slack: drop the eyes-emoji ack post; light the pill on the prior anchor

cli/src/slack/bridge.ts

@@ -518,34 +518,30 @@ export async function runSlackBridge(opts: BridgeOptions): Promise<void> {
     recentRelays.set(decision.slug, relays);
     pasteTmuxPrompt(decision.slug, prompt); // tmux session name == slug
 
-    // Post a tiny bot-authored anchor so the working pill has something to
-    // attach to immediately. Without this the channel looks dead for the
-    // 10-20s between the human typing and the agent's first text post (which
-    // is the natural pill anchor for the existing flow). Slack's
-    // assistant.threads.setStatus only accepts thread roots authored by the
-    // app itself, so we cannot pin the pill to the human's own message
-    // (that would 400 with invalid_thread_ts).
+    // Light the working pill immediately so the channel shows activity
+    // within seconds of the relay, instead of looking dead until the
+    // agent's first text post (10-20s later). Slack's
+    // assistant.threads.setStatus only accepts thread roots authored by
+    // the app itself, so the human's own message ts is not a valid
+    // anchor. Instead, reuse the most recent app-authored anchor for the
+    // slug (the previous turn's text post). No new bot message is
+    // posted — the pill rides on the existing prior reply. The agent's
+    // first text post in this turn will then become the new anchor via
+    // the existing post loop, moving the pill onto it.
     //
-    // The eyes emoji is chosen for minimal noise: a single character that
-    // reads as "received, on it" and renders cleanly next to the working
-    // pill ("is working…"). The agent's first real text post will become
-    // the new thread root via the existing post loop, draining any tools
-    // accumulated in the meantime under this anchor and moving the pill
-    // onto the narrative text.
+    // If there is no prior anchor (very first interaction with the
+    // agent), skip: the pill will appear the natural way on the first
+    // reply. That happens once per agent lifetime, so the one-time
+    // delay is acceptable to avoid a noise-only bot post.
     if (event.channel) {
-      try {
-        const ts = await client.post(event.channel, "👀");
-        if (ts) {
-          writeThreadRoot(decision.slug, ts);
-          try {
-            await client.setStatus(event.channel, ts, WORKING_LABEL);
-            setActivePill(decision.slug, { channelId: event.channel, threadTs: ts, label: WORKING_LABEL, lastSetMs: Date.now() });
-          } catch (e) {
-            console.error(`setStatus on relay anchor for ${decision.slug} failed:`, e);
-          }
+      const prevAnchor = readThreadRoot(decision.slug);
+      if (prevAnchor) {
+        try {
+          await client.setStatus(event.channel, prevAnchor, WORKING_LABEL);
+          setActivePill(decision.slug, { channelId: event.channel, threadTs: prevAnchor, label: WORKING_LABEL, lastSetMs: Date.now() });
+        } catch (e) {
+          console.error(`setStatus on prior anchor for ${decision.slug} failed:`, e);
         }
-      } catch (e) {
-        console.error(`relay anchor post for ${decision.slug} failed:`, e);
       }
     }
   });