coderouter-cli 2.0.0__py3-none-any.whl → 2.1.0__py3-none-any.whl

coderouter/routing/fallback.py

@@ -26,7 +26,10 @@ from __future__ import annotations
 
 import time
 from collections.abc import AsyncIterator
-from typing import Final
+from typing import TYPE_CHECKING, Any, Final
+
+if TYPE_CHECKING:
+    from coderouter.guards.drift_detection import DriftVerdict
 
 from coderouter.adapters.anthropic_native import AnthropicAdapter
 from coderouter.adapters.base import (
@@ -469,11 +472,16 @@ class _StreamUsageAccumulator:
     """
 
     __slots__ = (
+        "_current_block_text",
+        "_current_block_type",
         "_observed",
+        "_text_blocks",
         "cache_creation_input_tokens",
         "cache_read_input_tokens",
+        "has_tool_use",
         "input_tokens",
         "output_tokens",
+        "stop_reason",
     )
 
     def __init__(self) -> None:
@@ -482,6 +490,32 @@ class _StreamUsageAccumulator:
         self.cache_read_input_tokens = 0
         self.cache_creation_input_tokens = 0
         self._observed = False
+        # v2.0-G: tracked for drift detection observation at stream end.
+        self.has_tool_use: bool = False
+        self.stop_reason: str | None = None
+        # v2.0-H: partial content accumulation for mid-stream recovery.
+        # Completed text blocks are moved to _text_blocks on content_block_stop.
+        # In-progress text is in _current_block_text (list of str fragments).
+        self._text_blocks: list[str] = []
+        self._current_block_type: str | None = None
+        self._current_block_text: list[str] = []
+
+    @property
+    def partial_content(self) -> list[dict[str, Any]]:
+        """Return accumulated text content as Anthropic content blocks.
+
+        Includes both completed blocks and any in-progress text block
+        (useful when the stream is interrupted mid-block). Tool_use blocks
+        are excluded because partial JSON is unusable.
+        """
+        blocks: list[dict[str, Any]] = []
+        for text in self._text_blocks:
+            if text:
+                blocks.append({"type": "text", "text": text})
+        # Include in-progress text block if any
+        if self._current_block_type == "text" and self._current_block_text:
+            blocks.append({"type": "text", "text": "".join(self._current_block_text)})
+        return blocks
 
     def observe(self, event: AnthropicStreamEvent) -> None:
         """Update counters from one stream event (no-op for non-usage events)."""
@@ -494,6 +528,33 @@ class _StreamUsageAccumulator:
             usage = event.data.get("usage") if isinstance(event.data, dict) else None
             if isinstance(usage, dict):
                 self._merge(usage)
+            # v2.0-G: capture stop_reason from the terminal message_delta.
+            delta = event.data.get("delta") if isinstance(event.data, dict) else None
+            if isinstance(delta, dict) and "stop_reason" in delta:
+                self.stop_reason = delta["stop_reason"]
+        elif event.type == "content_block_start":
+            # v2.0-G: detect tool_use content blocks for drift observation.
+            cb = event.data.get("content_block") if isinstance(event.data, dict) else None
+            if isinstance(cb, dict):
+                block_type = cb.get("type", "")
+                if block_type == "tool_use":
+                    self.has_tool_use = True
+                # v2.0-H: start tracking a new content block.
+                self._current_block_type = block_type
+                self._current_block_text = []
+        elif event.type == "content_block_delta":
+            # v2.0-H: accumulate text_delta fragments.
+            delta = event.data.get("delta") if isinstance(event.data, dict) else None
+            if isinstance(delta, dict) and delta.get("type") == "text_delta":
+                text = delta.get("text", "")
+                if text:
+                    self._current_block_text.append(text)
+        elif event.type == "content_block_stop":
+            # v2.0-H: finalize current block.
+            if self._current_block_type == "text" and self._current_block_text:
+                self._text_blocks.append("".join(self._current_block_text))
+            self._current_block_type = None
+            self._current_block_text = []
 
     def _merge(self, usage: dict[str, object]) -> None:
         any_nonzero = False
@@ -613,9 +674,18 @@ class MidStreamError(CodeRouterError):
     one chunk to the client. Fallback is not attempted (the client has
     received partial content, so switching providers would corrupt the
     stream). Callers should surface this as a terminal error event.
+
+    v2.0-H: carries ``partial_content`` — the accumulated text blocks
+    generated before the failure. The ingress uses this to synthesize
+    a graceful stream termination when ``partial_stitch_action: surface``.
     """
 
-    def __init__(self, provider: str, original: AdapterError) -> None:
+    def __init__(
+        self,
+        provider: str,
+        original: AdapterError,
+        partial_content: list[dict[str, Any]] | None = None,
+    ) -> None:
         """Wrap the underlying :class:`AdapterError` with the provider name.
 
         The ingress layer catches this and converts it into an in-stream
@@ -624,6 +694,7 @@ class MidStreamError(CodeRouterError):
         """
         self.provider = provider
         self.original = original
+        self.partial_content: list[dict[str, Any]] = partial_content or []
         super().__init__(f"provider {provider!r} failed mid-stream: {original}")
 
 
@@ -747,6 +818,32 @@ class FallbackEngine:
         # Distinct from v1.9-C ``adaptive`` which handles the
         # gradient case via a rolling window.
         self._backend_health_monitor: BackendHealthMonitor = BackendHealthMonitor()
+        # v2.0-G (L4): per-process drift detection window manager.
+        # Stores per-provider rolling observations; the detector is
+        # invoked after each provider-ok / provider-failed event and
+        # returns a verdict. Action dispatch (promote/reload) reuses
+        # the adaptive rank machinery.
+        from coderouter.guards.drift_detection import DriftWindow
+
+        self._drift_window: DriftWindow = DriftWindow()
+        # Track which providers are currently in drift-demoted state
+        # and when their cooldown expires (monotonic timestamp).
+        self._drift_demoted: dict[str, float] = {}
+        # Last drift verdict (set by _observe_drift_signal for ingress header).
+        self._last_drift_verdict: DriftVerdict | None = None
+
+    @property
+    def last_drift_severity(self) -> str | None:
+        """Return the severity string of the most recent drift verdict, or None.
+
+        The ingress reads this after generate_anthropic / stream_anthropic to
+        set the ``X-CodeRouter-Drift`` response header.  Returns ``"mild"`` or
+        ``"severe"`` when drift was detected, ``None`` otherwise.
+        """
+        v = self._last_drift_verdict
+        if v is None or not v.drifted:
+            return None
+        return v.severity
 
     @property
     def _adaptive(self) -> AdaptiveAdjuster:
@@ -794,12 +891,17 @@ class FallbackEngine:
         return existing
 
     @property
-    def _backend_health(self) -> BackendHealthMonitor:
+    def backend_health(self) -> BackendHealthMonitor:
         """Return the L5 backend-health monitor, lazily building one if absent.
 
         Same legacy-test compatibility pattern as the other guard
         properties — ``__new__``-constructed engines get a fresh
         empty monitor so ``state_for`` is always answerable.
+
+        v2.0-I: promoted from ``_backend_health`` to public ``backend_health``
+        so the continuous probe background task can feed results into the
+        same state machine. Internal callers continue to work (property
+        access is transparent).
         """
         existing = getattr(self, "_backend_health_monitor", None)
         if existing is None:
@@ -807,6 +909,11 @@ class FallbackEngine:
             existing = self._backend_health_monitor
         return existing
 
+    # Alias for backward compat with internal callers.
+    @property
+    def _backend_health(self) -> BackendHealthMonitor:
+        return self.backend_health
+
     def _observe_provider_failure(
         self,
         provider: str,
@@ -925,6 +1032,132 @@ class FallbackEngine:
                 consecutive_failures=transition.consecutive_failures,
             )
 
+    def _observe_drift_signal(
+        self,
+        provider: str,
+        *,
+        profile: str | None,
+        output_tokens: int = 0,
+        has_tool_use: bool = False,
+        request_had_tools: bool = False,
+        stop_reason: str | None = None,
+        is_error: bool = False,
+        stream: bool = False,
+    ) -> DriftVerdict | None:
+        """v2.0-G (L4): record an observation and check for drift.
+
+        Called after every provider-ok / provider-failed event on the
+        Anthropic-shaped paths. Returns a :class:`DriftVerdict` when
+        drift is detected (drifted=True), None otherwise.
+
+        Side effects on detection:
+        - Emits ``drift-detected`` log.
+        - If action is ``promote`` or ``reload``, demotes the provider
+          via the adaptive rank machinery.
+        """
+        from coderouter.guards.drift_detection import (
+            SENSITIVITY_PRESETS,
+            ResponseObservation,
+            detect_drift,
+        )
+        from coderouter.logging import log_drift_detected, log_drift_promoted
+
+        chosen = profile or self.config.default_profile
+        try:
+            chain_cfg = self.config.profile_by_name(chosen)
+        except (KeyError, ValueError):
+            return None
+        if chain_cfg.drift_detection_action == "off":
+            return None
+
+        # Update window size if config differs from default
+        self._drift_window.max_size = chain_cfg.drift_detection_window_size
+
+        # Record observation
+        obs = ResponseObservation(
+            provider=provider,
+            output_tokens=output_tokens,
+            has_tool_use=has_tool_use,
+            request_had_tools=request_had_tools,
+            stop_reason=stop_reason,
+            is_error=is_error,
+            stream=stream,
+        )
+        self._drift_window.record(obs)
+
+        # Check for cooldown recovery
+        import time as _time
+
+        demote_expires = self._drift_demoted.get(provider)
+        if demote_expires is not None and _time.monotonic() >= demote_expires:
+            # Cooldown expired — restore rank and clear drift state
+            from coderouter.logging import log_drift_recovered
+
+            elapsed = chain_cfg.drift_detection_cooldown_s
+            log_drift_recovered(logger, provider=provider, profile=chosen, after_s=elapsed)
+            self._drift_demoted.pop(provider, None)
+            self._drift_window.clear(provider)
+            return None
+
+        # Don't re-detect while in cooldown
+        if provider in self._drift_demoted:
+            return None
+
+        # Run detection
+        window = self._drift_window.get_window(provider)
+        thresholds = SENSITIVITY_PRESETS.get(
+            chain_cfg.drift_detection_sensitivity, SENSITIVITY_PRESETS["normal"]
+        )
+        verdict = detect_drift(window, thresholds)
+
+        if not verdict.drifted:
+            self._last_drift_verdict = None
+            return None
+
+        # Store for ingress response header.
+        self._last_drift_verdict = verdict
+
+        # Emit log
+        log_drift_detected(
+            logger,
+            provider=provider,
+            profile=chosen,
+            severity=verdict.severity,
+            reason=verdict.reason,
+            action=chain_cfg.drift_detection_action,
+            signals=verdict.signals,
+        )
+
+        # Action: promote / reload
+        if chain_cfg.drift_detection_action in ("promote", "reload"):
+            import time as _time_mod
+
+            # Demote via adaptive rank
+            self._adaptive.demote(provider, steps=2)
+            log_drift_promoted(
+                logger,
+                provider=provider,
+                profile=chosen,
+                demoted_to_rank=2,
+                cooldown_s=chain_cfg.drift_detection_cooldown_s,
+            )
+            # Record cooldown expiry
+            self._drift_demoted[provider] = (
+                _time_mod.monotonic() + chain_cfg.drift_detection_cooldown_s
+            )
+
+            # v2.0-G: reload action — attempt Ollama KV cache flush
+            # (best-effort, fire-and-forget background task).
+            if chain_cfg.drift_detection_action == "reload":
+                import asyncio
+
+                from coderouter.guards.drift_actions import attempt_reload
+
+                provider_config = self._adapters[provider].config
+                self._reload_task = asyncio.create_task(attempt_reload(provider_config))
+
+        return verdict
+
     def _resolve_profile_overrides(self, profile_name: str | None) -> ProviderCallOverrides:
         """v0.6-B: build the ProviderCallOverrides for the active profile.
 
@@ -1455,6 +1688,14 @@ class FallbackEngine:
                 self._observe_provider_failure(
                     adapter.name, exc, profile=request.profile
                 )
+                # v2.0-G (L4): drift detection observation (failure path).
+                self._observe_drift_signal(
+                    adapter.name,
+                    profile=request.profile,
+                    is_error=True,
+                    request_had_tools=bool(request.tools),
+                    stream=False,
+                )
                 errors.append(exc)
                 if not exc.retryable:
                     break
@@ -1482,6 +1723,18 @@ class FallbackEngine:
             self._observe_provider_success(
                 adapter.name, profile=request.profile
             )
+            # v2.0-G (L4): drift detection observation (success path).
+            self._observe_drift_signal(
+                adapter.name,
+                profile=request.profile,
+                output_tokens=resp.usage.output_tokens if resp.usage else 0,
+                has_tool_use=any(
+                    getattr(b, "type", None) == "tool_use" for b in (resp.content or [])
+                ),
+                request_had_tools=bool(request.tools),
+                stop_reason=resp.stop_reason,
+                stream=False,
+            )
             # v1.9-A: pair every successful Anthropic response with a
             # cache-observed log line. Native Anthropic / LM Studio
             # /v1/messages report cache_read_input_tokens /
@@ -1620,6 +1873,14 @@ class FallbackEngine:
                 self._observe_provider_failure(
                     adapter.name, exc, profile=request.profile
                 )
+                # v2.0-G (L4): drift detection observation (stream failure).
+                self._observe_drift_signal(
+                    adapter.name,
+                    profile=request.profile,
+                    is_error=True,
+                    request_had_tools=bool(request.tools),
+                    stream=True,
+                )
                 errors.append(exc)
                 if not exc.retryable:
                     break
@@ -1662,7 +1923,27 @@ class FallbackEngine:
                 self._observe_provider_failure(
                     adapter.name, exc, profile=request.profile
                 )
-                raise MidStreamError(adapter.name, exc) from exc
+                # v2.0-G (L4): drift detection observation (mid-stream failure).
+                self._observe_drift_signal(
+                    adapter.name,
+                    profile=request.profile,
+                    is_error=True,
+                    request_had_tools=bool(request.tools),
+                    stream=True,
+                )
+                raise MidStreamError(
+                    adapter.name, exc, partial_content=acc.partial_content
+                ) from exc
+            # v2.0-G (L4): drift detection observation (stream success).
+            self._observe_drift_signal(
+                adapter.name,
+                profile=request.profile,
+                output_tokens=acc.output_tokens,
+                has_tool_use=acc.has_tool_use,
+                request_had_tools=bool(request.tools),
+                stop_reason=acc.stop_reason,
+                stream=True,
+            )
             # v1.9-B2: pair the successful stream with a cache-observed
             # log line carrying the aggregated usage counters that the
             # ``_StreamUsageAccumulator`` collected from the

{coderouter_cli-2.0.0.dist-info → coderouter_cli-2.1.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coderouter-cli
-Version: 2.0.0
+Version: 2.1.0
 Summary: Local-first, free-first, fallback-built-in LLM router. Claude Code / OpenAI compatible.
 Project-URL: Homepage, https://github.com/zephel01/CodeRouter
 Project-URL: Repository, https://github.com/zephel01/CodeRouter
@@ -60,7 +60,7 @@ Description-Content-Type: text/markdown
 <p align="center">
   <a href="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml"><img src="https://github.com/zephel01/CodeRouter/actions/workflows/ci.yml/badge.svg?branch=main" alt="CI"></a>
   <a href=""><img src="https://img.shields.io/badge/status-stable-brightgreen" alt="status"></a>
-  <a href=""><img src="https://img.shields.io/badge/version-2.0.0-blue" alt="version"></a>
+  <a href=""><img src="https://img.shields.io/badge/version-2.1.0-blue" alt="version"></a>
   <a href=""><img src="https://img.shields.io/badge/python-3.12%2B-blue" alt="python"></a>
   <a href=""><img src="https://img.shields.io/badge/runtime%20deps-5-brightgreen" alt="deps"></a>
   <a href=""><img src="https://img.shields.io/badge/license-MIT-yellow" alt="license"></a>
@@ -91,7 +91,8 @@ Description-Content-Type: text/markdown
 - **v1.10.0 で Long-run reliability pillar が完成**: `cost.monthly_budget_usd` で provider 月次 USD 予算を強制、**L2 memory pressure detector**（Ollama / LM Studio が VRAM 切れで OOM になった時に自動クールダウン）、**L5 backend health 状態機械**（連続失敗で UNHEALTHY → chain 末尾に降格、1 回成功で即復帰）
 - **v1.10.0 で auto-router が 6 matcher に揃う**: `has_image` / `code_fence_ratio_min` / `content_contains` / `content_regex` / `model_pattern`（Opus/Sonnet/Haiku 分岐）/ `content_token_count_min`（長文 → 1M ctx Gemini Flash 等へ自動切替）
 - **v2.0.0 で Context Budget Management (L1 overflow 防止) を搭載**: 長時間 agent session で messages が context window に漸近 → backend 400 エラーで session 死亡する問題を根本解決。warn (80%) → auto trim (90%) の 2 段階 guard で **context overflow ゼロ**を実現。tool_use / tool_result ペアは atomic 保全、`X-CodeRouter-Context-Budget` ヘッダで状態通知、Prometheus メトリクス完備
-- ランタイム依存 5 個（`fastapi` / `uvicorn` / `httpx` / `pydantic` / `pyyaml`）— 純 Python、MIT、テスト 930 本緑
+- **v2.1.0 で Long-run Reliability 3 機能を追加搭載**: **Drift Detection** (L4 品質劣化検知 — 5 シグナル rolling window + warn/promote/reload 3 段階アクション)、**Partial Stitching** (L6 mid-stream 失敗時の蓄積テキスト返却)、**Continuous Probing** (P3 idle 時 1-token 定期 probe + model drift 検知 + backend health 自動更新)
+- ランタイム依存 5 個（`fastapi` / `uvicorn` / `httpx` / `pydantic` / `pyyaml`）— 純 Python、MIT、テスト 950 本緑
 
 → **Claude Code / gemini-cli / codex + Ollama / llama.cpp / NVIDIA NIM で、破綻しない local-first agent が組める**
 
@@ -198,7 +199,7 @@ CodeRouter / Voice Bridge ともに独立した repo で進化していて、HTT
 
 ## クイックスタート（3 コマンド）
 
-**v2.0.0 で Context Budget Management (L1 overflow 防止) を搭載** — 長時間 agent session が context window を使い切って死ぬ問題を根本解決。`uvx` 一発で動きます (Python 3.12 以上必須):
+**v2.1.0 で Long-run Reliability pillar 完成** — Context Budget (L1) + Drift Detection (L4) + Partial Stitching (L6) + Continuous Probing (P3)。`uvx` 一発で動きます (Python 3.12 以上必須):
 
 ```bash
 # 1. サンプル設定を置く
@@ -266,9 +267,9 @@ CodeRouter 自体は純 Python 3.12+ で、実質的な OS 対応範囲は `min(
 
 注意点や「ローカル GPU なし」向けレシピを含むフル版マトリクス: [利用ガイド §1](./docs/usage-guide.md#1-os-互換性)
 
-## ステータス — v2.0.0 (2026-05)
+## ステータス — v2.1.0 (2026-05)
 
-**テスト 930 本通過。ランタイム依存 5 個 (36 sub-release 連続据え置き)。macOS / Linux / Windows WSL2 で動作。** v2.0.0 で **Context Budget Management (L1 overflow 防止)** を搭載 — 長時間 agent session が context window を使い切って死ぬ問題を根本解決。v1.10.0 までに **Long-run Reliability** (L2/L3/L5)、**Cost pillar**、**Auto-router 6 matcher** が完成済み。v1.0 の総まとめは [`docs/retrospectives/v1.0.md`](./docs/retrospectives/v1.0.md)。
+**テスト 950 本通過。ランタイム依存 5 個 (39 sub-release 連続据え置き)。macOS / Linux / Windows WSL2 で動作。** v2.1.0 で **Long-run Reliability pillar が完成** — Context Budget (L1) / Drift Detection (L4) / Partial Stitching (L6) / Continuous Probing (P3) の 4 sub-release を統合出荷。v1.10.0 までに **Long-run Reliability** (L2/L3/L5)、**Cost pillar**、**Auto-router 6 matcher** が完成済み。v1.0 の総まとめは [`docs/retrospectives/v1.0.md`](./docs/retrospectives/v1.0.md)。
 
 今日の CodeRouter が届ける価値:
 

{coderouter_cli-2.0.0.dist-info → coderouter_cli-2.1.0.dist-info}/RECORD

@@ -7,7 +7,7 @@ coderouter/doctor.py,sha256=2luNk6BHSRvpQStJnHcqzNvNi-SKdOuKV0WZdorZhVk,82854
 coderouter/doctor_apply.py,sha256=r_J6xbu5-HivofPNriw4_vjNYs_VRs7GsGTS0oMEX10,24209
 coderouter/env_security.py,sha256=FEBZnXfJ0xE39kmMMn39zk0W_DRRnmcB_REmP9f4xWo,14796
 coderouter/errors.py,sha256=Xmq67lheyw8iv3Ox39jh2c4tvNI5RcUR4QkoxVDN6l4,1130
-coderouter/logging.py,sha256=tHG3V7N6GJFI4TUc5LKrBjlLFgRfRkmZw1UKMlPOmhg,42272
+coderouter/logging.py,sha256=DV1BwLsQK3wa8493omij5hiG7lU0WTnJFk9xA-ei5Zc,49831
 coderouter/output_filters.py,sha256=rI4YgKVv5vviDBl3Xkf7rp6LaSSkdWyEV004q6HrkB0,15706
 coderouter/token_estimation.py,sha256=1Ai1uT68hahpyr4LBhNyVRGq7y4yXItd6J4k5ApGX7M,5995
 coderouter/adapters/__init__.py,sha256=7dIDSZ-FE_0iSqLSDc_lK1idRdLTKcM2hP9tCJipgPI,463
@@ -19,35 +19,38 @@ coderouter/config/__init__.py,sha256=FODEn74fN-qZnt4INPSHswqhOlEgpL6-_onxsitSx8g
 coderouter/config/capability_registry.py,sha256=F6DetVL5oM03R4QeK1g6h_Q_zrXH0opnYDp3duZmkN4,15808
 coderouter/config/env_file.py,sha256=CoMK27fuAXm-NtoLzXb8yN2E-wDFjHQuFwiIlmgTBQw,10356
 coderouter/config/loader.py,sha256=FUEe8m4Tnmj_aul0vSctD8vKvNW-oLRoMRbTpSKqSmc,4077
-coderouter/config/schemas.py,sha256=pXWYDpW9p5QWRSA9pPNZ3ukUySf3ulfX8hfhSejhB_Q,42044
+coderouter/config/schemas.py,sha256=F0xHOnn9knwuaYNBL2sKOwum08mvKFP57JWKkS_-Sho,46407
 coderouter/data/__init__.py,sha256=uNyfD9jaCvTWsBAWtaw1Fr25OSxzv3psGMfBjT1z0Cc,328
 coderouter/data/model-capabilities.yaml,sha256=2ztY4PUbGN_cWG-UUB-iPy-baeVFnGV8OcZHJUfZE7c,19290
 coderouter/guards/__init__.py,sha256=eYjuKo0OvE-GjJo7drYtU2XavUPF9OdiTB5IS76c92Y,855
 coderouter/guards/backend_health.py,sha256=bW148sadK60ZHaz9n6HKeRMcWxfAEvieBVGQhGh_MwY,7686
 coderouter/guards/context_budget.py,sha256=moWulVr5NtVci13vXxS0ucV4EvX2b7tbA1W1d9eQnkw,13281
+coderouter/guards/continuous_probe.py,sha256=AKNMbJ7hUJG-FDoU160BCbSEQQUyw0hBxFYMTaBZg84,11681
+coderouter/guards/drift_actions.py,sha256=A6pY5CR480Ct5rCVyjlBvjPFVc93eu_r5qcUpK9mWKc,3602
+coderouter/guards/drift_detection.py,sha256=W6f9yhVZO-J4PCeyeM9MIc40BY0tYoHGewTPU0eze0g,11011
 coderouter/guards/memory_pressure.py,sha256=mul1KXO9oE1i424cs92Sk6uzoRrV6Seck2Lk3bu-w68,7903
 coderouter/guards/tool_loop.py,sha256=QRUh0fM9LJx-KFd4751dRp2Tv_8ewjcXaW-TIeWsfJQ,12724
 coderouter/ingress/__init__.py,sha256=WQsCH2CGJCAhy0mS6GSEdeYZRkkQu2OHDsP4CJWTLug,155
-coderouter/ingress/anthropic_routes.py,sha256=eGS-z7hi_UMni8PtEWylIPuJs6YdjoKAPhMNaq-xptQ,12924
-coderouter/ingress/app.py,sha256=b9s6NpBGckaTQJCaw-6MZaY7Eq3rxgtX_cCkgWq-0Ig,6620
+coderouter/ingress/anthropic_routes.py,sha256=pfk0rWWH5IJvdPF36h6egxYvX40RBrP5yTVfzOJArcw,15726
+coderouter/ingress/app.py,sha256=E9Ze1hCWeZVqhh1DrDjrc1uJpjmjoHMMfv89tbReO6g,8031
 coderouter/ingress/dashboard_routes.py,sha256=jVFc_4Q67YrP0VIppuQfBLnLvnvaCWJJ4z7aZsHmu1s,21822
 coderouter/ingress/metrics_routes.py,sha256=M22dwOGn24P05Ge4W3c7d7mYytSGWjIR-pPSPOAiHJY,3965
 coderouter/ingress/openai_routes.py,sha256=Zw1efPw9DI6GgV8ZcLrzS6Cda0KLrFkKn2GBZWSe6Vo,6322
 coderouter/metrics/__init__.py,sha256=7Es351DPS7yLM0yVF_F0eesmiD83n7Zzhie44chht38,1465
-coderouter/metrics/collector.py,sha256=BOQcmD3UUPH17J6kONmeyldvuLQyxQKLShsbb2sV4sI,38501
-coderouter/metrics/prometheus.py,sha256=2uyr0AxzL2i6RQAr9ODeY-TkDXkaAdJbfrtT1KRC9B4,19192
+coderouter/metrics/collector.py,sha256=JbFqXex9VzInhPIhHcPtuXeGR1tXsMD2r7o0uF2qLuw,43717
+coderouter/metrics/prometheus.py,sha256=YRqyT931s40zVkIj07D-M2UNfDhIEElVFRz3izdJcnQ,24419
 coderouter/routing/__init__.py,sha256=g2vhutbozRx5QBThReqwPN3imk5qXdpDiaogILd3IRc,257
-coderouter/routing/adaptive.py,sha256=pRUphp2_NJ7i6I2HRNk3AkB7Wiu0amZ3vtgBHgmVLBU,20233
+coderouter/routing/adaptive.py,sha256=G2o377twGSjbUh65wiIFx6klnpFGjsD_nI3oDvcBwhY,21257
 coderouter/routing/auto_router.py,sha256=4_sQR0ztSED9FgQSvQqgqSiydyQVY_qOSRvwyZ5BfRc,12909
 coderouter/routing/budget.py,sha256=XZQL24YIkusB2MwfgbIJ1uhW1ODtMMdpbZ5e_A7oSU4,7164
 coderouter/routing/capability.py,sha256=ziIDuE5keH_jxYDlXSKufRVxxSYOAvUxJ6Rw5QkYDDU,18436
-coderouter/routing/fallback.py,sha256=Rj5RRb5_EaxkbnMft8hHjlswn4PtPMLOawSrQk_B9C4,74294
+coderouter/routing/fallback.py,sha256=cJW_LYghVKVHOsHmJFaGgEEAkJ4ZsfhFtstOODOI4y8,85983
 coderouter/translation/__init__.py,sha256=PYXN7XVEwpG1uC8RLy6fvnGbzEZhhrEuUapH8IYOtG8,1788
 coderouter/translation/anthropic.py,sha256=JpvIWNXHUPVqOGvps7o_6ZADhXuJuvpU7RdMqQFtwwM,6421
 coderouter/translation/convert.py,sha256=-qyzFzmmr9hhQV6_Sg75kJnvCZvHe3n7vRdaZtk_JqQ,47269
 coderouter/translation/tool_repair.py,sha256=fyxDb4kWHytO5JWq5y0i4tinJUtWqhMCkyfoCf5BjeM,8314
-coderouter_cli-2.0.0.dist-info/METADATA,sha256=rCenT_ZMrkR1GWvJHvs5cNG7sJEbonjdh40OKPURYS0,48192
-coderouter_cli-2.0.0.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
-coderouter_cli-2.0.0.dist-info/entry_points.txt,sha256=-dnLfD1YZ2WjH2zSdNCvlO65wYltM9bsHt9Fhg3yGss,51
-coderouter_cli-2.0.0.dist-info/licenses/LICENSE,sha256=wkEzoR86jFw33jvfOHjULqmkGEfxTFMgMaJnpR8mPRw,1065
-coderouter_cli-2.0.0.dist-info/RECORD,,
+coderouter_cli-2.1.0.dist-info/METADATA,sha256=5Jyg_kLzgs6PAuzyDdqjIVbabeUOwqu2k8c5V0ZTLrc,48570
+coderouter_cli-2.1.0.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
+coderouter_cli-2.1.0.dist-info/entry_points.txt,sha256=-dnLfD1YZ2WjH2zSdNCvlO65wYltM9bsHt9Fhg3yGss,51
+coderouter_cli-2.1.0.dist-info/licenses/LICENSE,sha256=wkEzoR86jFw33jvfOHjULqmkGEfxTFMgMaJnpR8mPRw,1065
+coderouter_cli-2.1.0.dist-info/RECORD,,