feat(whatsapp): gate group replies on mention and share group sessions

Stop replying to every group message by filtering on @bot/triggers in the sidecar and gateway, and map group chats to one shared session while preserving per-sender identity.

Co-authored-by: Cursor <cursoragent@cursor.com>

runtime/application/gateway/inbound_service.py

@@ -315,6 +315,18 @@ def _build_channel_session_title(*, channel: str, account_name: str, external_us
     return f"{ch}|{body}"
 
 
+def _session_title_user_label(*, is_group: bool, external_user_id: str, group_name: str, external_chat_id: str) -> str:
+    if not is_group:
+        return str(external_user_id or "").strip() or "unknown"
+    name = str(group_name or "").strip()
+    if name:
+        return name
+    chat = str(external_chat_id or "").strip()
+    if chat and "@" in chat:
+        return chat.split("@", 1)[0] or "group"
+    return "group"
+
+
 def _parse_generic_inbound(channel_name: str, payload: dict[str, Any]) -> InboundMessage:
     meta = payload.get("metadata") if isinstance(payload.get("metadata"), dict) else {}
     user_id = str(payload.get("user_id") or payload.get("external_user_id") or "").strip()
@@ -655,6 +667,19 @@ def process_inbound_payload(payload: dict[str, Any]) -> dict[str, Any]:
     if not account_id:
         raise ValueError(f"missing {channel_name} account_id")
 
+    account = store.find_user_by_channel_account(channel=inbound.channel, account_id=account_id) or {}
+    from runtime.orchestration.group_ingest import (
+        build_group_sender_context,
+        resolve_group_policy,
+        session_user_key,
+        should_process_group_inbound,
+    )
+
+    group_policy = resolve_group_policy(account=account)
+    bot_jid = None
+    if isinstance(inbound.metadata, dict):
+        bot_jid = str(inbound.metadata.get("bot_jid") or "").strip() or None
+
     text = inbound.text.strip()
     preface = ""
     channel_session_id = ""
@@ -673,6 +698,15 @@ def process_inbound_payload(payload: dict[str, Any]) -> dict[str, Any]:
         )
         reply = ("绑定成功。\n\n" + _menu_text()) if info else "绑定失败：无效或已使用的绑定码。"
     else:
+        if inbound.is_group and not should_process_group_inbound(
+            is_group=inbound.is_group,
+            text=text,
+            mentions=list(inbound.mentions or []),
+            bot_jid=bot_jid,
+            require_mention=group_policy.require_mention,
+            triggers=list(group_policy.triggers),
+        ):
+            return {"ok": True, "replies": []}
         reply = ""
         reply_attachments: list[dict[str, Any]] = []
         ident = store.resolve_user_by_channel_identity_v2(
@@ -714,19 +748,28 @@ def process_inbound_payload(payload: dict[str, Any]) -> dict[str, Any]:
             tenant_id = str(ident.get("tenant_id") or "")
             user_id = str(ident.get("user_id") or "")
             role = str(ident.get("role") or "member")
-            account = store.find_user_by_channel_account(channel=inbound.channel, account_id=account_id) or {}
             account_name = str(account.get("name") or "").strip() or account_id
             group_name = _extract_group_name(inbound)
+            session_external_user_id = session_user_key(
+                is_group=inbound.is_group,
+                external_user_id=inbound.external_user_id,
+            )
+            title_user_label = _session_title_user_label(
+                is_group=inbound.is_group,
+                external_user_id=inbound.external_user_id,
+                group_name=group_name,
+                external_chat_id=inbound.external_chat_id,
+            )
             session_id = store.get_or_create_channel_session_v2(
                 tenant_id=tenant_id,
                 channel=inbound.channel,
                 account_id=account_id,
-                external_user_id=inbound.external_user_id,
+                external_user_id=session_external_user_id,
                 external_chat_id=inbound.external_chat_id,
                 session_title=_build_channel_session_title(
                     channel=inbound.channel,
                     account_name=account_name,
-                    external_user_id=inbound.external_user_id,
+                    external_user_id=title_user_label,
                     is_group=inbound.is_group,
                     group_name=group_name,
                 ),
@@ -770,6 +813,12 @@ def process_inbound_payload(payload: dict[str, Any]) -> dict[str, Any]:
                         reply = cmd_reply
                     elif not reply:
                         user_text = (inbound.text or "").strip()
+                        if inbound.is_group:
+                            sender_ctx = build_group_sender_context(
+                                metadata=inbound.metadata if isinstance(inbound.metadata, dict) else {},
+                                external_user_id=inbound.external_user_id,
+                            )
+                            user_text = f"{sender_ctx}\n{user_text}" if user_text else sender_ctx
                         gw_attachments = _channel_attachments_for_gateway(
                             list(inbound.attachments or [])
                         )
@@ -806,6 +855,10 @@ def process_inbound_payload(payload: dict[str, Any]) -> dict[str, Any]:
                                         "account_id": account_id,
                                         "interaction_mode": interaction_mode,
                                         "selected_specialist": selected_specialist,
+                                        "is_group": inbound.is_group,
+                                        "external_user_id": inbound.external_user_id,
+                                        "external_chat_id": inbound.external_chat_id,
+                                        "group_sender_id": inbound.external_user_id,
                                     },
                                 )
                                 manager = _build_admin_gateway_executor(

runtime/operations/whatsapp_bridge/baileys_runner.ts

@@ -30,6 +30,11 @@ const PROXY_URL = (
   process.env.http_proxy ||
   ""
 ).trim();
+const GROUP_REQUIRE_MENTION = String(process.env.AIA_WHATSAPP_GROUP_REQUIRE_MENTION ?? "1").trim() !== "0";
+const GROUP_TRIGGERS = String(process.env.AIA_WHATSAPP_GROUP_TRIGGERS || "/oclaw,|oclaw")
+  .split(",")
+  .map((s) => s.trim())
+  .filter(Boolean);
 
 function log(msg: string): void {
   process.stdout.write(`${new Date().toISOString()} [baileys-whatsapp] ${msg}\n`);
@@ -57,22 +62,68 @@ function isStatusOrBroadcastJid(jid: string): boolean {
   return low === "status@broadcast" || low.endsWith("@broadcast");
 }
 
+function extractMentions(m: proto.IMessage | null | undefined): string[] {
+  if (!m) return [];
+  const ctx =
+    m.extendedTextMessage?.contextInfo ||
+    m.imageMessage?.contextInfo ||
+    m.videoMessage?.contextInfo ||
+    m.documentMessage?.contextInfo ||
+    null;
+  const raw = ctx?.mentionedJid;
+  if (!Array.isArray(raw)) return [];
+  return raw.map((j) => jidNormalizedUser(String(j || "").trim())).filter(Boolean);
+}
+
+function shouldProcessGroupMessage(params: {
+  isGroup: boolean;
+  text: string;
+  mentions: string[];
+  botJid: string;
+}): boolean {
+  if (!params.isGroup) return true;
+  const bot = jidNormalizedUser(params.botJid || "");
+  if (bot) {
+    const botUser = bot.split("@")[0] || "";
+    for (const m of params.mentions || []) {
+      const norm = jidNormalizedUser(String(m || "").trim());
+      if (!norm) continue;
+      if (norm === bot) return true;
+      if (botUser && norm.split("@")[0] === botUser) return true;
+    }
+  }
+  const body = String(params.text || "");
+  for (const t of GROUP_TRIGGERS) {
+    if (t && body.includes(t)) return true;
+  }
+  return !GROUP_REQUIRE_MENTION;
+}
+
 function buildInboundPayload(params: {
   chatId: string;
   userId: string;
   text: string;
   raw: unknown;
+  isGroup: boolean;
+  mentions: string[];
+  groupName?: string;
+  botJid?: string;
 }): Json {
+  const metadata: Json = {
+    source: "whatsapp_baileys",
+    raw: params.raw,
+  };
+  if (params.groupName) metadata.group_name = params.groupName;
+  if (params.botJid) metadata.bot_jid = params.botJid;
   return {
     channel: "whatsapp",
     account_id: ACCOUNT_ID,
     user_id: params.userId,
     chat_id: params.chatId,
     text: params.text,
-    metadata: {
-      source: "whatsapp_baileys",
-      raw: params.raw,
-    },
+    is_group: params.isGroup,
+    mentions: params.mentions,
+    metadata,
   };
 }
 
@@ -234,6 +285,26 @@ async function main(): Promise<void> {
   let reconnectAttempt = 0;
   let sock: ReturnType<typeof makeWASocket> | null = null;
   const wsAgent = PROXY_URL ? new HttpsProxyAgent(PROXY_URL) : undefined;
+  const groupNameCache = new Map<string, { name: string; ts: number }>();
+  const GROUP_NAME_TTL_MS = 10 * 60 * 1000;
+
+  const resolveGroupName = async (chatId: string): Promise<string> => {
+    const key = String(chatId || "").trim();
+    if (!key) return "";
+    const cached = groupNameCache.get(key);
+    const now = Date.now();
+    if (cached && now - cached.ts < GROUP_NAME_TTL_MS) return cached.name;
+    const s = sock;
+    if (!s) return cached?.name || "";
+    try {
+      const meta = await s.groupMetadata(key);
+      const name = String(meta?.subject || "").trim();
+      groupNameCache.set(key, { name, ts: now });
+      return name;
+    } catch {
+      return cached?.name || "";
+    }
+  };
 
   const connectOnce = async () => {
     sock = makeWASocket({
@@ -300,6 +371,22 @@ async function main(): Promise<void> {
           const from = isGroup ? String(key.participant || "").trim() : remoteJid;
           const userId = from ? jidNormalizedUser(from) : jidNormalizedUser(remoteJid);
           const chatId = jidNormalizedUser(remoteJid);
+          const mentions = extractMentions(msg.message);
+          const botJid = sock?.user?.id ? jidNormalizedUser(sock.user.id) : "";
+
+          if (
+            !shouldProcessGroupMessage({
+              isGroup,
+              text,
+              mentions,
+              botJid,
+            })
+          ) {
+            if (VERBOSE) log(`skip group message chat=${chatId} user=${userId} (no mention/trigger)`);
+            continue;
+          }
+
+          const groupName = isGroup ? await resolveGroupName(chatId) : "";
 
           const raw = {
             id,
@@ -309,7 +396,16 @@ async function main(): Promise<void> {
             messageTimestamp: (msg as any).messageTimestamp || null,
           };
 
-          const inbound = buildInboundPayload({ chatId, userId, text, raw });
+          const inbound = buildInboundPayload({
+            chatId,
+            userId,
+            text,
+            raw,
+            isGroup,
+            mentions,
+            groupName: groupName || undefined,
+            botJid: botJid || undefined,
+          });
           if (VERBOSE) log(`inbound posting chat=${chatId} user=${userId} textLen=${text.length}`);
           const out = await postInbound(inbound);
           const replies = Array.isArray(out.replies) ? (out.replies as Json[]) : [];

runtime/orchestration/group_ingest.py

@@ -0,0 +1,135 @@
+from __future__ import annotations
+
+import os
+from dataclasses import dataclass
+from typing import Any
+
+GROUP_SESSION_USER_SENTINEL = "__group__"
+
+
+def normalize_jid(jid: str) -> str:
+    s = str(jid or "").strip().lower()
+    if not s:
+        return ""
+    if "@" in s:
+        user, domain = s.split("@", 1)
+        user = user.split(":")[0]
+        return f"{user}@{domain}"
+    return s.split(":")[0]
+
+
+def normalize_jids(jids: list[str]) -> set[str]:
+    out: set[str] = set()
+    for raw in jids or []:
+        n = normalize_jid(raw)
+        if n:
+            out.add(n)
+    return out
+
+
+def session_user_key(*, is_group: bool, external_user_id: str) -> str:
+    return GROUP_SESSION_USER_SENTINEL if is_group else str(external_user_id or "").strip()
+
+
+@dataclass(frozen=True)
+class GroupPolicyConfig:
+    require_mention: bool = True
+    triggers: tuple[str, ...] = ("/oclaw",)
+    session_scope: str = "chat"
+
+
+def _parse_bool_env(name: str, default: bool) -> bool:
+    raw = str(os.environ.get(name) or "").strip().lower()
+    if not raw:
+        return default
+    return raw not in {"0", "false", "no", "off"}
+
+
+def _parse_triggers_env(name: str, default: tuple[str, ...]) -> tuple[str, ...]:
+    raw = str(os.environ.get(name) or "").strip()
+    if not raw:
+        return default
+    parts = [p.strip() for p in raw.split(",") if p.strip()]
+    return tuple(parts) if parts else default
+
+
+def _parse_group_policy_dict(raw: Any) -> GroupPolicyConfig | None:
+    if not isinstance(raw, dict):
+        return None
+    require_mention = raw.get("require_mention")
+    triggers_raw = raw.get("triggers")
+    session_scope = raw.get("session_scope")
+    triggers: tuple[str, ...] | None = None
+    if isinstance(triggers_raw, list):
+        triggers = tuple(str(x).strip() for x in triggers_raw if str(x).strip())
+    return GroupPolicyConfig(
+        require_mention=bool(require_mention) if require_mention is not None else True,
+        triggers=triggers if triggers is not None else ("/oclaw",),
+        session_scope=str(session_scope or "chat").strip() or "chat",
+    )
+
+
+def resolve_group_policy(*, account: dict[str, Any] | None = None) -> GroupPolicyConfig:
+    cfg = (account or {}).get("config")
+    if isinstance(cfg, dict):
+        gp = _parse_group_policy_dict(cfg.get("group_policy"))
+        if gp is not None:
+            return gp
+        gp = _parse_group_policy_dict(cfg.get("group"))
+        if gp is not None:
+            return gp
+    return GroupPolicyConfig(
+        require_mention=_parse_bool_env("AIA_WHATSAPP_GROUP_REQUIRE_MENTION", True),
+        triggers=_parse_triggers_env("AIA_WHATSAPP_GROUP_TRIGGERS", ("/oclaw", "|oclaw")),
+    )
+
+
+def should_process_group_inbound(
+    *,
+    is_group: bool,
+    text: str,
+    mentions: list[str],
+    bot_jid: str | None,
+    require_mention: bool = True,
+    triggers: list[str] | tuple[str, ...] | None = None,
+) -> bool:
+    if not is_group:
+        return True
+    mention_set = normalize_jids(list(mentions or []))
+    bot = normalize_jid(str(bot_jid or ""))
+    if bot and bot in mention_set:
+        return True
+    # Baileys may omit bot from mentionedJid when user uses display-name @; compare user part.
+    if bot and "@" in bot:
+        bot_user = bot.split("@", 1)[0]
+        for m in mention_set:
+            if "@" in m and m.split("@", 1)[0] == bot_user:
+                return True
+    trigger_list = [str(t) for t in (triggers or []) if str(t)]
+    body = str(text or "")
+    if trigger_list and any(t in body for t in trigger_list):
+        return True
+    return not require_mention
+
+
+def build_group_sender_context(*, metadata: dict[str, Any] | None, external_user_id: str) -> str:
+    meta = metadata if isinstance(metadata, dict) else {}
+    raw = meta.get("raw") if isinstance(meta.get("raw"), dict) else {}
+    push_name = str(raw.get("pushName") or meta.get("push_name") or meta.get("display_name") or "").strip()
+    sender = str(external_user_id or "").strip()
+    label = push_name or sender or "unknown"
+    if sender and push_name and sender not in push_name:
+        return f"[群成员: {label} ({sender})]"
+    return f"[群成员: {label}]"
+
+
+__all__ = [
+    "GROUP_SESSION_USER_SENTINEL",
+    "GroupPolicyConfig",
+    "build_group_sender_context",
+    "normalize_jid",
+    "normalize_jids",
+    "resolve_group_policy",
+    "session_user_key",
+    "should_process_group_inbound",
+]

runtime/orchestration/session_mapping.py

@@ -20,9 +20,9 @@ def get_or_create_session_for_wecom(
     external_chat_id: str,
     is_group: bool,
 ) -> SessionMapResult:
-    """Group: session per chat_id; DM: session per user_id.
+    """Group chats share one session per external_chat_id (sentinel user key).
 
-    We store mapping under (tenant, channel, external_chat_id, external_user_id) to keep it simple.
+    Direct chats use one session per external_user_id. Identity binding always uses the real sender id.
     """
     scope = "group" if bool(is_group) else "dm"
     # For DM, external_chat_id is often contact id; we still keep it as provided.