surface reactions: messages get short event_id prefix, reactions appear as standalone timeline events

src/main.rs

@@ -51,12 +51,52 @@ struct PersistedSession {
 }
 
 #[derive(Clone, Debug)]
-struct ChatMessage {
+enum TimelineItem {
+    Message {
+        event_id: OwnedEventId,
         sender: OwnedUserId,
         body: String,
         is_self: bool,
         /// Unix seconds. 0 if unknown.
         ts: i64,
+        in_reply_to: Option<OwnedEventId>,
+    },
+    Reaction {
+        sender: OwnedUserId,
+        target_event_id: OwnedEventId,
+        key: String,
+        is_self: bool,
+        ts: i64,
+    },
+}
+
+impl TimelineItem {
+    fn ts(&self) -> i64 {
+        match self {
+            TimelineItem::Message { ts, .. } | TimelineItem::Reaction { ts, .. } => *ts,
+        }
+    }
+
+    fn event_id(&self) -> Option<&OwnedEventId> {
+        match self {
+            TimelineItem::Message { event_id, .. } => Some(event_id),
+            TimelineItem::Reaction { .. } => None,
+        }
+    }
+
+    fn sender(&self) -> &OwnedUserId {
+        match self {
+            TimelineItem::Message { sender, .. } | TimelineItem::Reaction { sender, .. } => sender,
+        }
+    }
+
+    fn is_self(&self) -> bool {
+        match self {
+            TimelineItem::Message { is_self, .. } | TimelineItem::Reaction { is_self, .. } => {
+                *is_self
+            }
+        }
+    }
 }
 
 struct DaemonState {
@@ -493,28 +533,39 @@ async fn process_loop(state: Arc<Mutex<DaemonState>>, client: Client) {
             )
         };
 
-        // Load recent history from matrix-sdk's persistent event cache
+        // Load recent timeline (messages + reactions) from matrix-sdk's
-        let mut history = match load_recent_messages(&room, max_history).await {
+        // persistent event cache.
-            Ok(h) => h,
+        let mut timeline = match load_timeline(&room, max_history, &own_user).await {
+            Ok(t) => t,
             Err(e) => {
-                tracing::error!(room = %room_id, "failed to load history: {e}");
+                tracing::error!(room = %room_id, "failed to load timeline: {e}");
                 continue;
             }
         };
 
         // For any new messages that reply to events outside the window, fetch
         // the replied-to event from cache and prepend it as extra context.
+        let in_window: std::collections::HashSet<OwnedEventId> = timeline
+            .iter()
+            .filter_map(TimelineItem::event_id)
+            .cloned()
+            .collect();
         let seen_idx_initial = prev_last_shown
             .as_ref()
-            .and_then(|id| history.iter().position(|(eid, _, _, _, _)| eid == id))
+            .and_then(|id| {
+                timeline.iter().position(|t| match t {
+                    TimelineItem::Message { event_id, .. } => event_id == id,
+                    _ => false,
+                })
+            })
             .map_or(0, |pos| pos + 1);
-        let in_window: std::collections::HashSet<OwnedEventId> = history
-            .iter()
-            .map(|(eid, _, _, _, _)| eid.clone())
-            .collect();
         let mut reply_targets: Vec<OwnedEventId> = Vec::new();
-        for (_, _, _, _, in_reply_to) in history.iter().skip(seen_idx_initial) {
+        for item in timeline.iter().skip(seen_idx_initial) {
-            if let Some(target) = in_reply_to {
+            if let TimelineItem::Message {
+                in_reply_to: Some(target),
+                ..
+            } = item
+            {
                 if !in_window.contains(target) && !reply_targets.contains(target) {
                     reply_targets.push(target.clone());
                 }
@@ -523,32 +574,32 @@ async fn process_loop(state: Arc<Mutex<DaemonState>>, client: Client) {
         if !reply_targets.is_empty() {
             if let Ok((cache, _h)) = room.event_cache().await {
                 for target in &reply_targets {
-                    if let Some(found) = fetch_message(&cache, target).await {
+                    if let Some(found) = fetch_message(&cache, target, &own_user).await {
-                        history.insert(0, found);
+                        timeline.insert(0, found);
                     }
                 }
             }
         }
 
-        let chat_msgs: Vec<ChatMessage> = history
+        // Determine seen split based on the last_shown message event id
-            .iter()
-            .map(|(_, sender, body, ts, _)| ChatMessage {
-                sender: sender.clone(),
-                body: body.clone(),
-                is_self: sender == &own_user,
-                ts: *ts,
-            })
-            .collect();
-
-        // Determine seen split: everything before (and including) prev_last_shown is "seen"
         let seen_idx = prev_last_shown
             .as_ref()
-            .and_then(|id| history.iter().position(|(eid, _, _, _, _)| eid == id))
+            .and_then(|id| {
+                timeline.iter().position(|t| match t {
+                    TimelineItem::Message { event_id, .. } => event_id == id,
+                    _ => false,
+                })
+            })
             .map_or(0, |pos| pos + 1);
 
-        let new_last_event_id = history.last().map(|(eid, _, _, _, _)| eid.clone());
+        // The "last shown" pointer should advance to the latest message we've
+        // loaded (not a reaction).
+        let new_last_event_id = timeline.iter().rev().find_map(|t| match t {
+            TimelineItem::Message { event_id, .. } => Some(event_id.clone()),
+            _ => None,
+        });
 
-        let docs = match invoke_claude(&room_id, &room_name, &chat_msgs, seen_idx, &model).await {
+        let docs = match invoke_claude(&room_id, &room_name, &timeline, seen_idx, &model).await {
             Ok(d) => d,
             Err(e) => {
                 tracing::error!(room = %room_id, "claude invocation failed: {e}");
@@ -645,62 +696,151 @@ async fn send_read_receipt(room: &Room, event_id: Option<OwnedEventId>) {
     }
 }
 
-/// Load the last N text messages from the room's persistent event cache.
+/// Load the last N timeline items (messages + reactions) from the room's
-/// Returns oldest-first list of (event_id, sender, body, ts_secs).
+/// persistent event cache. Returns oldest-first.
-/// Returns oldest-first list of (event_id, sender, body, ts_secs, in_reply_to).
+///
-async fn load_recent_messages(
+/// We walk events newest-first, collect messages until we have `limit`, then
+/// also include any reactions whose timestamps fall within the message window.
+async fn load_timeline(
     room: &Room,
     limit: usize,
-) -> anyhow::Result<Vec<(OwnedEventId, OwnedUserId, String, i64, Option<OwnedEventId>)>> {
+    own_user: &OwnedUserId,
+) -> anyhow::Result<Vec<TimelineItem>> {
     use matrix_sdk::ruma::events::AnySyncTimelineEvent;
 
     let (cache, _handles) = room.event_cache().await?;
     let events = cache.events().await;
 
-    let mut out: Vec<(OwnedEventId, OwnedUserId, String, i64, Option<OwnedEventId>)> = Vec::new();
+    let mut messages: Vec<TimelineItem> = Vec::new();
+    let mut reactions: Vec<TimelineItem> = Vec::new();
+    let mut earliest_message_ts: Option<i64> = None;
+
     for ev in events.iter().rev() {
-        if out.len() >= limit {
-            break;
-        }
         let raw = ev.raw();
         let Ok(deserialized) = raw.deserialize() else {
             continue;
         };
-        if let AnySyncTimelineEvent::MessageLike(msg) = deserialized {
+        let AnySyncTimelineEvent::MessageLike(msg) = deserialized else {
-            if let matrix_sdk::ruma::events::AnySyncMessageLikeEvent::RoomMessage(
+            continue;
+        };
+
+        match msg {
+            matrix_sdk::ruma::events::AnySyncMessageLikeEvent::RoomMessage(
                 matrix_sdk::ruma::events::SyncMessageLikeEvent::Original(orig),
-            ) = msg
+            ) => {
-            {
+                if messages.len() >= limit {
-                if let MessageType::Text(text) = &orig.content.msgtype {
+                    continue;
-                    let ts_ms: u64 = orig.origin_server_ts.0.into();
+                }
-                    let ts_secs: i64 = i64::try_from(ts_ms).unwrap_or(0) / 1000;
+                let MessageType::Text(text) = &orig.content.msgtype else {
+                    continue;
+                };
+                let ts = ts_secs_from(orig.origin_server_ts.0);
                 let in_reply_to = match &orig.content.relates_to {
                     Some(matrix_sdk::ruma::events::room::message::Relation::Reply {
                         in_reply_to,
                     }) => Some(in_reply_to.event_id.clone()),
                     _ => None,
                 };
-                    out.push((
+                if earliest_message_ts.is_none_or(|e| ts < e) {
-                        orig.event_id.clone(),
+                    earliest_message_ts = Some(ts);
-                        orig.sender.clone(),
+                }
-                        text.body.clone(),
+                messages.push(TimelineItem::Message {
-                        ts_secs,
+                    event_id: orig.event_id.clone(),
+                    sender: orig.sender.clone(),
+                    body: text.body.clone(),
+                    is_self: &orig.sender == own_user,
+                    ts,
                     in_reply_to,
-                    ));
+                });
             }
+            matrix_sdk::ruma::events::AnySyncMessageLikeEvent::Reaction(
+                matrix_sdk::ruma::events::SyncMessageLikeEvent::Original(orig),
+            ) => {
+                let ts = ts_secs_from(orig.origin_server_ts.0);
+                reactions.push(TimelineItem::Reaction {
+                    sender: orig.sender.clone(),
+                    target_event_id: orig.content.relates_to.event_id.clone(),
+                    key: orig.content.relates_to.key.clone(),
+                    is_self: &orig.sender == own_user,
+                    ts,
+                });
             }
+            _ => {}
         }
     }
 
-    out.reverse();
+    // Drop reactions older than the oldest visible message, to avoid
-    Ok(out)
+    // referencing messages we don't have in scope.
+    if let Some(min_ts) = earliest_message_ts {
+        reactions.retain(|r| r.ts() >= min_ts);
+    }
+
+    let mut combined: Vec<TimelineItem> = Vec::with_capacity(messages.len() + reactions.len());
+    combined.extend(messages);
+    combined.extend(reactions);
+    combined.sort_by_key(TimelineItem::ts);
+    Ok(combined)
+}
+
+/// Render one timeline item into the prompt.
+/// Messages: `[ts] $eid... [(you) ]@user: body`
+/// Reactions: `[ts] [(you) ]@user reacted to $eid... with KEY`
+fn render_timeline_item(prompt: &mut String, item: &TimelineItem) {
+    match item {
+        TimelineItem::Message {
+            event_id,
+            sender,
+            body,
+            is_self,
+            ts,
+            ..
+        } => {
+            let ts_str = format_ts(*ts);
+            let id = short_event_id(event_id);
+            let prefix = if *is_self { "(you) " } else { "" };
+            writeln!(prompt, "[{ts_str}] {id} {prefix}{sender}: {body}").unwrap();
+        }
+        TimelineItem::Reaction {
+            sender,
+            target_event_id,
+            key,
+            is_self,
+            ts,
+        } => {
+            let ts_str = format_ts(*ts);
+            let id = short_event_id(target_event_id);
+            let prefix = if *is_self { "(you) " } else { "" };
+            writeln!(
+                prompt,
+                "[{ts_str}] {prefix}{sender} reacted to {id} with {key}"
+            )
+            .unwrap();
+        }
+    }
+}
+
+/// Shorten an event id for prompt display: `$abc123def456...` → `$abc123de`.
+fn short_event_id(id: &OwnedEventId) -> String {
+    let s = id.as_str();
+    let prefix: String = s.chars().take(9).collect();
+    if s.len() > 9 {
+        format!("{prefix}…")
+    } else {
+        prefix
+    }
+}
+
+fn ts_secs_from(ts: matrix_sdk::ruma::UInt) -> i64 {
+    let ms: u64 = ts.into();
+    i64::try_from(ms).unwrap_or(0) / 1000
 }
 
 /// Fetch a single text message by event_id from the room's event cache.
 async fn fetch_message(
     cache: &matrix_sdk::event_cache::RoomEventCache,
     event_id: &matrix_sdk::ruma::EventId,
-) -> Option<(OwnedEventId, OwnedUserId, String, i64, Option<OwnedEventId>)> {
+    own_user: &OwnedUserId,
+) -> Option<TimelineItem> {
     use matrix_sdk::ruma::events::AnySyncTimelineEvent;
 
     let ev = cache.find_event(event_id).await?;
@@ -717,15 +857,15 @@ async fn fetch_message(
     let MessageType::Text(text) = &orig.content.msgtype else {
         return None;
     };
-    let ts_ms: u64 = orig.origin_server_ts.0.into();
+    let ts = ts_secs_from(orig.origin_server_ts.0);
-    let ts_secs: i64 = i64::try_from(ts_ms).unwrap_or(0) / 1000;
+    Some(TimelineItem::Message {
-    Some((
+        event_id: orig.event_id.clone(),
-        orig.event_id.clone(),
+        sender: orig.sender.clone(),
-        orig.sender.clone(),
+        body: text.body.clone(),
-        text.body.clone(),
+        is_self: &orig.sender == own_user,
-        ts_secs,
+        ts,
-        None,
+        in_reply_to: None,
-    ))
+    })
 }
 
 enum ResponseTarget {
@@ -750,7 +890,7 @@ enum ClaudeDoc {
 async fn invoke_claude(
     source_room: &OwnedRoomId,
     room_name: &str,
-    history: &[ChatMessage],
+    timeline: &[TimelineItem],
     seen_idx: usize,
     model: &str,
 ) -> anyhow::Result<Vec<ClaudeDoc>> {
@@ -766,11 +906,11 @@ async fn invoke_claude(
     )
     .unwrap();
 
-    // Collect unique non-self senders for per-person note discovery
+    // Collect unique non-self participants (message senders + reactors)
-    let mut senders: Vec<&OwnedUserId> = history
+    let mut senders: Vec<&OwnedUserId> = timeline
         .iter()
-        .filter(|m| !m.is_self)
+        .filter(|t| !t.is_self())
-        .map(|m| &m.sender)
+        .map(TimelineItem::sender)
         .collect();
     senders.sort();
     senders.dedup();
@@ -785,30 +925,37 @@ async fn invoke_claude(
         }
     }
 
-    let seen = seen_idx.min(history.len());
+    let seen = seen_idx.min(timeline.len());
-    let (old, new) = history.split_at(seen);
+    let (old, new) = timeline.split_at(seen);
 
     if !old.is_empty() {
-        writeln!(prompt, "\n[previously seen messages — for context]").unwrap();
+        writeln!(prompt, "\n[previously seen events — for context]").unwrap();
-        for msg in old {
+        for item in old {
-            let prefix = if msg.is_self { "(you) " } else { "" };
+            render_timeline_item(&mut prompt, item);
-            let ts = format_ts(msg.ts);
-            writeln!(prompt, "[{ts}] {prefix}{}: {}", msg.sender, msg.body).unwrap();
         }
     }
 
-    writeln!(prompt, "\n[new messages — respond to these]").unwrap();
+    writeln!(prompt, "\n[new events — respond to these]").unwrap();
     if new.is_empty() {
         writeln!(prompt, "(none)").unwrap();
     } else {
-        for msg in new {
+        for item in new {
-            let prefix = if msg.is_self { "(you) " } else { "" };
+            render_timeline_item(&mut prompt, item);
-            let ts = format_ts(msg.ts);
-            writeln!(prompt, "[{ts}] {prefix}{}: {}", msg.sender, msg.body).unwrap();
         }
     }
 
-    tracing::info!("invoking claude: {} new, {} seen", new.len(), old.len());
+    let new_msg_count = new
+        .iter()
+        .filter(|t| matches!(t, TimelineItem::Message { .. }))
+        .count();
+    let new_react_count = new.len() - new_msg_count;
+    tracing::info!(
+        "invoking claude: {} new ({} msg + {} react), {} seen",
+        new.len(),
+        new_msg_count,
+        new_react_count,
+        old.len()
+    );
     tracing::trace!("full prompt:\n{prompt}");
 
     use tokio::process::Command;