core: Registry, Skills, Router, Executor, Verifier, Runner mit Tests

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

core/src/knaster/llm.py

@@ -0,0 +1,71 @@
+"""Chat-Client für das OpenAI-kompatible API von LiteLLM."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from typing import Any, Protocol
+
+import httpx
+
+from .problems import is_context_overflow_message
+
+
+class LLMError(Exception):
+    """Basisklasse; `infra=True` bedeutet: Endpunkt/Netz, nicht Modellqualität."""
+
+    infra = True
+
+
+class EndpointError(LLMError):
+    pass
+
+
+class ContextOverflowError(LLMError):
+    pass
+
+
+@dataclass
+class ChatResult:
+    content: str
+    finish_reason: str | None = None
+    usage: dict[str, Any] = field(default_factory=dict)
+    model: str | None = None
+
+
+class ChatClient(Protocol):
+    def chat(self, model: str, messages: list[dict[str, str]], *, max_tokens: int | None = None,
+             temperature: float | None = None) -> ChatResult: ...
+
+
+class OpenAIChatClient:
+    def __init__(self, base_url: str, api_key: str = "", timeout: float = 120.0,
+                 client: httpx.Client | None = None):
+        self.base_url = base_url.rstrip("/")
+        self.headers = {"Authorization": f"Bearer {api_key}"} if api_key else {}
+        self.client = client or httpx.Client(timeout=timeout)
+
+    def chat(self, model: str, messages: list[dict[str, str]], *, max_tokens: int | None = None,
+             temperature: float | None = None) -> ChatResult:
+        body: dict[str, Any] = {"model": model, "messages": messages}
+        if max_tokens is not None:
+            body["max_tokens"] = max_tokens
+        if temperature is not None:
+            body["temperature"] = temperature
+        try:
+            resp = self.client.post(f"{self.base_url}/v1/chat/completions", json=body, headers=self.headers)
+        except httpx.HTTPError as exc:
+            raise EndpointError(f"{type(exc).__name__}: {exc}") from exc
+        if resp.status_code == 400 and is_context_overflow_message(resp.text):
+            raise ContextOverflowError(resp.text[:500])
+        if resp.status_code == 429 or resp.status_code >= 500:
+            raise EndpointError(f"HTTP {resp.status_code}: {resp.text[:300]}")
+        if resp.status_code >= 400:
+            raise EndpointError(f"HTTP {resp.status_code}: {resp.text[:300]}")
+        data = resp.json()
+        choice = data["choices"][0]
+        return ChatResult(
+            content=choice["message"].get("content") or "",
+            finish_reason=choice.get("finish_reason"),
+            usage=data.get("usage") or {},
+            model=data.get("model"),
+        )