattr(exc, "code", None) or getattr(exc, "status_code", None)
+    return isinstance(status, int) and (status == 429 or status >= 500)
+
+
+def _contents(req: LlmRequest) -> str:
+    if isinstance(req.input, str):
+        return req.input
+    return "\n\n".join(b.text for b in req.input)
+
+
+def _system_text(req: LlmRequest) -> str:
+    return "\n\n".join(b.text for b in req.system)
+
+
+def _usage_from(meta: Any) -> ProviderUsage:
+    if meta is None:
+        return ProviderUsage(input_tokens=0, output_tokens=0)
+    return ProviderUsage(
+        input_tokens=int(getattr(meta, "prompt_token_count", 0) or 0),
+        output_tokens=int(getattr(meta, "candidates_token_count", 0) or 0),
+        cache_read_tokens=int(getattr(meta, "cached_content_token_count", 0) or 0),
+    )
+
+
+def _config(req: LlmRequest) -> dict[str, Any]:
+    cfg: dict[str, Any] = {}
+    system = _system_text(req)
+    if system:
+        cfg["system_instruction"] = system
+    if req.max_tokens is not None:
+        cfg["max_output_tokens"] = req.max_tokens
+    if req.output_schema is not None:
+        cfg["response_mime_type"] = "application/json"
+        cfg["response_schema"] = req.output_schema
+    return cfg
+
+
+class GeminiAdapter:
+    def __init__(self, provider: str, client: GeminiClient) -> None:
+        self.provider = provider
+        self._client = client
+
+    def capabilities(self) -> Capabilities:
+        return Capabilities(structured_output=True, prefix_cache=False, thinking=True)
+
+    async def complete(self, req: LlmRequest, model: str) -> ProviderResult:
+        try:
+            resp = await self._client.aio.models.generate_content(
+                model=model, contents=_contents(req), config=_config(req)
+            )
+        except Exception as exc:
+            if _is_transient(exc):
+                raise TransientProviderError(str(exc), provider=self.provider) from exc
+            raise
+        text = getattr(resp, "text", "") or ""
+        usage = _usage_from(getattr(resp, "usage_metadata", None))
+        if req.output_schema is not None:
+            parsed = req.output_schema.model_validate_json(text)
+            return ProviderResult(text=text, usage=usage, parsed=parsed)
+        return ProviderResult(text=text, usage=usage)
+
+    async def stream(self, req: LlmRequest, model: str) -> AsyncIterator[StreamChunk]:
+        try:
+            stream = await self._client.aio.models.generate_content_stream(
+                model=model, contents=_contents(req), config=_config(req)
+            )
+            async for chunk in stream:
+                text = getattr(chunk, "text", "") or ""
+                if text:
+                    yield StreamChunk(text=text)
+                meta = getattr(chunk, "usage_metadata", None)
+                if meta is not None:
+                    yield StreamChunk(usage=_usage_from(meta))
+        except Exception as exc:
+            if _is_transient(exc):
+                raise TransientProviderError(str(exc), provider=self.provider) from exc
+            raise