import os, sys, subprocess from urllib.request import urlopen from urllib.parse import quote if sys.stdout is None: sys.stdout = open(os.devnull, "w") if sys.stderr is None: sys.stderr = open(os.devnull, "w") try: sys.stdout.reconfigure(errors='replace') except: pass try: sys.stderr.reconfigure(errors='replace') except: pass script_dir = os.path.dirname(__file__) sys.path.append(os.path.abspath(os.path.join(script_dir, '..'))) sys.path.append(os.path.abspath(script_dir)) import streamlit as st import time, json, re, threading, queue from agentmain import GeneraticAgent import chatapp_common # activate /continue command (monkey patches GeneraticAgent) from continue_cmd import handle_frontend_command, reset_conversation, list_sessions, extract_ui_messages st.set_page_config(page_title="Cowork", layout="wide") @st.cache_resource def init(): agent = GeneraticAgent() if agent.llmclient is None: st.error("⚠️ 사용 가능한 LLM 인터페이스가 구성되지 않았습니다. mykey.py를 설정하세요.") st.stop() else: threading.Thread(target=agent.run, daemon=True).start() return agent agent = init() st.title("🖥️ Cowork") if 'autonomous_enabled' not in st.session_state: st.session_state.autonomous_enabled = False @st.fragment def render_sidebar(): current_idx = agent.llm_no st.caption(f"LLM Core: {current_idx}: {agent.get_llm_name()}", help="백업 링크 전환하려면 클릭하세요") last_reply_time = st.session_state.get('last_reply_time', 0) if last_reply_time > 0: st.caption(f"대기 시간：{int(time.time()) - last_reply_time}초", help="30분 이상 응답이 없으면 시스템이 자동으로 작업을 시작합니다") if st.button("백업 링크 전환"): agent.next_llm(); st.rerun(scope="fragment") if st.button("작업 강제 중지"): agent.abort(); st.toast("중지 신호가 전송되었습니다"); st.rerun() if st.button("도구 재주입"): agent.llmclient.last_tools = '' try: hist_path = os.path.join(script_dir, '..', 'assets', 'tool_usable_history.json') with open(hist_path, 'r', encoding='utf-8') as f: tool_hist = json.load(f) agent.llmclient.backend.history.extend(tool_hist) st.toast(f"도구가 재주입되었습니다. {len(tool_hist)}개의 예시 기록이 추가되었습니다") except Exception as e: st.toast(f"도구 예시 주입 실패: {e}") if st.button("🐱 데스크톱 펫"): kwargs = {'creationflags': 0x08} if sys.platform == 'win32' else {} pet_script = os.path.join(script_dir, 'desktop_pet_v2.pyw') if not os.path.exists(pet_script): pet_script = os.path.join(script_dir, 'desktop_pet.pyw') subprocess.Popen([sys.executable, pet_script], **kwargs) def _pet_req(q): def _do(): try: urlopen(f'http://127.0.0.1:41983/?{q}', timeout=2) except Exception: pass threading.Thread(target=_do, daemon=True).start() agent._pet_req = _pet_req if not hasattr(agent, '_turn_end_hooks'): agent._turn_end_hooks = {} def _pet_hook(ctx): parts = [f"Turn {ctx.get('turn','?')}"] if ctx.get('summary'): parts.append(ctx['summary']) if ctx.get('exit_reason'): parts.append('작업 완료') _pet_req(f'msg={quote(chr(10).join(parts))}') if ctx.get('exit_reason'): _pet_req('state=idle') agent._turn_end_hooks['pet'] = _pet_hook st.toast("데스크톱 펫이 시작되었습니다") st.divider() if st.button("대기 중 자율 행동 시작"): st.session_state.last_reply_time = int(time.time()) - 1800 st.toast("마지막 응답 시간을 1800초 전으로 설정했습니다"); st.rerun() if st.session_state.autonomous_enabled: if st.button("⏸️ 자율 행동 금지"): st.session_state.autonomous_enabled = False st.toast("⏸️ 자율 행동이 금지되었습니다"); st.rerun() st.caption("🟢 자율 행동이 실행 중입니다. 30분 후 자동으로 진행됩니다") else: if st.button("▶️ 자율 행동 허용", type="primary"): st.session_state.autonomous_enabled = True st.toast("✅ 자율 행동이 허용되었습니다"); st.rerun() st.caption("🔴 자율 행동이 중지되었습니다") with st.sidebar: render_sidebar() def fold_turns(text): """Return list of segments: [{'type':'text','content':...}, {'type':'fold','title':...,'content':...}]""" # 먼저 4개 이상의 백틱 블록을 플레이스홀더로 교체하여 하위 에이전트 중첩 LLM Running이 잘못 잘리는 것을 방지합니다. _ph = [] safe = re.sub(r'`{4,}.*?`{4,}', lambda m: (_ph.append(m.group(0)), f'\x00PH{len(_ph)-1}\x00')[1], text, flags=re.DOTALL) parts = re.split(r'(\**LLM Running \(Turn \d+\) \.\.\.\*\**)', safe) parts = [re.sub(r'\x00PH(\d+)\x00', lambda m: _ph[int(m.group(1))], p) for p in parts] if len(parts) < 4: return [{'type': 'text', 'content': text}] segments = [] if parts[0].strip(): segments.append({'type': 'text', 'content': parts[0]}) turns = [] for i in range(1, len(parts), 2): marker = parts[i] content = parts[i+1] if i+1 < len(parts) else '' turns.append((marker, content)) for idx, (marker, content) in enumerate(turns): if idx < len(turns) - 1: _c = re.sub(r'`{3,}.*?`{3,}|.*?', '', content, flags=re.DOTALL) matches = re.findall(r'\s*((?:(?!).)*?)\s*', _c, re.DOTALL) if matches: title = matches[0].strip() title = title.split('\n')[0] if len(title) > 50: title = title[:50] + '...' else: title = marker.strip('*') segments.append({'type': 'fold', 'title': title, 'content': content}) else: segments.append({'type': 'text', 'content': marker + content}) return segments def render_segments(segments, suffix=''): # 전체 다시 그리기: 호출자가 slot.container()로 감싸서 DOM 경로가 안정적이고 rerun 간 정렬이 보장되도록 함 ("회색 잔상" 제거). # heartbeat 대기 시 segments 변경 없음 → Streamlit 백엔드 diff 변화 없음 → 프론트엔드 깜빡임 제로; # 단 container/markdown 자체는 API 호출이므로 StopException은 여전히 발생함 (abort는 정상 작동). for seg in segments: if seg['type'] == 'fold': with st.expander(seg['title'], expanded=False): st.markdown(seg['content']) else: st.markdown(seg['content'] + suffix) def agent_backend_stream(prompt): display_queue = agent.put_task(prompt, source="user") response = '' try: while True: try: item = display_queue.get(timeout=1) except queue.Empty: yield response # heartbeat: let outer st.markdown() run → Streamlit checks StopException continue if 'next' in item: response = item['next']; yield response if 'done' in item: yield item['done']; break finally: agent.abort() if "messages" not in st.session_state: st.session_state.messages = [] for msg in st.session_state.messages: with st.chat_message(msg["role"]): # slot=st.empty() + with slot.container(): ... 외부 껍질을 사용하여 DOM 경로와 스트리밍 렌더링이 완전히 일치하고 rerun 간 정렬됨 slot = st.empty() with slot.container(): if msg["role"] == "assistant": render_segments(fold_turns(msg["content"])) else: st.markdown(msg["content"]) # Scroll-height ghost fix: during streaming, expander open/close mid-animation can leave # phantom height → scrollbar long but can't scroll to bottom. Periodically detect & reflow. try: from streamlit import iframe as _st_iframe # 1.56+ _embed_html = lambda html, **kw: _st_iframe(html, **{k: max(v, 1) if isinstance(v, int) else v for k, v in kw.items()}) except (ImportError, AttributeError): from streamlit.components.v1 import html as _embed_html # ≤1.55 _js_scroll_fix = ("!function(){var p=window.parent;if(p.__sfx)return;p.__sfx=1;" "var d=p.document;setInterval(function(){" "var m=d.querySelector('section.main');if(!m)return;" "var b=m.querySelector('.block-container');if(!b)return;" "if(m.scrollHeight>b.scrollHeight+150){" "m.style.overflow='hidden';void m.offsetHeight;m.style.overflow=''}" "},3000)}()") # IME composition fix (macOS only) - prevents Enter from submitting during CJK input _js_ime_fix = ("" if os.name == 'nt' else "!function(){if(window.parent.__imeFix)return;window.parent.__imeFix=1;" "var d=window.parent.document,c=0;" "d.addEventListener('compositionstart',()=>c=1,!0);" "d.addEventListener('compositionend',()=>c=0,!0);" "function f(){d.querySelectorAll('textarea[data-testid=stChatInputTextArea]')" ".forEach(t=>{t.__imeFix||(t.__imeFix=1,t.addEventListener('keydown',e=>{" "e.key==='Enter'&&!e.shiftKey&&(e.isComposing||c||e.keyCode===229)&&" "(e.stopImmediatePropagation(),e.preventDefault())},!0))})}" "f();new MutationObserver(f).observe(d.body,{childList:1,subtree:1})}()") _embed_html(f'', height=0) if prompt := st.chat_input("any task?"): ts = time.strftime("%Y-%m-%d %H:%M:%S") cmd = (prompt or "").strip() def _reset_and_rerun(): st.session_state.streaming = False st.session_state.stopping = False st.session_state.display_queue = None st.session_state.partial_response = "" st.session_state.reply_ts = "" st.session_state.current_prompt = "" st.session_state.last_reply_time = int(time.time()) st.rerun() if cmd == "/new": st.session_state.messages = [{"role": "assistant", "content": reset_conversation(agent), "time": ts}] _reset_and_rerun() if cmd.startswith("/continue"): m = re.match(r'/continue\s+(\d+)\s*$', cmd.strip()) sessions = list_sessions(exclude_pid=os.getpid()) if m else [] idx = int(m.group(1)) - 1 if m else -1 # Resolve target path BEFORE handle (which snapshots current log, shifting indices). target = sessions[idx][0] if 0 <= idx < len(sessions) else None result = handle_frontend_command(agent, cmd) history = extract_ui_messages(target) if target and result.startswith('✅') else None tail = [{"role": "assistant", "content": result, "time": ts}] if history: st.session_state.messages = history + tail else: st.session_state.messages = list(st.session_state.messages) + \ [{"role": "user", "content": cmd, "time": ts}] + tail _reset_and_rerun() st.session_state.messages.append({"role": "user", "content": prompt}) if hasattr(agent, '_pet_req') and not prompt.startswith('/'): agent._pet_req('state=walk') with st.chat_message("user"): st.markdown(prompt) with st.chat_message("assistant"): frozen = 0; live = st.empty(); response = '' CURSOR = ' ▌' for response in agent_backend_stream(prompt): segs = fold_turns(response) n_done = max(0, len(segs) - 1) while frozen < n_done: with live.container(): render_segments([segs[frozen]]) live = st.empty(); frozen += 1 with live.container(): render_segments([segs[-1]], suffix=CURSOR) # 라이브 영역 segs = fold_turns(response) for i in range(frozen, len(segs)): with live.container(): render_segments([segs[i]]) if i < len(segs) - 1: live = st.empty() st.session_state.messages.append({"role": "assistant", "content": response}) st.session_state.last_reply_time = int(time.time()) if st.session_state.autonomous_enabled: st.markdown(f"""""", unsafe_allow_html=True)