crabcode 0.1.6__py3-none-win_amd64.whl

crabcode_core/api/error_logging.py

@@ -0,0 +1,132 @@
+"""Provider error diagnostics shared by CLI and Gateway-backed clients."""
+
+from __future__ import annotations
+
+import json
+import os
+import re
+from typing import Any
+
+from crabcode_core.logging_utils import get_logger
+from crabcode_core.utf8_sanitize import safe_utf8_str
+
+logger = get_logger(__name__)
+
+_CREDENTIAL_KEYS = {
+    "authorization", "proxyauthorization", "apikey", "xapikey", "token",
+    "accesstoken", "refreshtoken", "idtoken", "password", "secret",
+    "clientsecret", "cookie", "setcookie",
+}
+
+
+def _credential_key(key: str) -> bool:
+    return re.sub(r"[^a-z0-9]", "", key.lower()) in _CREDENTIAL_KEYS
+
+
+def response_request_id(response: Any) -> str | None:
+    headers = getattr(response, "headers", {})
+    return next((headers.get(key) for key in (
+        "x-request-id", "request-id", "x-amzn-requestid",
+    ) if headers.get(key)), None)
+
+
+def _exception_details(exc: BaseException) -> list[dict[str, Any]]:
+    details = []
+    pending = [exc]
+    seen: set[int] = set()
+    while pending:
+        current = pending.pop()
+        if id(current) in seen:
+            continue
+        seen.add(id(current))
+        response = getattr(current, "response", None)
+        body = getattr(current, "body", None)
+        # HTTP errors may contain an envelope beyond SDK exception.body.
+        # Only inspect an already-read response; never consume a live stream.
+        if response is not None:
+            try:
+                body = response.text or body
+            except Exception:
+                pass
+        details.append({
+            "exception": type(current).__name__,
+            # SDK messages often repeat body as a Python repr, which would
+            # bypass structured credential redaction. The body retains it.
+            "message": str(current) if body is None else None,
+            "status_code": getattr(current, "status_code", None)
+            or getattr(response, "status_code", None),
+            "request_id": getattr(current, "request_id", None)
+            or response_request_id(response),
+            "body": body,
+        })
+        pending.extend(item for item in (current.__context__, current.__cause__)
+                       if item is not None)
+    return details
+
+
+def _redact(value: Any, secrets: set[str]) -> Any:
+    if hasattr(value, "model_dump"):
+        value = value.model_dump(mode="json")
+    if isinstance(value, dict):
+        return {str(key): "[REDACTED]" if _credential_key(str(key))
+                else _redact(item, secrets) for key, item in value.items()}
+    if isinstance(value, (list, tuple)):
+        return [_redact(item, secrets) for item in value]
+    if isinstance(value, bytes):
+        value = value.decode("utf-8", "replace")
+    if isinstance(value, str):
+        # Raw HTTP bodies can be JSON too. Preserve every diagnostic field.
+        try:
+            parsed = json.loads(value)
+        except (ValueError, TypeError):
+            parsed = None
+        if isinstance(parsed, (dict, list)):
+            return _redact(parsed, secrets)
+        for secret in sorted(secrets, key=len, reverse=True):
+            value = value.replace(secret, "[REDACTED]")
+        return re.sub(r"(?i)\bBearer\s+[^\s\"'<>]+", "Bearer [REDACTED]", value)
+    return value
+
+
+def log_api_error(
+    adapter: Any, *, session_id: str | None, model: str, attempt_id: str,
+    purpose: str, exception: BaseException | None = None, chunk: Any = None,
+) -> None:
+    """Log the full error body without dumping requests or credential headers.
+
+    Diagnostics must never replace the original error or change retry behavior.
+    Bodies are not truncated; credentials are the only redacted fields.
+    """
+    try:
+        config = getattr(adapter, "config", None)
+        secrets = set()
+        for name in (getattr(config, "api_key_env", None),
+                     "OPENAI_API_KEY", "ANTHROPIC_API_KEY"):
+            if name and os.environ.get(name):
+                secrets.add(os.environ[name])
+        for value in (getattr(adapter, "_api_key", None),
+                      getattr(getattr(adapter, "client", None), "api_key", None)):
+            if isinstance(value, str) and value:
+                secrets.add(value)
+        for key, value in (getattr(config, "http_headers", None) or {}).items():
+            if _credential_key(key) and value:
+                secrets.add(value)
+                if value.lower().startswith("bearer "):
+                    secrets.add(value[7:])
+        details = {
+            "session_id": session_id, "model": model,
+            "provider": getattr(config, "provider", None) or type(adapter).__name__,
+            "attempt_id": attempt_id, "purpose": purpose,
+            "errors": _exception_details(exception) if exception is not None else [{
+                "message": chunk.error if chunk.error_body is None else None,
+                "status_code": chunk.error_status_code,
+                "request_id": chunk.error_request_id,
+                "body": chunk.error_body,
+            }],
+        }
+        logger.warning("API error details: %s", safe_utf8_str(json.dumps(
+            _redact(details, secrets), ensure_ascii=False, default=str,
+        )))
+    except Exception:
+        logger.warning("Unable to serialize API error details: session=%s attempt=%s",
+                       session_id, attempt_id)

crabcode_core/api/gemini_adapter.py

@@ -0,0 +1,26 @@
+"""Gemini API adapter — uses Google's OpenAI-compatible endpoint."""
+
+from __future__ import annotations
+
+from crabcode_core.api.openai_adapter import OpenAIAdapter
+from crabcode_core.types.config import ApiConfig
+
+
+class GeminiAdapter(OpenAIAdapter):
+    """Adapter for Google Gemini via its OpenAI-compatible endpoint.
+
+    Google Gemini has provided an OpenAI-compatible endpoint since late 2024
+    (``https://generativelanguage.googleapis.com/v1beta/openai/``), so we
+    simply inherit from :class:`OpenAIAdapter` and pre-fill the base URL,
+    API key environment variable, and disable thinking (not supported on the
+    OpenAI-compatible surface).
+    """
+
+    def __init__(self, config: ApiConfig):
+        if not config.base_url:
+            config = config.model_copy(update={"base_url": "https://generativelanguage.googleapis.com/v1beta/openai/"})
+        if not config.api_key_env:
+            config = config.model_copy(update={"api_key_env": "GEMINI_API_KEY"})
+        if config.thinking_enabled is True:
+            config = config.model_copy(update={"thinking_enabled": False})
+        super().__init__(config)

crabcode_core/api/model_info.py

@@ -0,0 +1,257 @@
+"""Known model context window sizes and resolution helpers."""
+
+from __future__ import annotations
+
+# Maps model ID (or prefix) to context window size in tokens.
+# Used as fallback when the API doesn't provide model metadata.
+KNOWN_CONTEXT_WINDOWS: dict[str, int] = {
+    # Anthropic
+    "claude-mythos-5-1": 1_000_000,
+    "claude-fable-5-1": 1_000_000,
+    "claude-mythos-preview": 1_000_000,
+    "claude-mythos-5": 1_000_000,
+    "claude-fable-5": 1_000_000,
+    "claude-opus-5-5": 1_000_000,
+    "claude-sonnet-5-5": 1_000_000,
+    "claude-haiku-5-5": 1_000_000,
+    "claude-opus-5": 1_000_000,
+    "claude-sonnet-5": 1_000_000,
+    "claude-opus-4-8": 1_000_000,
+    "claude-opus-4-7": 1_000_000,
+    "claude-opus-4-6": 1_000_000,
+    "claude-sonnet-4-6": 1_000_000,
+    "claude-opus-4-5": 200_000,
+    "claude-sonnet-4-5": 200_000,
+    "claude-haiku-4-5": 200_000,
+    "claude-sonnet-4-20250514": 200_000,
+    "claude-opus-4-20250514": 200_000,
+    "claude-haiku-3-5-20241022": 200_000,
+    "claude-3-5-sonnet-20241022": 200_000,
+    "claude-3-5-haiku-20241022": 200_000,
+    "claude-3-opus-20240229": 200_000,
+    "claude-3-sonnet-20240229": 200_000,
+    "claude-3-haiku-20240307": 200_000,
+    # OpenAI
+    "gpt-daybreak-blue-latest": 1_050_000,
+    "gpt-daybreak-red-latest": 400_000,
+    "gpt-6-astra": 1_050_000,
+    "gpt-6.1-sol": 1_050_000,
+    "gpt-6-sol": 1_050_000,
+    "gpt-6-luna": 1_050_000,
+    "gpt-5.6-cyber": 400_000,
+    "gpt-5.6-sol": 1_050_000,
+    "gpt-5.6-terra": 1_050_000,
+    "gpt-5.6-luna": 1_050_000,
+    "gpt-5.6": 1_050_000,
+    "gpt-5.5-pro": 1_050_000,
+    "gpt-5.5": 1_050_000,
+    "gpt-5.4-mini": 400_000,
+    "gpt-5.4-nano": 400_000,
+    "gpt-5.4-pro": 1_050_000,
+    "gpt-5.4": 1_050_000,
+    "gpt-5.3-chat-latest": 128_000,
+    "gpt-5.3-codex": 400_000,
+    "gpt-5.2-chat-latest": 128_000,
+    "gpt-5.2-codex": 400_000,
+    "gpt-5.2-pro": 400_000,
+    "gpt-5.2": 400_000,
+    "gpt-5.1-chat-latest": 128_000,
+    "gpt-5.1-codex-mini": 400_000,
+    "gpt-5.1-codex-max": 400_000,
+    "gpt-5.1-codex": 400_000,
+    "gpt-5.1": 400_000,
+    "gpt-5-chat-latest": 128_000,
+    "gpt-5-codex": 400_000,
+    "gpt-5-mini": 400_000,
+    "gpt-5-nano": 400_000,
+    "gpt-5-pro": 400_000,
+    "gpt-5": 400_000,
+    "gpt-4.5-preview": 128_000,
+    "gpt-4.1-mini": 1_047_576,
+    "gpt-4.1-nano": 1_047_576,
+    "gpt-4.1": 1_047_576,
+    "gpt-4o-mini-realtime-preview": 16_000,
+    "gpt-4o-mini-transcribe": 16_000,
+    "gpt-4o-realtime-preview": 32_000,
+    "gpt-4o-transcribe-diarize": 16_000,
+    "gpt-4o-transcribe": 16_000,
+    "gpt-4o-mini": 128_000,
+    "gpt-4o": 128_000,
+    "gpt-4-turbo": 128_000,
+    "gpt-4": 8_192,
+    "o1-preview": 128_000,
+    "o1-mini": 128_000,
+    "o1-pro": 200_000,
+    "o1": 200_000,
+    "o3-deep-research": 200_000,
+    "o3-mini": 200_000,
+    "o3-pro": 200_000,
+    "o3": 200_000,
+    "o4-mini-deep-research": 200_000,
+    "o4-mini": 200_000,
+    "codex-mini-latest": 200_000,
+    # xAI
+    "grok-4.7": 500_000,
+    "grok-4.20": 1_000_000,
+    "grok-4.6": 500_000,
+    "grok-4.5": 500_000,
+    "grok-4.3": 1_000_000,
+    "grok-build-latest": 500_000,
+    "grok-build-0.1": 256_000,
+    "grok-code-fast": 256_000,
+    # DeepSeek
+    "deepseek-flash": 1_000_000,
+    "deepseek-v4-pro": 1_000_000,
+    "deepseek-v4-flash": 1_000_000,
+    "deepseek-v4.1": 1_000_000,
+    "deep-research-preview-04-2026": 1_048_576,
+    "deep-research-max-preview-04-2026": 1_048_576,
+    "antigravity-preview-09-2026": 1_048_576,
+    "deepseek-v4.1-flash": 1_000_000,
+    "deepseek-chat": 128_000,
+    "deepseek-reasoner": 128_000,
+    # GLM (Zhipu)
+    "glm-5.3-flash": 1_000_000,
+    "glm-5.3": 1_000_000,
+    "glm-5.2": 1_000_000,
+    "glm-5.1-fp8": 202_752,
+    "glm-5.1": 202_752,
+    "glm-4-plus": 128_000,
+    "glm-4": 128_000,
+    # Other OpenAI-compatible models
+    "qwen3.8-max": 1_000_000,
+    "qwen3.8-flash": 1_000_000,
+    "qwen3.7-plus": 1_000_000,
+    "qwen3.7-flash": 1_000_000,
+    "qwen3.6-plus": 1_000_000,
+    "qwen3.6-flash": 1_000_000,
+    "qwen3.5-plus": 1_000_000,
+    "qwen3.5-flash": 1_000_000,
+    "kimi-k3": 1_000_000,
+    "kimi-k2.7-code": 262_144,
+    "kimi-k2.6": 262_144,
+    "kimi-k2.5": 262_144,
+    "moonshot-v1-128k": 128_000,
+    "moonshot-v1-32k": 32_000,
+    "moonshot-v1-8k": 8_000,
+    "MiniMax-M3": 1_000_000,
+    "MiniMax-M2.7-highspeed": 204_800,
+    "MiniMax-M2.7": 204_800,
+    "minimax-m2.7": 204_800,
+    "skyclaw-v1-lite": 1_000_000,
+    "skyclaw-v1": 1_000_000,
+    # Meta
+    "Llama-4-Maverick-17B-128E-Instruct": 1_000_000,
+    "Llama-4-Scout-17B-16E-Instruct": 10_000_000,
+    "Llama-3.3": 128_000,
+    "Llama-3.2": 128_000,
+    "Llama-3.1": 128_000,
+    "Llama-3": 8_000,
+    "Llama-2": 4_000,
+    "Llama-1": 2_048,
+    "Llama-4-Maverick-17B-128E-Instruct": 1_000_000,
+    "Llama-4-Scout-17B-16E-Instruct": 10_000_000,
+    "muse-spark-1.3-contributor": 1_000_000,
+    "muse-spark-1.3": 1_000_000,
+    "mimo-v2.6-pro": 1_000_000,
+    "mimo-v2.6-pro-ultraspeed": 1_000_000,
+    "mimo-v2.6-flash": 1_000_000,
+    "mimo-v2.5": 1_000_000,
+    "mimo-v2.5-pro": 1_000_000,
+    "mimo-v2.5-asr": 8_000,
+    "mimo-v2.5-tts": 8_000,
+    "mimo-v2-flash": 256_000,
+    "mimo-v2-pro": 1_000_000,
+    "mimo-v2-omni": 256_000,
+    "mimo-7b": 32_768,
+    "IFM/K2-Horizon-375B-A23B": 524_288,
+    "IFM/K2-Horizon-MoVA-36B-A4B": 524_288,
+    "IFM/K2-Horizon-32B": 524_288,
+    "IFM/K2-Horizon-7B-Uno": 524_288,
+    "IFM/K2-Horizon-7B": 524_288,
+    "IFM/K2-Horizon-3.7B": 524_288,
+    "IFM/K2-Horizon-0.9B": 131_072,
+    "thinkingmachines/Inkling": 1_048_576,
+    "thinkingmachines/Inkling-Small": 1_048_576,
+    "inkling": 1_048_576,
+    "inkling-small": 1_048_576,
+    "NVIDIA-Nemotron-3-Ultra-550B-A55B": 1_000_000,
+    "NVIDIA-Nemotron-3.5-Lightning-30B-A3B": 1_000_000,
+    "NVIDIA-Nemotron-3-Nano-30B-A3B": 1_000_000,
+    "Nemotron-Cascade-2-30B-A3B": 1_000_000,
+    "NVIDIA-Nemotron-Nano-12B-v2-VL": 128_000,
+    "NVIDIA-Nemotron-Nano-12B-v2": 128_000,
+    "Llama-3.1-Nemotron-Nano-4B-v1.1": 131_072,
+    "Nemotron-H-4B-Instruct-128K": 128_000,
+    "Nemotron-Mini-4B-Instruct": 4_096,
+    "Nemotron-4-340B-Instruct": 4_096,
+    "nemotron-3-8b-base-4k": 4_096,
+    # Mistral
+    "mistral-medium-3-5": 256_000,
+    "mistral-medium-3": 256_000,
+    "mistral-medium-latest": 256_000,
+    "mistral-small-2603": 256_000,
+    "mistral-small-latest": 256_000,
+    "mistral-large-2512": 256_000,
+    "mistral-large-latest": 256_000,
+    "Mixtral-8x22B-Instruct-v0.1": 64_000,
+    "Mixtral-8x22B-v0.1": 64_000,
+    "Mixtral-8x7B-Instruct-v0.1": 32_000,
+    "Mixtral-8x7B-v0.1": 32_000,
+    "Mixtral-8x22B-Instruct-v0.1": 64_000,
+    "Mixtral-8x22B-v0.1": 64_000,
+    "Mixtral-8x7B-Instruct-v0.1": 32_000,
+    "Mixtral-8x7B-v0.1": 32_000,
+    "open-mixtral-8x22b": 64_000,
+    "open-mixtral-8x7b": 32_000,
+    # Ollama
+    "qwen3:32b": 128_000,
+    "qwen2.5-coder:32b": 128_000,
+    "deepseek-coder-v2": 128_000,
+    "llama3.1:8b": 128_000,
+    "codellama": 16_000,
+    "mistral": 32_000,
+    "mixtral": 32_000,
+    # Gemini
+    "gemini-3.8-live-extended-thinking": 131_072,
+    "gemini-3.8-live": 131_072,
+    "gemini-3.8-flash": 1_048_576,
+    "gemini-3.8-flash-tts": 1_048_576,
+    "gemini-3.8-flash-lite-tts": 1_048_576,
+    "gemini-nano-banana-2.1": 1_048_576,
+    "gemini-3.7-flash": 1_048_576,
+    "gemini-3.6-flash": 1_048_576,
+    "gemini-3.5-flash-lite": 1_048_576,
+    "gemini-3.5-flash": 1_048_576,
+    "gemini-3.1-pro": 1_048_576,
+    "gemini-3.1-flash-lite": 1_048_576,
+    "gemini-3-flash": 1_048_576,
+    "gemini-2.5-pro": 1_048_576,
+    "gemini-2.5-flash-lite": 1_048_576,
+    "gemini-2.5-flash": 1_048_576,
+    "gemini-2.0-flash-lite": 1_048_576,
+    "gemini-2.0-flash": 1_048_576,
+    "gemini-1.5-pro": 2_097_152,
+    "gemini-1.5-flash": 1_048_576,
+}
+
+DEFAULT_CONTEXT_WINDOW = 200_000
+
+
+def lookup_context_window(model: str | None) -> int | None:
+    """Look up the context window for a model ID.
+
+    Tries exact match first, then prefix matching for versioned model IDs
+    (e.g. "gpt-4o-2024-11-20" falls back to "gpt-4o").
+    """
+    if not model:
+        return None
+
+    if model in KNOWN_CONTEXT_WINDOWS:
+        return KNOWN_CONTEXT_WINDOWS[model]
+
+    for known_model in sorted(KNOWN_CONTEXT_WINDOWS, key=len, reverse=True):
+        if model.startswith((f"{known_model}-", f"{known_model}:")):
+            return KNOWN_CONTEXT_WINDOWS[known_model]
+
+    return None

crabcode_core/api/network.py

@@ -0,0 +1,62 @@
+"""Shared HTTP routing policy. Never fall back from a proxy to direct access."""
+from __future__ import annotations
+
+import ssl
+from typing import Any
+from urllib.parse import urlsplit
+
+import httpx
+
+
+def http_options(config: Any) -> dict[str, Any]:
+    mode = getattr(config, "network_mode", "inherit")
+    if mode == "inherit":
+        return {}
+    if mode == "direct":
+        return {"trust_env": False}
+    proxy = getattr(config, "proxy_url", None)
+    try:
+        parsed = urlsplit(proxy or "")
+        valid = parsed.scheme in {"http", "https"} and bool(parsed.hostname) and parsed.port != 0
+        valid = valid and not parsed.username and not parsed.password
+    except ValueError:
+        valid = False
+    if not valid:
+        raise ValueError("Proxy mode requires an HTTP(S) proxy URL without embedded credentials")
+    return {"trust_env": False, "proxy": proxy}
+
+
+def request_timeout(config: Any) -> httpx.Timeout:
+    timeout = float(getattr(config, "timeout", 300))
+    return httpx.Timeout(timeout, connect=min(timeout, 10.0))
+
+
+def sdk_options(config: Any) -> dict[str, Any]:
+    # The query loop owns retry budgets; nested SDK retries hide progress.
+    options: dict[str, Any] = {"max_retries": 0, "timeout": request_timeout(config)}
+    routing = http_options(config)
+    if routing:
+        options["http_client"] = httpx.AsyncClient(**routing, timeout=request_timeout(config))
+    return options
+
+
+def certificate_failure(exc: BaseException) -> bool:
+    seen: set[int] = set()
+    pending = [exc]
+    while pending:
+        current = pending.pop()
+        if id(current) in seen:
+            continue
+        seen.add(id(current))
+        if isinstance(current, ssl.SSLCertVerificationError) or "CERTIFICATE_VERIFY_FAILED" in str(current):
+            return True
+        pending.extend(item for item in (current.__cause__, current.__context__) if item is not None)
+    return False
+
+
+def network_error_message(exc: BaseException, config: Any) -> str:
+    if certificate_failure(exc):
+        return "模型连接证书校验失败，请检查证书或网络拦截；未自动重试"
+    mode = getattr(config, "network_mode", "inherit")
+    route = {"inherit": "继承进程代理环境", "direct": "直连", "proxy": "指定代理"}.get(mode, mode)
+    return f"模型连接中断（{type(exc).__name__}，{route}）。请检查网络或代理状态后重试；更改系统环境变量需重启 Gateway。"

crabcode_core/api/ollama_adapter.py

@@ -0,0 +1,27 @@
+"""Ollama API adapter — delegates to OpenAIAdapter with Ollama defaults."""
+
+from __future__ import annotations
+
+from crabcode_core.api.openai_adapter import OpenAIAdapter
+from crabcode_core.types.config import ApiConfig
+
+
+class OllamaAdapter(OpenAIAdapter):
+    """Adapter for Ollama via its OpenAI-compatible Chat Completions API.
+
+    Ollama natively exposes an OpenAI-compatible endpoint at
+    ``http://localhost:11434/v1`` and does not require a real API key,
+    but the OpenAI SDK still expects *some* value for ``api_key``.
+    This adapter pre-fills those defaults and disables extended thinking
+    (local models do not support it).
+    """
+
+    def __init__(self, config: ApiConfig):
+        # Pre-fill Ollama defaults before handing off to OpenAIAdapter
+        if not config.base_url:
+            config = config.model_copy(update={"base_url": "http://localhost:11434/v1"})
+        if not config.api_key_env:
+            config = config.model_copy(update={"api_key_env": "OLLAMA_API_KEY"})
+        if config.thinking_enabled is True:
+            config = config.model_copy(update={"thinking_enabled": False})
+        super().__init__(config)