havre: a photo or clip they sent is told to the model in their own turn, as a stage direction, not in yours (characters took "you sent me a photo" as yours and got confused); a reply that narrates sending a photo makes one likely

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VAvUH72XsdCitHdh3t7mjr

ventures/havre/abilities.js

@@ -270,25 +270,27 @@ export function clipOptions(model, last = {}) {
 
 /**
  * A conversation with pictures and voice notes in it, as plain messages for a text model. A photo or clip the
- * character sent is mentioned in the next thing the person says ("(You sent me a photo: …)"), so the model knows
+ * character sent is in their own turn, as a stage direction ("*sends a photo: …*"), the way characters narrate what
- * about it without learning to write "[photo]" itself. Your photo is "[a photo: what it shows]", your voice note is
+ * they do, so the model knows what it sent and who sent it (told in the person's turn, "you sent me a photo" read as
- * the words you said.
+ * the person's, and characters got confused about whose photo it was). Your photo is "[a photo: what it shows]", your
+ * voice note is the words you said.
  */
 export function apiHistory(history) {
   const out = [];
-  let notes = [];
   for (const m of history || []) {
     if (!m || m.demo) continue;
     const media = m.media || null;
     if (m.role === "assistant") {
       if (media && (media.kind === "photo" || media.kind === "clip")) {
         const shot = clip(str(media.shot).trim(), 300).replace(/[.!?…\s]+$/, "");
-        notes.push(`(You sent me ${media.kind === "clip" ? "a short video" : "a photo"}${shot ? ": " + shot : ""}.)`);
+        const sent = `*sends ${media.kind === "clip" ? "a short video" : "a photo"}${shot ? ": " + shot : ""}*`;
+        const last = out.at(-1);
+        if (last && last.role === "assistant") last.content += "\n\n" + sent;   // with the words it came after
+        else out.push({ role: "assistant", content: sent });
       } else if (str(m.content).trim()) out.push({ role: "assistant", content: str(m.content) });
     } else if (m.role === "user") {
       let content = str(m.content);
       if (media && media.kind === "photo-in") content = `[a photo${media.saw ? ": " + media.saw : ""}]` + (content ? " " + content : "");
-      if (notes.length) { content = notes.join(" ") + (content ? "\n\n" + content : ""); notes = []; }
       out.push({ role: "user", content, ...(m.hidden ? { hidden: true } : {}) });
     }
   }
@@ -357,7 +359,7 @@ export function normalizeMove(o) {
 
 // words that make a move likely: asking to see or hear them, or a reply about where they are and what they're doing
 const ASKS = /\b(pics?|picture|photo|selfie|snap|show me|see you|see it|what (?:do )?you look like|look like|wearing|outfit|send (?:me )?(?:a|one|it)|video|clip|film|hear you|your voice|say it|sing|whisper|out loud|aloud)\b/i;
-const SHOWS = /\b(look at (?:this|my|the)|wearing|outfit|dress|view from|sunset|sunrise|my (?:room|place|desk|garden|kitchen|car|boat|bed|window)|just got (?:back|a|my)|i'?m (?:at|in|on) the|picture|photo|selfie|whisper)\b/i;
+const SHOWS = /\b(sends? (?:you )?(?:a |another )?(?:photo|pic|selfie|snap)|look at (?:this|my|the)|wearing|outfit|dress|view from|sunset|sunrise|my (?:room|place|desk|garden|kitchen|car|boat|bed|window)|just got (?:back|a|my)|i'?m (?:at|in|on) the|picture|photo|selfie|whisper)\b/i;
 
 /**
  * Whether to think about a move after this reply: when the person asked to see or hear them, when the reply talks