feat(agent): v4.15.0 - panneau Agent style Notion AI (mentions @ inline + skills / + actions sous chaque réponse)

- Composeur contenteditable : mentions inline (pastilles icône+nom), barre de contexte groupée documents/skills, sélecteur @ (icône+titre+chemin), skills intégrés + agent_skills + commandes admin via '/'
- Actions sous chaque réponse agent : copier, insérer dans la page, feedback 👍/👎
- Backend : table agent_feedback, GET /api/agent/mentions, POST /api/agent/feedback, contexte document:<id>/page:<id>/collection:<id> résolu en contenu réel
- Inclut le travail v4.14.0 (documents/espace de travail + outils + auto-titre de conversation)

app/routers/agent.py

@@ -442,6 +442,98 @@ async def apply_skill(request: Request, skill_id: int):
     return {"conversation_id": cur.lastrowid, "skill": skill["name"], "status": "ready"}
 
 
+# ── Mentions (commande @ / +) & feedback (boutons 👍 / 👎) ──
+
+
+@router.get("/mentions")
+async def list_mentions(request: Request, q: str = ""):
+    """Éléments mentionnables dans le panneau agent (commande « @ » / bouton « + »).
+
+    Retourne un flux unique d'objets FlowDeck (documents éditeur, collections,
+    pages de collection) que l'utilisateur peut épingler au contexte du chat.
+    Chaque entrée porte un jeton ``token`` (``document:<id>``, ``collection:<id>``,
+    ``page:<id>``) que ContextBuilder résout en contenu réel pour le LLM.
+    """
+    query = (q or "").strip()[:60]
+    like = f"%{query}%"
+    limit = 8
+    items: list[dict] = []
+
+    with get_conn() as conn:
+        # 1) Documents (pages éditeur) — token document:<id>
+        sql_docs = (
+            "SELECT p.id, p.title, p.workspace_id, w.name AS ws_name "
+            "FROM pages p LEFT JOIN workspaces w ON w.id=p.workspace_id "
+            "WHERE p.deleted_at IS NULL AND (? = '' OR p.title LIKE ?) "
+            "ORDER BY p.updated_at DESC LIMIT ?"
+        )
+        for r in conn.execute(sql_docs, (query, like, limit)).fetchall():
+            items.append({
+                "type": "document", "icon": "📄",
+                "id": r["id"], "label": r["title"] or "Sans titre",
+                "sub": r["ws_name"] or "", "token": f"document:{r['id']}",
+            })
+
+        # 2) Collections (bases de données) — token collection:<id>
+        sql_colls = (
+            "SELECT id, name, icon FROM collections "
+            "WHERE (? = '' OR name LIKE ?) ORDER BY updated_at DESC LIMIT ?"
+        )
+        for r in conn.execute(sql_colls, (query, like, limit)).fetchall():
+            items.append({
+                "type": "collection", "icon": r["icon"] or "🗄️",
+                "id": r["id"], "label": r["name"],
+                "sub": "Base de données", "token": f"collection:{r['id']}",
+            })
+
+        # 3) Pages de collection — token page:<id>
+        sql_pages = (
+            "SELECT cp.id, cp.title, cp.icon, c.name AS coll_name "
+            "FROM collection_pages cp LEFT JOIN collections c ON c.id=cp.collection_id "
+            "WHERE (? = '' OR cp.title LIKE ?) ORDER BY cp.updated_at DESC LIMIT ?"
+        )
+        for r in conn.execute(sql_pages, (query, like, limit)).fetchall():
+            items.append({
+                "type": "page", "icon": r["icon"] if str(r["icon"]).startswith(("📄", "📑", "🗂️", "📁")) else "📑",
+                "id": r["id"], "label": r["title"] or "Sans titre",
+                "sub": f"Page · {r['coll_name']}" if r["coll_name"] else "Page",
+                "token": f"page:{r['id']}",
+            })
+
+    return {"mentions": items}
+
+
+@router.post("/feedback")
+async def add_feedback(request: Request):
+    """Enregistre le retour (👍 / 👎) porté sur une réponse de l'agent."""
+    user_id = await _current_user_id(request)
+    body = await request.json() if request.headers.get("content-type") else {}
+    rating = (body.get("rating") or "").strip().lower()
+    if rating not in ("up", "down"):
+        raise HTTPException(status_code=400, detail="rating doit être 'up' ou 'down'")
+    conversation_id = body.get("conversation_id")
+    message_id = body.get("message_id")
+    snippet = (body.get("content") or "")[:1000].strip()
+    comment = (body.get("comment") or "").strip()[:2000]
+    try:
+        conversation_id = int(conversation_id) if conversation_id not in (None, "") else None
+    except (TypeError, ValueError):
+        conversation_id = None
+    try:
+        message_id = int(message_id) if message_id not in (None, "") else None
+    except (TypeError, ValueError):
+        message_id = None
+    with get_conn() as conn:
+        cur = conn.execute(
+            """INSERT INTO agent_feedback
+               (conversation_id, message_id, user_id, rating, snippet, comment)
+               VALUES (?,?,?,?,?,?)""",
+            (conversation_id, message_id, user_id, rating, snippet, comment),
+        )
+        conn.commit()
+    return {"id": cur.lastrowid, "rating": rating, "status": "recorded"}
+
+
 # ── Triggers & tools ──