feat(frontends): stapp2 UI enhancements with modular stapp2_extensions

frontends/stapp2.py

@@ -1,3 +1,14 @@
+"""
+GenericAgent Streamlit UI (stapp2)
+
+Upstream baseline: lsdefine/GenericAgent frontends/stapp2.py
+Extensions: see stapp2_extensions.py (attachments, sidebar extras, upload UI)
+
+Code review | 审查顺序
+  1. stapp2_extensions.py — attachment & sidebar enhancements
+  2. stapp2.py            — upstream shell + turn expanders (Module F)
+"""
+
 import os, sys
 import html
 if sys.stdout is None: sys.stdout = open(os.devnull, "w")
@@ -18,9 +29,11 @@
 from datetime import datetime
 from agentmain import GeneraticAgent
 
+from frontends import stapp2_extensions as ext
+
 st.set_page_config(page_title="Cowork", layout="wide")
 
-# ─── Anthropic Light Theme CSS ───
+# ── Upstream: Anthropic light theme CSS | 上游：Anthropic 浅色主题 ─────────
 ANTHROPIC_CSS = """
 <style>
 /* ===== Root variables ===== */
@@ -656,6 +669,7 @@
 </style>
 """
 
+# ── Upstream: sidebar selectbox width fix JS | 上游：侧栏下拉框宽度修正 ───
 ANTHROPIC_SELECTBOX_SCRIPT = """
 <div></div>
 <script>
@@ -797,6 +811,8 @@
 </script>
 """
 
+# ── Upstream: agent init & theme helpers | 上游：Agent 与主题工具 ─────────
+
 @st.cache_resource
 def init():
     agent = GeneraticAgent()
@@ -941,29 +957,36 @@ def build_header_agent_badge_script() -> str:
 </script>
 """
 
+# ── App bootstrap | 应用启动 ─────────────────────────────────────────────
+
 agent = init()
 
+_UPSTREAM_SESSION_DEFAULTS = {
+    'agent_name': 'GenericAgent', 'streaming': False, 'stopping': False, 'display_queue': None,
+    'partial_response': '', 'reply_ts': '', 'current_prompt': '', 'selected_llm_idx': agent.llm_no,
+    'autonomous_enabled': False, 'messages': [],
+}
+
 def init_session_state():
-    for key, value in {
-        'agent_name': 'GenericAgent', 'streaming': False, 'stopping': False, 'display_queue': None,
-        'partial_response': '', 'reply_ts': '', 'current_prompt': '', 'selected_llm_idx': agent.llm_no,
-        'autonomous_enabled': False, 'messages': [],
-    }.items(): st.session_state.setdefault(key, value)
+    for key, value in _UPSTREAM_SESSION_DEFAULTS.items():
+        st.session_state.setdefault(key, value)
+    ext.register_extension_session_state()
 
 init_session_state()
 
-# Inject Anthropic theme
 st.markdown(ANTHROPIC_CSS, unsafe_allow_html=True)
 st.markdown(build_dynamic_font_css(110.0), unsafe_allow_html=True)
 _embed_html(ANTHROPIC_SELECTBOX_SCRIPT, height=0, width=0)
 _embed_html(build_header_agent_badge_script(), height=0, width=0)
+ext.inject_extension_assets()
 
 st.session_state.agent_name = 'Generic Agent'
 with st.chat_message("assistant"):
     st.markdown(f'<div class="msg-timestamp">{datetime.now().strftime("%Y-%m-%d %H:%M:%S")}</div>', unsafe_allow_html=True)
     st.write("欢迎使用GenericAgent~")
 
 
+# ── Upstream: sidebar (LLM switch) + fork extras | 上游侧栏 + fork 扩展 ──
 @st.fragment
 def render_sidebar():
     llm_options, current_idx = agent.list_llms(), agent.llm_no
@@ -978,13 +1001,13 @@ def render_sidebar():
         st.toast(f"已切换到备用链路：{llm_labels[selected_idx]}")
         st.rerun()
     st.divider()
-    if st.button("重新注入System Prompt"):
-        agent.llmclient.last_tools = ''
-        st.toast("下次将重新注入System Prompt")
+    ext.render_sidebar_extras(agent)
 
 with st.sidebar: render_sidebar()
 
 
+# ── Upstream: agent task & chat loop | 上游：任务与对话循环 ───────────────
+
 def start_agent_task(prompt):
     st.session_state.display_queue = agent.put_task(prompt, source="user")
     st.session_state.streaming, st.session_state.stopping, st.session_state.partial_response = True, False, ''
@@ -1015,17 +1038,44 @@ def poll_agent_output(max_items=20):
 def _get_response_segments(text):
     return [p for p in re.split(r'(?=\*\*LLM Running \(Turn \d+\) \.\.\.\*\*)', text) if p.strip()] or [text]
 
-def render_message(role, content, ts='', unsafe_allow_html=True):
+
+_TURN_RUNNING_RE = re.compile(r'^\*\*LLM Running \(Turn \d+\)')
+
+
+def _is_intermediate_turn_segment(segment: str, *, is_last: bool) -> bool:
+    """Fold non-final **LLM Running (Turn N)** blocks into expanders | 将非最终 Turn 块折叠为 expander"""
+    return (not is_last) and bool(_TURN_RUNNING_RE.match(segment.strip()))
+
+
+def _turn_expander_label(content: str) -> str:
+    m = re.match(r'^\*\*LLM Running \(Turn (\d+)\)', content.strip())
+    return f"Turn {m.group(1)} · 推理过程" if m else "推理过程"
+
+
+def render_message(role, content, ts='', unsafe_allow_html=True, intermediate=False):
     with st.chat_message(role):
         if ts: st.markdown(f'<div class="msg-timestamp">{ts}</div>', unsafe_allow_html=True)
-        st.markdown(content, unsafe_allow_html=unsafe_allow_html)
+        if intermediate:
+            with st.expander(_turn_expander_label(content), expanded=False):
+                st.markdown(content, unsafe_allow_html=unsafe_allow_html)
+        else:
+            st.markdown(content, unsafe_allow_html=unsafe_allow_html)
+
 
 def finish_streaming_message():
     reply_ts = st.session_state.reply_ts
-    st.session_state.messages.extend({"role": "assistant", "content": seg, "time": reply_ts} for seg in _get_response_segments(st.session_state.partial_response))
+    segments = _get_response_segments(st.session_state.partial_response)
+    for i, seg in enumerate(segments):
+        st.session_state.messages.append({
+            "role": "assistant",
+            "content": seg,
+            "time": reply_ts,
+            "intermediate": _is_intermediate_turn_segment(seg, is_last=(i == len(segments) - 1)),
+        })
     st.session_state.last_reply_time = int(time.time())
     st.session_state.partial_response = st.session_state.reply_ts = st.session_state.current_prompt = ''
 
+
 def render_streaming_area():
     if not st.session_state.streaming: return
     with st.container():
@@ -1035,15 +1085,34 @@ def render_streaming_area():
     reply_ts = st.session_state.reply_ts
     with st.empty().container():
         segments = _get_response_segments(st.session_state.partial_response)
-        for i, seg in enumerate(segments): render_message("assistant", seg + ("" if i < len(segments) - 1 else "▌"), ts=reply_ts, unsafe_allow_html=False)
+        for i, seg in enumerate(segments):
+            is_last = (i == len(segments) - 1)
+            render_message(
+                "assistant",
+                seg + ("" if not is_last else "▌"),
+                ts=reply_ts,
+                unsafe_allow_html=False,
+                intermediate=_is_intermediate_turn_segment(seg, is_last=is_last),
+            )
     if poll_agent_output(): finish_streaming_message()
     else: time.sleep(0.2)
     st.rerun()
 
-for msg in st.session_state.messages: render_message(msg["role"], msg["content"], ts=msg.get("time", ""), unsafe_allow_html=True)
-if st.session_state.streaming: render_streaming_area()
-if prompt := st.chat_input("请输入指令", disabled=st.session_state.streaming):
-    st.session_state.messages.append({"role": "user", "content": prompt, "time": datetime.now().strftime("%Y-%m-%d %H:%M:%S")})
-    start_agent_task(prompt)
+
+# ── Main UI loop | 主界面 ─────────────────────────────────────────────────
+
+for msg in st.session_state.messages:
+    render_message(msg["role"], msg["content"], ts=msg.get("time", ""),
+                   unsafe_allow_html=True, intermediate=msg.get("intermediate", False))
+
+if st.session_state.streaming:
+    render_streaming_area()
+
+if ext.process_paste_signal() or ext.process_delete_signal():
     st.rerun()
 
+ext.render_signal_inputs()
+prompt = ext.render_attachment_input_row(streaming=st.session_state.streaming)
+
+if prompt:
+    ext.handle_user_submit(prompt, start_task=start_agent_task)