feat(gateway): support [[reply_to]] directive for gateway platforms

docs/feishu.md

@@ -200,6 +200,22 @@ To start a threaded conversation: reply to any bot message in a group chat (long
 
 Streaming (typewriter) mode works in threads — edits target the same message regardless of thread context.
 
+## Agent-Controlled Reply-To
+
+Agents can reply to a specific message using the `[[reply_to:message_id]]` output directive (see [docs/output-directives.md](output-directives.md)). The gateway sends the reply via Feishu's native Reply API, showing a quote reference in the UI.
+
+```
+Agent output:
+  [[reply_to:om_xxx]]
+  This is my reply to that specific message.
+```
+
+**How agents get message IDs:** Every incoming message includes `message_id` in the `SenderContext` injected into the agent prompt. Agents can store and reference these IDs to reply to specific messages.
+
+**Fallback:** If the specified message ID is invalid or the Reply API fails, the gateway automatically falls back to a plain send (no quote).
+
+**Use case:** In multi-bot threads, each bot can reply to a different message, creating clear visual conversation threads within a Feishu thread.
+
 ## Bot-to-Bot Collaboration (Gateway-Side Only)
 
 The gateway adapter includes bot identification and filtering scaffolding (`AllowBots` enum, `FEISHU_TRUSTED_BOT_IDS`, `FEISHU_MAX_BOT_TURNS` with human-reset safety valve), matching Discord's `allow_bot_messages` design.

docs/output-directives.md

@@ -35,6 +35,7 @@ Here is my reply to that specific message.
 
 **Behavior**:
 - Discord: sends with `message_reference`, showing the native "replying to..." UI
+- Feishu: sends via Reply API (`POST /im/v1/messages/{id}/reply`), showing native quote UI
 - Invalid/non-existent message ID: silently falls back to plain send
 - Works in both streaming and send-once modes
 
@@ -73,4 +74,4 @@ This creates clear visual conversation threads within a Discord thread — essen
 | Hermes Agent | `DISCORD_REPLY_TO_MODE` env var | ❌ Platform decides, always to trigger msg |
 | **OAB** | `[[reply_to:message_id]]` directive | ✅ Agent chooses any message |
 
-> **Note:** `reply_to` is currently implemented for Discord only. Slack message IDs (ts format like `1234567890.123456`) are accepted by the parser but the Slack adapter does not yet send threaded replies via this directive — it falls back to plain send. Slack support can be added in a future PR.
+> **Note:** `reply_to` is currently implemented for Discord and Feishu (gateway). Slack message IDs (ts format like `1234567890.123456`) are accepted by the parser but the Slack adapter does not yet send threaded replies via this directive — it falls back to plain send. Slack support can be added in a future PR.

gateway/src/adapters/feishu.rs

@@ -1904,14 +1904,25 @@ pub async fn handle_reply(
     let api_base = adapter.config.api_base();
     let text = &reply.content.text;
     let limit = adapter.config.message_limit;
+    // quote_message_id (agent-controlled reply-to) takes priority over thread_id
+    let reply_target = reply.quote_message_id.as_deref()
+        .or(reply.channel.thread_id.as_deref());
     let thread_id = reply.channel.thread_id.as_deref();
 
     // Split long messages; store sent message_ids in dedupe to prevent
     // self-echo (Feishu pushes bot's own messages back via WebSocket)
     // Use post (rich text) format for markdown rendering.
     // When in a thread (thread_id present), use reply API to stay in the same thread.
     if text.len() <= limit {
-        match send_post_message(&adapter.client, &api_base, &token, &reply.channel.id, thread_id, text).await {
+        let result = send_post_message(&adapter.client, &api_base, &token, &reply.channel.id, reply_target, text).await;
+        // Fallback: if quote_message_id caused failure, retry without it
+        let result = if result.is_none() && reply.quote_message_id.is_some() {
+            tracing::warn!(quote_message_id = ?reply.quote_message_id, channel_id = %reply.channel.id, "reply-to failed, falling back to plain send");
+            send_post_message(&adapter.client, &api_base, &token, &reply.channel.id, thread_id, text).await
+        } else {
+            result
+        };
+        match result {
             Some(msg_id) => {
                 adapter.dedupe.is_duplicate(&msg_id);
                 // Record thread participation for mention bypass
@@ -1953,11 +1964,21 @@ pub async fn handle_reply(
     } else {
         let mut sent_any = false;
         for chunk in split_text(text, limit) {
-            if let Some(msg_id) = send_post_message(&adapter.client, &api_base, &token, &reply.channel.id, thread_id, chunk).await {
+            if let Some(msg_id) = send_post_message(&adapter.client, &api_base, &token, &reply.channel.id, reply_target, chunk).await {
                 adapter.dedupe.is_duplicate(&msg_id);
                 sent_any = true;
             }
         }
+        // Fallback: if quote_message_id caused all chunks to fail, retry without it
+        if !sent_any && reply.quote_message_id.is_some() {
+            tracing::warn!(quote_message_id = ?reply.quote_message_id, channel_id = %reply.channel.id, "chunked reply-to failed, falling back to plain send");
+            for chunk in split_text(text, limit) {
+                if let Some(msg_id) = send_post_message(&adapter.client, &api_base, &token, &reply.channel.id, thread_id, chunk).await {
+                    adapter.dedupe.is_duplicate(&msg_id);
+                    sent_any = true;
+                }
+            }
+        }
         if sent_any {
             if let Some(tid) = thread_id {
                 record_participation(&adapter.participated_threads, tid, adapter.config.session_ttl_secs);
@@ -2942,4 +2963,80 @@ mod tests {
         // (caller would pass false because Mentions mode always returns false)
         assert!(parse_message_event(&env, Some("ou_bot"), &cfg, false).is_none());
     }
+
+    #[test]
+    fn quote_message_id_takes_priority_over_thread_id() {
+        use crate::schema::{GatewayReply, ReplyChannel, Content};
+        let reply = GatewayReply {
+            schema: "openab.gateway.reply.v1".into(),
+            reply_to: "evt_123".into(),
+            platform: "feishu".into(),
+            channel: ReplyChannel {
+                id: "chat_123".into(),
+                thread_id: Some("om_root".into()),
+            },
+            content: Content {
+                content_type: "text".into(),
+                text: "hello".into(),
+                attachments: vec![],
+            },
+            command: None,
+            request_id: None,
+            quote_message_id: Some("om_specific".into()),
+        };
+        // quote_message_id should take priority
+        let reply_target = reply.quote_message_id.as_deref()
+            .or(reply.channel.thread_id.as_deref());
+        assert_eq!(reply_target, Some("om_specific"));
+    }
+
+    #[test]
+    fn reply_target_falls_back_to_thread_id_when_no_quote() {
+        use crate::schema::{GatewayReply, ReplyChannel, Content};
+        let reply = GatewayReply {
+            schema: "openab.gateway.reply.v1".into(),
+            reply_to: "evt_123".into(),
+            platform: "feishu".into(),
+            channel: ReplyChannel {
+                id: "chat_123".into(),
+                thread_id: Some("om_root".into()),
+            },
+            content: Content {
+                content_type: "text".into(),
+                text: "hello".into(),
+                attachments: vec![],
+            },
+            command: None,
+            request_id: None,
+            quote_message_id: None,
+        };
+        let reply_target = reply.quote_message_id.as_deref()
+            .or(reply.channel.thread_id.as_deref());
+        assert_eq!(reply_target, Some("om_root"));
+    }
+
+    #[test]
+    fn reply_target_is_none_when_both_absent() {
+        use crate::schema::{GatewayReply, ReplyChannel, Content};
+        let reply = GatewayReply {
+            schema: "openab.gateway.reply.v1".into(),
+            reply_to: "evt_123".into(),
+            platform: "feishu".into(),
+            channel: ReplyChannel {
+                id: "chat_123".into(),
+                thread_id: None,
+            },
+            content: Content {
+                content_type: "text".into(),
+                text: "hello".into(),
+                attachments: vec![],
+            },
+            command: None,
+            request_id: None,
+            quote_message_id: None,
+        };
+        let reply_target = reply.quote_message_id.as_deref()
+            .or(reply.channel.thread_id.as_deref());
+        assert_eq!(reply_target, None);
+    }
 }

gateway/src/adapters/googlechat.rs

@@ -1375,6 +1375,7 @@ mod tests {
             },
             command: None,
             request_id: Some("req_123".into()),
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;
@@ -1418,6 +1419,7 @@ mod tests {
             },
             command: None,
             request_id: Some("req_fail".into()),
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;
@@ -1465,6 +1467,7 @@ mod tests {
             },
             command: None,
             request_id: Some("req_empty".into()),
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;
@@ -1509,6 +1512,7 @@ mod tests {
             },
             command: None,
             request_id: Some("req_multi_fail".into()),
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;
@@ -1543,6 +1547,7 @@ mod tests {
             },
             command: None,
             request_id: Some("req_notoken".into()),
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;
@@ -1588,6 +1593,7 @@ mod tests {
             },
             command: Some("edit_message".into()),
             request_id: None,
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;
@@ -1630,6 +1636,7 @@ mod tests {
             },
             command: None,
             request_id: Some("req_multi".into()),
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;
@@ -1687,6 +1694,7 @@ mod tests {
             },
             command: None,
             request_id: Some("req_partial".into()),
+            quote_message_id: None,
         };
 
         adapter.handle_reply(&reply, &event_tx).await;

gateway/src/main.rs

@@ -427,6 +427,7 @@ mod tests {
             },
             command: None,
             request_id: None,
+            quote_message_id: None,
         }
     }
 

gateway/src/schema.rs

@@ -64,6 +64,12 @@ pub struct GatewayReply {
     pub command: Option<String>,
     #[serde(default)]
     pub request_id: Option<String>,
+    /// When set, send this message as a reply/quote to the specified platform message ID.
+    /// Unlike `reply_to` (which identifies the triggering event for routing/dedup),
+    /// this field controls the visual reply/quote UI on the platform.
+    /// If quoting fails, the gateway MUST fall back to sending without quoting.
+    #[serde(default)]
+    pub quote_message_id: Option<String>,
 }
 
 #[derive(Clone, Debug, Serialize, Deserialize)]