feat(avatar): understand BOXIM image messages

This commit is contained in:
stefanfeng
2026-09-01 15:35:41 +08:00
parent 3d999f9472
commit 0fc43908ae
8 changed files with 523 additions and 13 deletions
@@ -0,0 +1,151 @@
"""Parse and safely download image payloads from BOXIM private messages."""
import ipaddress
import json
import os
import socket
from dataclasses import dataclass
from pathlib import PurePosixPath
from urllib.parse import unquote, urljoin, urlsplit
import httpx
MAX_REDIRECTS = 3
class BoxIMImageError(RuntimeError):
pass
@dataclass(frozen=True)
class DownloadedBoxIMImage:
content: bytes
filename: str
mime_type: str
source_url: str
def parse_boxim_image_url(content: str, *, base_url: str = "") -> str:
try:
payload = json.loads(content or "")
except (TypeError, ValueError) as exc:
raise BoxIMImageError("BOXIM 图片消息格式无效") from exc
if not isinstance(payload, dict):
raise BoxIMImageError("BOXIM 图片消息格式无效")
value = payload.get("originUrl") or payload.get("thumbUrl") or payload.get("url")
if not isinstance(value, str) or not value.strip():
raise BoxIMImageError("BOXIM 图片消息缺少图片地址")
value = value.strip()
if value.startswith("/"):
if not base_url:
raise BoxIMImageError("BOXIM 图片地址不完整")
value = urljoin(f"{base_url.rstrip('/')}/", value)
return value
def _configured_hosts(name: str) -> set[str]:
return {
value.strip().lower().rstrip(".")
for value in os.getenv(name, "").split(",")
if value.strip()
}
def _host_matches(host: str, configured: set[str]) -> bool:
return any(host == value or host.endswith(f".{value}") for value in configured)
def _resolved_addresses(host: str, port: int) -> set[ipaddress.IPv4Address | ipaddress.IPv6Address]:
try:
return {
ipaddress.ip_address(item[4][0])
for item in socket.getaddrinfo(host, port, type=socket.SOCK_STREAM)
}
except (OSError, ValueError) as exc:
raise BoxIMImageError("BOXIM 图片地址无法解析") from exc
def _is_safe_remote_url(url: str) -> None:
parsed = urlsplit(url)
scheme = parsed.scheme.lower()
allow_http = os.getenv("BOXIM_IMAGE_ALLOW_HTTP", "").lower() in {"1", "true", "yes"}
if scheme not in ({"https", "http"} if allow_http else {"https"}):
raise BoxIMImageError("BOXIM 图片地址必须使用 HTTPS")
if parsed.username or parsed.password or not parsed.hostname:
raise BoxIMImageError("BOXIM 图片地址无效")
host = parsed.hostname.lower().rstrip(".")
allowed_hosts = _configured_hosts("BOXIM_IMAGE_ALLOWED_HOSTS")
if allowed_hosts and not _host_matches(host, allowed_hosts):
raise BoxIMImageError("BOXIM 图片地址不在允许的域名范围内")
private_hosts = _configured_hosts("BOXIM_IMAGE_PRIVATE_HOSTS")
try:
addresses = {ipaddress.ip_address(host)}
except ValueError:
addresses = _resolved_addresses(host, parsed.port or (443 if scheme == "https" else 80))
if not addresses:
raise BoxIMImageError("BOXIM 图片地址无法解析")
if _host_matches(host, private_hosts):
return
if any(not address.is_global for address in addresses):
raise BoxIMImageError("BOXIM 图片地址指向受限网络")
def _filename_from_url(url: str) -> str:
value = unquote(PurePosixPath(urlsplit(url).path).name).strip()
value = value.replace("\x00", "")
return (value or "boxim-image")[:255]
def download_boxim_image(
content: str,
*,
base_url: str = "",
transport: httpx.BaseTransport | None = None,
) -> DownloadedBoxIMImage:
"""Download one BOXIM image without redirects or oversized responses escaping checks."""
url = parse_boxim_image_url(content, base_url=base_url)
max_bytes = max(1024, int(os.getenv("CHAT_IMAGE_MAX_BYTES", str(8 * 1024 * 1024))))
timeout = max(1.0, min(float(os.getenv("BOXIM_IMAGE_TIMEOUT_SECONDS", "15")), 60.0))
with httpx.Client(
timeout=timeout,
follow_redirects=False,
trust_env=False,
transport=transport,
) as client:
for _ in range(MAX_REDIRECTS + 1):
_is_safe_remote_url(url)
try:
with client.stream("GET", url, headers={"Accept": "image/*"}) as response:
if response.status_code in {301, 302, 303, 307, 308}:
location = response.headers.get("location", "").strip()
if not location:
raise BoxIMImageError("BOXIM 图片跳转地址无效")
url = urljoin(url, location)
continue
response.raise_for_status()
raw_length = response.headers.get("content-length", "")
if raw_length.isdigit() and int(raw_length) > max_bytes:
raise BoxIMImageError("BOXIM 图片超过大小限制")
chunks = bytearray()
for chunk in response.iter_bytes():
chunks.extend(chunk)
if len(chunks) > max_bytes:
raise BoxIMImageError("BOXIM 图片超过大小限制")
if not chunks:
raise BoxIMImageError("BOXIM 图片内容为空")
return DownloadedBoxIMImage(
content=bytes(chunks),
filename=_filename_from_url(url),
mime_type=response.headers.get("content-type", "").split(";", 1)[0][:100],
source_url=url,
)
except BoxIMImageError:
raise
except (httpx.HTTPError, OSError) as exc:
raise BoxIMImageError("BOXIM 图片下载失败") from exc
raise BoxIMImageError("BOXIM 图片跳转次数过多")
@@ -3,6 +3,7 @@
import asyncio
import hashlib
import logging
import os
import re
import secrets
import time
@@ -13,12 +14,19 @@ from sqlalchemy.orm import Session
from models import (
Avatar,
ChatAttachment,
TakeoverCursor,
TakeoverMessage,
TakeoverReplyTask,
User,
)
from services.boxim_client import BoxIMClient, BoxIMError
from services.boxim_image_service import (
BoxIMImageError,
download_boxim_image,
parse_boxim_image_url,
)
from services.vision_service import ImageValidationError
logger = logging.getLogger(__name__)
@@ -37,6 +45,10 @@ HUMAN_PAUSE_SECONDS = 600
RATE_LIMIT_WINDOW_SECONDS = 300
RATE_LIMIT_MAX_REPLIES = 5
AVATAR_LOCAL_ID_PREFIX = "880"
BOXIM_TEXT_MESSAGE_TYPE = 0
BOXIM_IMAGE_MESSAGE_TYPE = 1
BOXIM_IMAGE_PROMPT = "请看看这张图片。"
BOXIM_IMAGE_UNAVAILABLE_REPLY = "这张图片我暂时没看清,麻烦重新发送一张清晰的原图。"
def _utcnow() -> datetime:
@@ -107,6 +119,14 @@ def _configured_reply_delay(avatar: Avatar, fallback: int | None = None) -> int:
return delay
def _event_prompt(event: TakeoverMessage) -> str:
if event.message_type == BOXIM_TEXT_MESSAGE_TYPE:
return event.content.strip()
if event.message_type == BOXIM_IMAGE_MESSAGE_TYPE:
return BOXIM_IMAGE_PROMPT
return ""
class TakeoverService:
"""Poll BOXIM, honor the owner grace period, then generate and send one reply."""
@@ -498,8 +518,27 @@ class TakeoverService:
if not is_avatar:
self._cancel_conversation(db, avatar.owner_id, peer_id, "owner_replied")
return
if not schedule_reply or event.message_type != 0 or not event.content.strip():
if not schedule_reply or event.message_type not in {
BOXIM_TEXT_MESSAGE_TYPE,
BOXIM_IMAGE_MESSAGE_TYPE,
}:
return
if event.message_type == BOXIM_TEXT_MESSAGE_TYPE and not event.content.strip():
return
if event.message_type == BOXIM_IMAGE_MESSAGE_TYPE:
try:
parse_boxim_image_url(
event.content,
base_url=getattr(self.boxim, "im_base_url", ""),
)
except BoxIMImageError as exc:
logger.warning(
"Ignored invalid BOXIM image message %s for avatar %s: %s",
message_id,
avatar.id,
exc,
)
return
if (now - send_time).total_seconds() > self.max_message_age_seconds:
logger.info(
"Ignored stale BOXIM message %s for avatar %s (age=%ss)",
@@ -606,7 +645,7 @@ class TakeoverService:
task.status = "cancelled"
task.cancel_reason = "newer_incoming_message"
task.locked_at = None
prompt_parts.append(event.content.strip())
prompt_parts.append(_event_prompt(event))
source_ids.append(event.boxim_message_id)
prompt = "\n".join(part for part in prompt_parts if part).strip()[-MAX_PROMPT_LENGTH:]
due_at = max(
@@ -665,6 +704,50 @@ class TakeoverService:
results = await asyncio.gather(*(generate(task_id) for task_id in task_ids))
return sum(bool(result) for result in results)
def _takeover_image_attachment(
self,
db: Session,
avatar: Avatar,
event: TakeoverMessage,
) -> ChatAttachment:
now = self.now()
if event.attachment_id:
cached = db.get(ChatAttachment, event.attachment_id)
if cached and cached.status == "ready" and cached.expires_at > now:
cached.used_at = now
db.commit()
return cached
downloaded = download_boxim_image(
event.content,
base_url=getattr(
self.boxim,
"im_base_url",
os.getenv("BOXIM_API_BASE_URL", "https://im.99hui.com/api"),
),
)
from routers.chat import _analyze_image_bytes
attachment = _analyze_image_bytes(
db,
avatar,
downloaded.content,
filename=downloaded.filename,
mime_type=downloaded.mime_type,
uploader_kind="boxim",
)
event.attachment_id = attachment.id
attachment.used_at = now
db.commit()
logger.info(
"BOXIM image analyzed message=%s attachment=%s avatar=%s category=%s",
event.boxim_message_id,
attachment.id,
avatar.id,
attachment.category,
)
return attachment
def _generate_reply(self, task_id: str) -> bool:
db = self.session_factory()
try:
@@ -694,9 +777,36 @@ class TakeoverService:
.limit(30)
.all()
)
source_events = {
event.boxim_message_id: event
for event in events
if event.boxim_message_id in excluded_ids
}
image_attachments = []
image_failed = False
for message_id in (task.source_message_ids or [])[-3:]:
event = source_events.get(message_id)
if not event or event.message_type != BOXIM_IMAGE_MESSAGE_TYPE:
continue
try:
image_attachments.append(
self._takeover_image_attachment(db, avatar, event)
)
except (BoxIMImageError, ImageValidationError) as exc:
image_failed = True
logger.warning(
"BOXIM image unavailable message=%s avatar=%s: %s",
event.boxim_message_id,
avatar.id,
exc,
)
history = []
for event in reversed(events):
if event.boxim_message_id in excluded_ids or not event.content.strip():
if (
event.boxim_message_id in excluded_ids
or event.message_type != BOXIM_TEXT_MESSAGE_TYPE
or not event.content.strip()
):
continue
if event.direction == "incoming" and event.is_avatar:
continue
@@ -708,10 +818,25 @@ class TakeoverService:
)
history = history[-10:]
from routers.chat import _resolve_reply
from routers.chat import _attachment_contexts, _resolve_reply
result = _resolve_reply(db, avatar, task.prompt, history, usage_source="takeover")
answer = _plain_text_reply(result.get("answer", ""))
image_contexts = _attachment_contexts(image_attachments)
has_source_text = any(
event.message_type == BOXIM_TEXT_MESSAGE_TYPE and event.content.strip()
for event in source_events.values()
)
if image_failed and not image_contexts and not has_source_text:
answer = BOXIM_IMAGE_UNAVAILABLE_REPLY
else:
result = _resolve_reply(
db,
avatar,
task.prompt,
history,
usage_source="takeover",
image_contexts=image_contexts,
)
answer = _plain_text_reply(result.get("answer", ""))
db.refresh(task)
if task.status != "generating":
return False