fix(whatsapp): detect group @mentions when self is in allowFrom (#49317)

`isBotMentionedFromTargets` derives a "self-chat" flag from
`isSelfChatMode(self.e164, allowFrom)`, which is a config-shaped check
("is the bot's own E.164 in allowFrom?") rather than a conversation-shaped
check. The flag flipped to true for ordinary group conversations whenever
the operator put their own number in allowFrom — the common config — and
the function then took the `hasMentions && isSelfChat` branch, dropping
the identity-overlap check entirely.

Group `@mention` JIDs (including WhatsApp LID mentions like
`216372600647751@lid` that resolve to the bot's E.164 via the LID reverse
mapping) silently failed mention detection. The empty fall-through then
fell back to body-regex / E.164-digit matching, which does not match
WhatsApp's tap-mention markup, so `wasMentioned` stayed false even when
the bot was clearly mentioned.

Confine the implicit `isSelfChat` derivation to non-group conversations.
Real group `@mentions` go through the identity-overlap check regardless of
allowFrom. Explicit `mentionCfg.isSelfChat` overrides from callers
(multi-account / precomputed paths) keep their existing behavior so we do
not regress callers that intentionally set the flag.

Closes #49317

CHANGELOG.md

@@ -90,6 +90,7 @@ Docs: https://docs.openclaw.ai
 - Agents/context engines: preserve the child agent's configured `agentDir` when subagent cleanup re-resolves a context engine, so `onSubagentEnded` hooks keep operating on the correct per-agent state. (#67243) Thanks @jarimustonen.
 - Channels/WhatsApp: restrict pairing verification replies to real inbound user content, preventing unsolicited prompts from receipts, typing indicators, presence updates, and other non-message Baileys upserts. Fixes #73797. (#73823) Thanks @hclsys.
 - Configure/Ollama: show the configured Ollama model allowlist after Cloud only or Cloud + Local setup and skip slow per-model cloud metadata fetches. (#73995) Thanks @obviyus.
+- Channels/WhatsApp: stop dropping group `@mention` detection when the bot's own E.164 is in `allowFrom`, so LID-style group mentions resolve correctly through the identity-overlap check; the original 1:1 self-DM mention-skip suppression still applies for direct chats and explicit `isSelfChat` overrides. Fixes #49317. Thanks @juan-flores077.
 
 ## 2026.4.27
 

extensions/whatsapp/src/auto-reply/mentions.ts

@@ -10,6 +10,7 @@ import {
   identitiesOverlap,
   type WhatsAppIdentity,
 } from "../identity.js";
+import { isWhatsAppGroupJid } from "../normalize-target.js";
 import { isSelfChatMode, normalizeE164 } from "../text-runtime.js";
 import type { WebInboundMsg } from "./types.js";
 
@@ -44,10 +45,21 @@ export function isBotMentionedFromTargets(
     // Remove zero-width and directionality markers WhatsApp injects around display names
     normalizeMentionText(text);
 
-  const isSelfChat =
-    typeof mentionCfg.isSelfChat === "boolean"
-      ? mentionCfg.isSelfChat
-      : isSelfChatMode(targets.self.e164, mentionCfg.allowFrom);
+  const explicitSelfChatOverride = typeof mentionCfg.isSelfChat === "boolean";
+  // `isSelfChatMode` is a config-shaped check ("is the bot's own E.164 in
+  // allowFrom?"), not a conversation-shaped check, so it returns true even
+  // for group conversations whenever the operator put their own number in
+  // allowFrom — which is the common config. The original mention-skip path
+  // was designed to prevent owner-mentioning-self in a true 1:1 self DM
+  // from falsely triggering the bot, so when we derive the flag implicitly
+  // from `allowFrom`, confine the suppression to non-group conversations
+  // and let real group @mentions go through the identity-overlap check
+  // (#49317). Explicit `mentionCfg.isSelfChat` overrides from the caller
+  // are honored as-is so multi-account / precomputed paths keep working.
+  const isGroupConversation = isWhatsAppGroupJid(msg.from);
+  const isSelfChat = explicitSelfChatOverride
+    ? Boolean(mentionCfg.isSelfChat)
+    : isSelfChatMode(targets.self.e164, mentionCfg.allowFrom) && !isGroupConversation;
 
   const hasMentions = targets.normalizedMentions.length > 0;
   if (hasMentions && !isSelfChat) {
@@ -59,7 +71,7 @@ export function isBotMentionedFromTargets(
     // If the message explicitly mentions someone else, do not fall back to regex matches.
     return false;
   } else if (hasMentions && isSelfChat) {
-    // Self-chat mode: ignore WhatsApp @mention JIDs, otherwise @mentioning the owner in group chats triggers the bot.
+    // Self-chat mode: ignore WhatsApp @mention JIDs, otherwise @mentioning the owner in self-chat triggers the bot.
   }
   const bodyClean = clean(msg.body);
   if (mentionCfg.mentionRegexes.some((re) => re.test(bodyClean))) {

extensions/whatsapp/src/auto-reply/web-auto-reply-utils.test.ts

@@ -70,9 +70,15 @@ describe("isBotMentionedFromTargets", () => {
     expectMentioned(msg, mentionCfg, true);
   });
 
-  it("ignores JID mentions in self-chat mode", () => {
+  it("ignores JID mentions in a true 1:1 self-chat (not a group)", () => {
     const cfg = { mentionRegexes: [/\bopenclaw\b/i], allowFrom: ["+999"] };
     const msg = makeMsg({
+      // Direct chat with self, not a group — the original "ignore mentions
+      // in self-chat" suppression still applies here so that mentioning the
+      // owner in their own DM does not falsely trigger the bot.
+      from: "999@s.whatsapp.net",
+      conversationId: "999@s.whatsapp.net",
+      chatType: "direct",
       body: "@owner ping",
       mentionedJids: ["999@s.whatsapp.net"],
       selfE164: "+999",
@@ -81,6 +87,9 @@ describe("isBotMentionedFromTargets", () => {
     expectMentioned(msg, cfg, false);
 
     const msgTextMention = makeMsg({
+      from: "999@s.whatsapp.net",
+      conversationId: "999@s.whatsapp.net",
+      chatType: "direct",
       body: "openclaw ping",
       selfE164: "+999",
       selfJid: "999@s.whatsapp.net",
@@ -88,6 +97,25 @@ describe("isBotMentionedFromTargets", () => {
     expectMentioned(msgTextMention, cfg, true);
   });
 
+  it("detects an explicit group @mention even when self is in allowFrom (#49317)", () => {
+    // Operator config commonly puts their own E.164 in allowFrom so they can
+    // run owner-only commands in groups; previously, that flipped the gate
+    // to "self-chat mode" and silently dropped mention detection in groups,
+    // including LID-style WhatsApp mentions that resolve to the bot's own
+    // E.164. After the fix, group conversations honor the identity-overlap
+    // check regardless of allowFrom.
+    const cfg = { mentionRegexes: [/\bopenclaw\b/i], allowFrom: ["+15551234567"] };
+    const msg = makeMsg({
+      // Default `from` is the @g.us group JID from `makeMsg`.
+      body: "@216372600647751 can you see this?",
+      mentionedJids: ["216372600647751@lid"],
+      selfE164: "+15551234567",
+      selfJid: "15551234567@s.whatsapp.net",
+      selfLid: "216372600647751@lid",
+    });
+    expectMentioned(msg, cfg, true);
+  });
+
   it("honors explicit self-chat overrides without recomputing from allowFrom", () => {
     const cfg = {
       mentionRegexes: [/\bopenclaw\b/i],