[Feature]: QQ Official adapter passes raw faceType emoji tags into Plain text context #2291

Runtime environment

LangBot v4.10.x / latest, Docker deployment, Community Edition, QQ Official Bot API (qqofficial).

Observed when receiving QQ Official group-at messages containing QQ emoji / face tags in the raw inbound content field.

Deployment version

Community Edition

Exception

QQ Official inbound messages may include QQ emoji / face markup in the raw text content, for example:

@justbot 欢迎我<faceType=1,faceId="300",ext="eyJ0ZXh0Ijoi6IOW5LiJ5pakIn0=">

The ext payload is base64-encoded JSON. Decoding this example gives:

{"text":"胖三斤"}

Expected behavior:

The QQ Official adapter should normalize this platform-specific face tag before the message enters LangBot's generic message chain / model context. For example:

欢迎我[胖三斤]

or, at minimum, remove or replace the raw tag with a safe placeholder such as:

欢迎我[表情]

Actual behavior:

The current legacy QQ Official converter appears to pass the raw inbound text directly into Plain:

yiri_msg_list.append(platform_message.Plain(text=message))

As a result, the model can receive the raw platform tag as ordinary user text:

@HENU_Bot 欢迎我<faceType=1,faceId="300",ext="...">

This can cause the model to interpret or discuss the implementation detail (faceType, faceId, ext, base64 payload) instead of responding naturally to the user message.

Related location:

src/langbot/pkg/platform/sources/qqofficial.py

Relevant path:

• QQOfficialEventConverter.target2yiri() passes event.content to QQOfficialMessageConverter.target2yiri().
• QQOfficialMessageConverter.target2yiri() appends the whole message as Plain(text=message).
• For GROUP_AT_MESSAGE_CREATE / AT_MESSAGE_CREATE, LangBot inserts a synthetic At(target='justbot'), but the original leading textual mention can remain in the raw content.

Suggested fix direction:

Normalize QQ Official inbound text in the adapter-level converter before creating Plain text.

Example approach:

import base64
import json
import re

_QQ_FACE_RE = re.compile(r'<faceType=\d+,faceId="[^"]+",ext="([^"]*)">')
_LEADING_BOT_MENTION_RE = re.compile(r'^\s*@\S+\s*')


def _decode_qq_face_ext(ext: str) -> str:
    try:
        padded = ext + "=" * ((4 - len(ext) % 4) % 4)
        payload = json.loads(base64.b64decode(padded).decode("utf-8"))
        text = str(payload.get("text") or "").strip()
        return f"[{text}]" if text else "[表情]"
    except Exception:
        return "[表情]"


def _normalize_qq_official_text(message: str, strip_leading_mention: bool = False) -> str:
    text = str(message or "")
    text = _QQ_FACE_RE.sub(lambda m: _decode_qq_face_ext(m.group(1)), text)
    if strip_leading_mention:
        text = _LEADING_BOT_MENTION_RE.sub("", text, count=1)
    return text.strip()

Then use the normalized text in QQOfficialMessageConverter.target2yiri() before appending Plain.

For group-at events, it may also be helpful to strip the leading textual bot mention because LangBot already inserts a synthetic At(target='justbot') for those events.

Reproduction steps

1. Configure a QQ Official bot in LangBot.
2. Send a group-at message containing a QQ emoji / face tag, such as:

@Bot 欢迎我<faceType=1,faceId="300",ext="eyJ0ZXh0Ijoi6IOW5LiJ5pakIn0=">

3. Observe the message that enters the model / conversation context.
4. The raw <faceType=...> tag can appear as plain text instead of a normalized emoji label such as [胖三斤] or [表情].

Enabled plugins

This issue is adapter-level and can be observed before plugin business logic. A plugin-assisted chat made it visible because the model responded to the raw faceType / ext text, but the normalization should happen in the QQ Official adapter rather than in individual plugins.