dsh-plugin-image-workbench 0.0.0-stage → 1.1.5

package/vendor/chatgpt2api/source/services/content_filter.py

@@ -0,0 +1,242 @@
+from __future__ import annotations
+
+import re
+
+from curl_cffi import requests
+from fastapi import HTTPException
+
+from services.config import config
+from services.proxy_service import proxy_settings
+from utils.log import logger
+
+DEFAULT_REVIEW_PROMPT = "判断用户请求是否允许。只回答 ALLOW 或 REJECT。"
+
+# Strip base64 image data URIs before review: a text-only review model can't
+# analyze image bytes, and a single inlined image easily blows past the token
+# budget of the upstream review service.
+_BASE64_DATA_URI = re.compile(r"data:[\w/.+;-]+;base64,[A-Za-z0-9+/=]+")
+
+# Cap aligned to the upstream review service's max context. If text still
+# exceeds the cap after base64 stripping, keep equal head/tail halves so both
+# the system prompt and the most recent user message survive.
+_MAX_REVIEW_TEXT_LEN = 100_000
+_TRUNCATION_MARKER = "\n…[truncated]…\n"
+
+
+def _text(value: object) -> str:
+    if isinstance(value, str):
+        return value
+    if isinstance(value, list):
+        return "\n".join(_text(item) for item in value)
+    if isinstance(value, dict):
+        return "\n".join(_text(value.get(key)) for key in ("text", "input_text", "content", "input", "instructions", "system", "prompt"))
+    return ""
+
+
+def request_text(*values: object) -> str:
+    return "\n".join(part for value in values if (part := _text(value).strip()))
+
+
+def request_shape(*values: object) -> dict[str, int]:
+    """Return a safe structural summary without logging prompts or image bytes."""
+    stats = {
+        "response_message_items": 0,
+        "input_image_parts": 0,
+        "image_url_parts": 0,
+        "image_parts": 0,
+        "data_url_images": 0,
+        "remote_image_urls": 0,
+        "literal_image_placeholders": 0,
+    }
+
+    def walk(value: object, key: str = "") -> None:
+        if isinstance(value, str):
+            text = value.strip()
+            lower = text.lower()
+            if "<image>" in lower:
+                stats["literal_image_placeholders"] += 1
+            if lower.startswith("data:image/"):
+                stats["data_url_images"] += 1
+            elif key in {"image_url", "url"} and lower.startswith(("http://", "https://")):
+                stats["remote_image_urls"] += 1
+            return
+        if isinstance(value, list):
+            for item in value:
+                walk(item, key)
+            return
+        if not isinstance(value, dict):
+            return
+        item_type = str(value.get("type") or "").strip()
+        if item_type == "message":
+            stats["response_message_items"] += 1
+        elif item_type == "input_image":
+            stats["input_image_parts"] += 1
+        elif item_type == "image_url":
+            stats["image_url_parts"] += 1
+        elif item_type == "image":
+            stats["image_parts"] += 1
+        for child_key, child in value.items():
+            walk(child, str(child_key))
+
+    for value in values:
+        walk(value)
+    return {key: value for key, value in stats.items() if value}
+
+
+def _sanitize_for_review(text: str) -> tuple[str, dict[str, int]]:
+    """Strip base64 data URIs and truncate to the review-service context limit.
+
+    Returns (sanitized_text, stats) where stats carries base64_blocks_stripped
+    and truncated_chars so callers can emit structured logs.
+    """
+    sanitized, base64_blocks_stripped = _BASE64_DATA_URI.subn("[image]", text)
+    truncated_chars = 0
+    if len(sanitized) > _MAX_REVIEW_TEXT_LEN:
+        # Reserve marker space so the result stays within the cap.
+        half = (_MAX_REVIEW_TEXT_LEN - len(_TRUNCATION_MARKER)) // 2
+        truncated_chars = len(sanitized) - 2 * half
+        sanitized = sanitized[:half] + _TRUNCATION_MARKER + sanitized[-half:]
+    stats = {
+        "base64_blocks_stripped": base64_blocks_stripped,
+        "truncated_chars": truncated_chars,
+    }
+    return sanitized, stats
+
+
+def _extract_review_decision(data: object) -> str | None:
+    """Defensively pull the decision text out of the review service response.
+
+    Returns None when the response shape doesn't match the OpenAI chat-completion
+    contract (e.g. {"error": ...} with no choices). The caller treats None as
+    "undecided" and applies the configured fail-open policy.
+    """
+    if not isinstance(data, dict):
+        return None
+    choices = data.get("choices")
+    if not isinstance(choices, list) or not choices:
+        return None
+    first = choices[0]
+    if not isinstance(first, dict):
+        return None
+    message = first.get("message")
+    if not isinstance(message, dict):
+        return None
+    content = message.get("content")
+    if content is None:
+        return None
+    return str(content).strip().lower()
+
+
+def _is_allow_decision(decision: str) -> bool:
+    return decision.startswith(("allow", "pass", "true", "yes", "通过", "允许", "安全"))
+
+
+def _is_reject_decision(decision: str) -> bool:
+    return decision.startswith(("reject", "deny", "block", "false", "no", "拒绝", "不允许", "违规", "禁止"))
+
+
+def _resolve_fail_open(review: dict) -> bool:
+    """Resolve fail_open from review config. Defaults to True."""
+    value = review.get("fail_open")
+    if value is None:
+        return True
+    if isinstance(value, bool):
+        return value
+    if isinstance(value, str):
+        return value.strip().lower() in {"1", "true", "yes", "on"}
+    return bool(value)
+
+
+def check_request(text: str) -> None:
+    text = str(text or "")
+    if not text.strip():
+        return
+    # Local sensitive-word match runs on the raw text (cheap, no network).
+    for word in config.sensitive_words:
+        if word in text:
+            raise HTTPException(status_code=400, detail={"error": "检测到敏感词，拒绝本次任务"})
+    review = config.ai_review
+    if not review.get("enabled"):
+        return
+    base_url = str(review.get("base_url") or "").strip().rstrip("/")
+    api_key = str(review.get("api_key") or "").strip()
+    model = str(review.get("model") or "").strip()
+    if not base_url or not api_key or not model:
+        raise HTTPException(status_code=400, detail={"error": "ai review config is incomplete"})
+
+    fail_open = _resolve_fail_open(review)
+
+    review_text, sanitize_stats = _sanitize_for_review(text)
+    if sanitize_stats["base64_blocks_stripped"] or sanitize_stats["truncated_chars"]:
+        logger.info({
+            "event": "ai_review_text_sanitized",
+            "original_text_len": len(text),
+            "review_text_len": len(review_text),
+            **sanitize_stats,
+        })
+    prompt = str(review.get("prompt") or DEFAULT_REVIEW_PROMPT).strip()
+    content = f"{prompt}\n\n用户请求:\n{review_text}\n\n只回答 ALLOW 或 REJECT。"
+
+    # fail_open=True (default): on upstream failure or ambiguous reply, let the
+    # request through. The review is a soft safety net; one missed review is
+    # preferable to a 5xx storm when the review service is flaky. Set
+    # config.ai_review.fail_open=false for strict-compliance deployments.
+    def _on_failure(event_payload: dict) -> None:
+        logger.warning(event_payload)
+        if not fail_open:
+            raise HTTPException(
+                status_code=503,
+                detail={"error": "AI 审核服务暂时不可用，请稍后重试"},
+            )
+
+    try:
+        response = requests.post(
+            f"{base_url}/v1/chat/completions",
+            headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
+            json={"model": model, "messages": [{"role": "user", "content": content}], "temperature": 0},
+            timeout=60,
+            **proxy_settings.build_session_kwargs(),
+        )
+    except Exception as exc:
+        _on_failure({
+            "event": "ai_review_request_failed",
+            "error": str(exc),
+            "error_type": exc.__class__.__name__,
+            "review_text_len": len(review_text),
+            "original_text_len": len(text),
+        })
+        return
+
+    try:
+        data = response.json()
+    except Exception as exc:
+        _on_failure({
+            "event": "ai_review_response_not_json",
+            "status_code": response.status_code,
+            "body_preview": str(response.text or "")[:200],
+            "error": str(exc),
+        })
+        return
+
+    decision = _extract_review_decision(data)
+    if decision is None:
+        _on_failure({
+            "event": "ai_review_malformed_response",
+            "status_code": response.status_code,
+            "body_preview": str(data)[:300],
+            "review_text_len": len(review_text),
+            "original_text_len": len(text),
+        })
+        return
+
+    if _is_allow_decision(decision):
+        return
+    if _is_reject_decision(decision):
+        raise HTTPException(status_code=400, detail={"error": "AI 审核未通过，拒绝本次任务"})
+    # Ambiguous decisions (e.g. "MAYBE", empty content) fall back to fail-open policy.
+    _on_failure({
+        "event": "ai_review_ambiguous_decision",
+        "decision": decision[:100],
+        "review_text_len": len(review_text),
+    })
+    return

package/vendor/chatgpt2api/source/services/cpa_service.py

@@ -0,0 +1,315 @@
+"""CLIProxyAPI integration for browsing remote auth files and importing selected tokens."""
+
+from __future__ import annotations
+
+import json
+import threading
+import uuid
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from datetime import datetime, timezone
+from pathlib import Path
+from threading import Lock
+
+from curl_cffi.requests import Session
+
+from services.account_service import account_service
+from services.config import DATA_DIR
+from services.proxy_service import proxy_settings
+
+
+CPA_CONFIG_FILE = DATA_DIR / "cpa_config.json"
+
+
+def _new_id() -> str:
+    return uuid.uuid4().hex[:12]
+
+
+def _now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def _normalize_import_job(raw: object, *, fail_unfinished: bool) -> dict | None:
+    if not isinstance(raw, dict):
+        return None
+    status = str(raw.get("status") or "failed").strip() or "failed"
+    if fail_unfinished and status in {"pending", "running"}:
+        status = "failed"
+    return {
+        "job_id": str(raw.get("job_id") or uuid.uuid4().hex).strip(),
+        "status": status,
+        "created_at": str(raw.get("created_at") or _now_iso()).strip() or _now_iso(),
+        "updated_at": str(raw.get("updated_at") or raw.get("created_at") or _now_iso()).strip() or _now_iso(),
+        "total": int(raw.get("total") or 0),
+        "completed": int(raw.get("completed") or 0),
+        "added": int(raw.get("added") or 0),
+        "skipped": int(raw.get("skipped") or 0),
+        "refreshed": int(raw.get("refreshed") or 0),
+        "failed": int(raw.get("failed") or 0),
+        "errors": raw.get("errors") if isinstance(raw.get("errors"), list) else [],
+    }
+
+
+def _normalize_pool(raw: dict) -> dict:
+    return {
+        "id": str(raw.get("id") or _new_id()).strip(),
+        "name": str(raw.get("name") or "").strip(),
+        "base_url": str(raw.get("base_url") or "").strip(),
+        "secret_key": str(raw.get("secret_key") or "").strip(),
+        "import_job": _normalize_import_job(raw.get("import_job"), fail_unfinished=True),
+    }
+
+
+def _management_headers(secret_key: str) -> dict[str, str]:
+    return {
+        "Authorization": f"Bearer {secret_key}",
+        "Accept": "application/json",
+    }
+
+
+class CPAConfig:
+    def __init__(self, store_file: Path):
+        self._store_file = store_file
+        self._lock = Lock()
+        self._pools: list[dict] = self._load()
+
+    def _load(self) -> list[dict]:
+        if not self._store_file.exists():
+            return []
+        try:
+            raw = json.loads(self._store_file.read_text(encoding="utf-8"))
+            if isinstance(raw, dict) and "base_url" in raw:
+                pool = _normalize_pool(raw)
+                return [pool] if pool["base_url"] else []
+            if isinstance(raw, list):
+                return [_normalize_pool(item) for item in raw if isinstance(item, dict)]
+        except Exception:
+            pass
+        return []
+
+    def _save(self) -> None:
+        self._store_file.parent.mkdir(parents=True, exist_ok=True)
+        self._store_file.write_text(json.dumps(self._pools, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+
+    def list_pools(self) -> list[dict]:
+        with self._lock:
+            return [dict(pool) for pool in self._pools]
+
+    def get_pool(self, pool_id: str) -> dict | None:
+        with self._lock:
+            for pool in self._pools:
+                if pool["id"] == pool_id:
+                    return dict(pool)
+        return None
+
+    def add_pool(self, name: str, base_url: str, secret_key: str) -> dict:
+        pool = _normalize_pool({"id": _new_id(), "name": name, "base_url": base_url, "secret_key": secret_key})
+        with self._lock:
+            self._pools.append(pool)
+            self._save()
+        return dict(pool)
+
+    def update_pool(self, pool_id: str, updates: dict) -> dict | None:
+        with self._lock:
+            for index, pool in enumerate(self._pools):
+                if pool["id"] != pool_id:
+                    continue
+                merged = {**pool, **{key: value for key, value in updates.items() if value is not None}, "id": pool_id}
+                self._pools[index] = _normalize_pool(merged)
+                self._save()
+                return dict(self._pools[index])
+        return None
+
+    def delete_pool(self, pool_id: str) -> bool:
+        with self._lock:
+            before = len(self._pools)
+            self._pools = [pool for pool in self._pools if pool["id"] != pool_id]
+            if len(self._pools) < before:
+                self._save()
+                return True
+        return False
+
+    def set_import_job(self, pool_id: str, import_job: dict | None) -> dict | None:
+        with self._lock:
+            for index, pool in enumerate(self._pools):
+                if pool["id"] != pool_id:
+                    continue
+                next_pool = dict(pool)
+                next_pool["import_job"] = _normalize_import_job(import_job, fail_unfinished=False)
+                self._pools[index] = next_pool
+                self._save()
+                return dict(next_pool)
+        return None
+
+    def get_import_job(self, pool_id: str) -> dict | None:
+        with self._lock:
+            for pool in self._pools:
+                if pool["id"] == pool_id:
+                    job = pool.get("import_job")
+                    return dict(job) if isinstance(job, dict) else None
+        return None
+
+
+def list_remote_files(pool: dict) -> list[dict]:
+    base_url = str(pool.get("base_url") or "").strip()
+    secret_key = str(pool.get("secret_key") or "").strip()
+    if not base_url or not secret_key:
+        return []
+
+    url = f"{base_url.rstrip('/')}/v0/management/auth-files"
+    session = Session(**proxy_settings.build_session_kwargs(verify=True))
+    try:
+        response = session.get(url, headers=_management_headers(secret_key), timeout=30)
+        if not response.ok:
+            raise RuntimeError(f"remote list failed: HTTP {response.status_code}")
+        payload = response.json()
+    finally:
+        session.close()
+
+    files = payload.get("files") if isinstance(payload, dict) else None
+    if not isinstance(files, list):
+        raise RuntimeError("remote list payload is invalid")
+
+    items: list[dict] = []
+    for item in files:
+        if not isinstance(item, dict):
+            continue
+        name = str(item.get("name") or "").strip()
+        email = str(item.get("email") or item.get("account") or "").strip()
+        if not name:
+            continue
+        items.append({"name": name, "email": email})
+    return items
+
+
+def fetch_remote_access_token(pool: dict, file_name: str) -> tuple[str | None, str | None]:
+    base_url = str(pool.get("base_url") or "").strip()
+    secret_key = str(pool.get("secret_key") or "").strip()
+    file_name = str(file_name or "").strip()
+    if not base_url or not secret_key or not file_name:
+        return None, "invalid request"
+
+    url = f"{base_url.rstrip('/')}/v0/management/auth-files/download"
+    session = Session(**proxy_settings.build_session_kwargs(verify=True))
+    try:
+        response = session.get(url, headers=_management_headers(secret_key), params={"name": file_name}, timeout=30)
+        if not response.ok:
+            return None, f"HTTP {response.status_code}"
+        payload = response.json()
+    except Exception as exc:
+        return None, str(exc)
+    finally:
+        session.close()
+
+    if not isinstance(payload, dict):
+        return None, "invalid payload"
+
+    access_token = str(payload.get("access_token") or "").strip()
+    if not access_token:
+        return None, "missing access_token"
+    return access_token, None
+
+
+class CPAImportService:
+    def __init__(self, cpa_config: CPAConfig):
+        self._config = cpa_config
+
+    def start_import(self, pool: dict, selected_files: list[str]) -> dict:
+        names = [str(name or "").strip() for name in selected_files if str(name or "").strip()]
+        if not names:
+            raise ValueError("selected files is required")
+
+        pool_id = str(pool.get("id") or "").strip()
+        job = {
+            "job_id": uuid.uuid4().hex,
+            "status": "pending",
+            "created_at": _now_iso(),
+            "updated_at": _now_iso(),
+            "total": len(names),
+            "completed": 0,
+            "added": 0,
+            "skipped": 0,
+            "refreshed": 0,
+            "failed": 0,
+            "errors": [],
+        }
+        saved_pool = self._config.set_import_job(pool_id, job)
+        if saved_pool is None:
+            raise ValueError("pool not found")
+
+        thread = threading.Thread(
+            target=self._run_import,
+            args=(pool_id, pool, names),
+            name=f"cpa-import-{pool_id}",
+            daemon=True,
+        )
+        thread.start()
+        return dict(saved_pool.get("import_job") or job)
+
+    def _update_job(self, pool_id: str, **updates) -> dict | None:
+        current = self._config.get_import_job(pool_id)
+        if current is None:
+            return None
+        next_job = {**current, **updates, "updated_at": _now_iso()}
+        pool = self._config.set_import_job(pool_id, next_job)
+        if pool is None:
+            return None
+        job = pool.get("import_job")
+        return dict(job) if isinstance(job, dict) else None
+
+    def _append_error(self, pool_id: str, file_name: str, message: str) -> None:
+        current = self._config.get_import_job(pool_id)
+        if current is None:
+            return
+        errors = list(current.get("errors") or [])
+        errors.append({"name": file_name, "error": message})
+        self._update_job(pool_id, errors=errors, failed=len(errors))
+
+    def _run_import(self, pool_id: str, pool: dict, names: list[str]) -> None:
+        self._update_job(pool_id, status="running")
+
+        tokens: list[str] = []
+        max_workers = min(16, max(1, len(names)))
+        with ThreadPoolExecutor(max_workers=max_workers) as executor:
+            future_map = {executor.submit(fetch_remote_access_token, pool, name): name for name in names}
+            for future in as_completed(future_map):
+                file_name = future_map[future]
+                try:
+                    token, error = future.result()
+                except Exception as exc:
+                    token, error = None, str(exc)
+
+                if token:
+                    tokens.append(token)
+                else:
+                    self._append_error(pool_id, file_name, error or "unknown error")
+
+                current = self._config.get_import_job(pool_id) or {}
+                failed = len(current.get("errors") or [])
+                self._update_job(pool_id, completed=int(current.get("completed") or 0) + 1, failed=failed)
+
+        if not tokens:
+            current = self._config.get_import_job(pool_id) or {}
+            self._update_job(
+                pool_id,
+                status="failed",
+                completed=int(current.get("total") or 0),
+                failed=len(current.get("errors") or []),
+            )
+            return
+
+        add_result = account_service.add_accounts(tokens, source_type="codex")
+        refresh_result = account_service.refresh_accounts(tokens)
+        current = self._config.get_import_job(pool_id) or {}
+        self._update_job(
+            pool_id,
+            status="completed",
+            completed=len(names),
+            added=int(add_result.get("added") or 0),
+            skipped=int(add_result.get("skipped") or 0),
+            refreshed=int(refresh_result.get("refreshed") or 0),
+            failed=len(current.get("errors") or []),
+        )
+
+
+cpa_config = CPAConfig(CPA_CONFIG_FILE)
+cpa_import_service = CPAImportService(cpa_config)