tagword-codeagent 1.1.6__py3-none-any.whl

codeagent/cli/main.py

@@ -0,0 +1,979 @@
+from __future__ import annotations
+
+import json
+import os
+import subprocess
+import sys
+from datetime import datetime
+
+from seed_tools import setup_builtin_tools
+
+from seed.core.persistence import ensure_session_dir, list_sessions, save_session
+from seed.core.routing import find_commands, get_all_commands, get_command
+
+
+def cmd_restart(args):
+    tgt = getattr(args, "restart_cmd", "")
+    if tgt != "serve":
+        print("Usage: codeagent restart serve [--host ... --port ...]", file=sys.stderr)
+        sys.exit(2)
+
+    port = int(getattr(args, "port", 8765))
+    timeout_sec = float(getattr(args, "timeout_sec", 5.0))
+
+    from codeagent.core.process_ports import stop_listeners_on_port
+
+    stopped = stop_listeners_on_port(
+        port,
+        timeout_sec=timeout_sec,
+        exclude_pid=os.getpid(),
+        log=print,
+    )
+    if not stopped and os.name != "nt":
+        import shutil
+
+        if not shutil.which("lsof") and not shutil.which("ss") and not shutil.which("fuser"):
+            print(
+                "restart serve: install lsof (macOS) or iproute2/ss (Linux) "
+                "to stop the previous listener automatically.",
+                file=sys.stderr,
+            )
+
+    # Start server in current process (foreground).
+    cmd_serve(args)
+
+def cmd_run(args):
+    """Handle run subcommand."""
+    prompt = getattr(args, 'prompt', '')
+    if not prompt:
+        print("Usage: codeagent run [options] <prompt>")
+        sys.exit(1)
+
+    ensure_session_dir()
+    session_id = args.save if args.save else f"agent-{datetime.now().strftime('%Y%m%d-%H%M%S')}"
+
+    matched = find_commands(prompt, limit=5)
+    matches = [c for c in matched if hasattr(c, 'name') and c]
+
+    print("=== CodeAgent Run Result")
+    print(f"Session ID: {session_id}")
+    print(f"Prompt: {prompt}")
+    print(f"Matched commands: {len(matches)}")
+
+    for cmd_entry in matched:
+        name = cmd_entry.name if hasattr(cmd_entry, 'name') else repr(cmd_entry)
+        cmd = get_command(name)
+        if cmd:
+            print(f" - {name}: {cmd.description}")
+
+    if args.save:
+        save_session(session_id, [prompt], 0, 0)
+        from seed.core.llm_sess import agent_sessions_dir
+
+        print(f"Session saved to: {agent_sessions_dir() / f'{session_id}.json'}")
+
+def cmd_route(args):
+    """Handle route subcommand."""
+    prompt = getattr(args, 'prompt', '')
+    limit = getattr(args, 'limit', 20)
+
+    print("=== CodeAgent Route Result")
+    print(f"Query: {prompt}")
+
+    matched = find_commands(prompt, limit=limit)
+    print(f"Matches found: {len(matched)}")
+
+    for cmd in matched[:limit]:
+        name = cmd.name if hasattr(cmd, 'name') else repr(cmd)
+        desc = cmd.description if hasattr(cmd, 'description') else ""
+        print(f" - {name}: {desc}")
+
+def cmd_commands(args):
+    """Handle commands subcommand."""
+    all_cmds = get_all_commands()
+    limit = getattr(args, 'limit', 50)
+    result_cmds = all_cmds[:limit]
+
+    print("=== CodeAgent Commands")
+    print(f"Total commands: {len(result_cmds)}")
+
+    for cmd in result_cmds:
+        cat = "default"
+        if hasattr(cmd, 'category') and cmd.category:
+            cat = cmd.category
+        if hasattr(cmd, 'name'):
+            print(f" [{cat}] {cmd.name}")
+            if hasattr(cmd, 'description') and cmd.description:
+                print(f"        {cmd.description}")
+
+def cmd_tools(args):
+    """Handle tools subcommand."""
+    registry, _executor = setup_builtin_tools()
+    tools = registry.list_all()
+    limit = getattr(args, 'limit', 50)
+    result_tools = tools[:limit]
+
+    print("=== CodeAgent Tools")
+    print(f"Total tools: {len(result_tools)}")
+
+    for tool in result_tools:
+        print(f" - [basic]: {tool.name} - {tool.description}")
+
+def cmd_summary(args):
+    """Handle summary subcommand."""
+    session_id = getattr(args, 'session', None)
+
+    print("=== CodeAgent Session Summary")
+
+    from seed.core.llm_sess import agent_sessions_dir, list_stored_sessions_meta
+
+    if session_id:
+        from codeagent.core import env as ca_env
+
+        aid = ca_env.default_agent_id()
+        file_path = agent_sessions_dir(aid) / f"{session_id}.json"
+
+        if not file_path.is_file():
+            print(f"Warning: Session file not found: {file_path}")
+            rows = list_stored_sessions_meta(limit=50, agent_id=aid)
+            print(f"Available chat sessions: {len(rows)}")
+        else:
+            print(f"Session: {session_id}")
+            print(f"File: {file_path}")
+    else:
+        sessions = list_sessions()
+        if not sessions:
+            print("No saved sessions (legacy QueryEngine store).")
+        print(f"Chat sessions directory: {agent_sessions_dir()}")
+        rows = list_stored_sessions_meta(limit=200)
+        print(f"Stored chat sessions: {len(rows)}")
+        print(f"Legacy persistence sessions: {len(sessions)}")
+
+
+
+"""
+CodeAgent CLI - Main entry point
+Provides command routing, tool execution, and session management.
+"""
+import argparse  # noqa: E402
+from typing import Any  # noqa: E402
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+
+def _win_no_window_kwargs() -> dict[str, Any]:
+    """Suppress the black cmd window when this CLI helper runs under a windowed
+    parent (e.g. ``CodeAgentTray.exe``). No-op on POSIX."""
+    if os.name != "nt":
+        return {}
+    si = subprocess.STARTUPINFO()  # type: ignore[attr-defined]
+    si.dwFlags |= subprocess.STARTF_USESHOWWINDOW  # type: ignore[attr-defined]
+    si.wShowWindow = 0  # SW_HIDE
+    return {"startupinfo": si, "creationflags": 0x08000000}  # CREATE_NO_WINDOW
+
+def main():
+    """Main CLI entry point."""
+
+    from codeagent.core.bootstrap import bootstrap_codeagent_runtime
+
+    bootstrap_codeagent_runtime()
+
+    parser = argparse.ArgumentParser(
+        prog='codeagent',
+        description='CodeAgent: Autonomous agent for task execution with command routing and session management',
+        formatter_class=argparse.RawDescriptionHelpFormatter
+    )
+
+    parser.add_argument('--version', action='version', version='CodeAgent v1.1.0')
+    subparsers = parser.add_subparsers(dest='command', help='Available commands')
+
+    # Run command
+    run_parser = subparsers.add_parser('run', help='Run agent with a prompt')
+    run_parser.add_argument('prompt', nargs='?', help='Prompt to execute', default='')
+    run_parser.add_argument('--max-turns', type=int, default=8, help='Maximum turns')
+    run_parser.add_argument('--budget', type=int, default=2000, help='Token budget')
+    run_parser.add_argument('--save', '-s', help='Save session ID')
+
+    # Route command
+    route_parser = subparsers.add_parser('route', help='Route prompt to commands')
+    route_parser.add_argument('prompt', help='Prompt to route')
+    route_parser.add_argument('--limit', type=int, default=20, help='Limit results')
+
+    # Commands command
+    cmd_parser = subparsers.add_parser('commands', help='List all commands')
+    cmd_parser.add_argument('--query', '-q', help='Filter by query', default=None)
+    cmd_parser.add_argument('--limit', '-n', type=int, default=50, help='Limit output')
+    cmd_parser.add_argument('--category', '-c', help='Filter by category', default=None)
+
+    # Tools command
+    tools_parser = subparsers.add_parser('tools', help='List all tools')
+    tools_parser.add_argument('--query', '-i', help='Filter by query')
+    tools_parser.add_argument('--limit', '-n', type=int, default=50, help='Limit output')
+    tools_parser.add_argument('--simple', '-s', action='store_true', help='Basic tools only')
+
+    # Summary command
+    summary_parser = subparsers.add_parser('summary', help='Show session summary')
+    summary_parser.add_argument('--session', '-S', help='Session ID')
+
+    # Session command
+    sess_parser = subparsers.add_parser('session', help='Manage chat sessions (agents/<id>/sessions/)')
+    sess_parser.add_argument(
+        'action',
+        choices=['list', 'migrate', 'delete', 'audit-list', 'audit-show', 'help'],
+        help='Action',
+    )
+    sess_parser.add_argument(
+        '--seq',
+        type=int,
+        default=None,
+        help='With audit-show: snapshot sequence number from audit-list',
+    )
+    sess_parser.add_argument(
+        '--project-id',
+        default=None,
+        help='Project id when session lives under projects-data/<id>/sessions/',
+    )
+    sess_parser.add_argument('session_id', nargs='?', help='Session ID (for delete)')
+    sess_parser.add_argument(
+        '--agent-id',
+        default=None,
+        help='Agent id (default: CODEAGENT_AGENT_ID or default)',
+    )
+    sess_parser.add_argument(
+        '--dry-run',
+        action='store_true',
+        help='With migrate: print planned moves without writing',
+    )
+
+    # Interactive chat (command routing only until LLM runtime is wired)
+    chat_parser = subparsers.add_parser(
+        'chat',
+        help='Interactive REPL: use --llm for full tool loop; else command routing only',
+    )
+    chat_parser.add_argument('--limit', '-n', type=int, default=8, help='Max matches per line (without --llm)')
+    chat_parser.add_argument(
+        '--session',
+        '-S',
+        help='Chat session id: resume/save under agents/<agent>/sessions (override SEED_AGENT_SESSIONS_DIR or SEED_LLM_SESSIONS_DIR)',
+    )
+    chat_parser.add_argument(
+        '--llm',
+        action='store_true',
+        help='Use SEED_LLM_BASEURL / SEED_LLM_MODEL (aliases CODEAGENT_*) and builtin tools; optional config/env',
+    )
+    chat_parser.add_argument('--max-tool-rounds', type=int, default=16, help='Max LLM/tool cycles per user line')
+    chat_parser.add_argument(
+        '--image',
+        action='append',
+        metavar='PATH',
+        help='Attach image/document file (repeatable); included with next user line',
+    )
+    chat_parser.add_argument(
+        '--image-dir',
+        metavar='PATH',
+        help='Stage all images under directory (or reference with --image-dir-ref)',
+    )
+    chat_parser.add_argument(
+        '--image-pattern',
+        help='Glob patterns with --image-dir, e.g. "*.png,*.jpg"',
+    )
+    chat_parser.add_argument(
+        '--image-dir-ref',
+        action='store_true',
+        help='Write [image_dir:path] reference instead of copying files',
+    )
+    chat_parser.add_argument(
+        '--vision-llm',
+        metavar='PRESET_ID',
+        help='Vision preset for vision_analyze when sending images (or CODEAGENT_VISION_PRESET_ID)',
+    )
+    chat_parser.add_argument(
+        '--image-gen-llm',
+        metavar='PRESET_ID',
+        help='Image generation preset for image_generate (or CODEAGENT_IMAGE_GEN_PRESET_ID)',
+    )
+    chat_parser.add_argument(
+        '--audio-llm',
+        metavar='PRESET_ID',
+        help='Audio transcription preset for audio_transcribe (or CODEAGENT_AUDIO_PRESET_ID)',
+    )
+    chat_parser.add_argument(
+        '--music-llm',
+        metavar='PRESET_ID',
+        help='Music generation preset for music_generate (or CODEAGENT_MUSIC_GEN_PRESET_ID)',
+    )
+    chat_parser.add_argument(
+        '--video-gen-llm',
+        metavar='PRESET_ID',
+        help='Video generation preset for video_generate (or CODEAGENT_VIDEO_GEN_PRESET_ID)',
+    )
+
+    serve_parser = subparsers.add_parser(
+        'serve',
+        help=(
+            'HTTP: GET /, POST /api/chat, WS /ws; optional Web UI token (codeagent webui-token init); '
+            'webhooks. 默认 --host 0.0.0.0：启动时会打印本机与局域网 URL，便于其它设备打开配置或填写 Webhook。'
+        ),
+    )
+    serve_parser.add_argument(
+        '--host',
+        default='0.0.0.0',
+        help='Bind address（0.0.0.0 表示局域网可访问）',
+    )
+    serve_parser.add_argument('--port', type=int, default=8765, help='Port')
+
+    restart_parser = subparsers.add_parser(
+        'restart',
+        help='Restart helpers (e.g. restart serve). Tries to stop existing listener first.',
+    )
+    restart_sub = restart_parser.add_subparsers(dest='restart_cmd', required=True, metavar='target')
+    restart_serve = restart_sub.add_parser('serve', help='Restart HTTP server (kill listener on port, then run serve)')
+    restart_serve.add_argument('--host', default='0.0.0.0', help='Bind address')
+    restart_serve.add_argument('--port', type=int, default=8765, help='Port')
+    restart_serve.add_argument(
+        '--timeout-sec',
+        type=float,
+        default=5.0,
+        help='How long to wait for port to be released after killing (seconds)',
+    )
+
+    cfg_parser = subparsers.add_parser('config', help='Manage Markdown config under config/')
+    cfg_sub = cfg_parser.add_subparsers(dest='cfg_cmd', help='Config action')
+    cfg_sub.add_parser('init', help='Create default agent.md, identity.md, soul.md, tools.md, user.md')
+
+    wt_parser = subparsers.add_parser(
+        "webui-token",
+        help="Web UI: create/show/reset access token (local CLI only; file config/webui.token)",
+    )
+    wt_sub = wt_parser.add_subparsers(dest="wt_action", required=True, metavar="action")
+    wt_sub.add_parser("init", help="Create token file if missing (fails if file already exists)")
+    wt_sub.add_parser("reset", help="Write a new token (invalidates existing browser sessions)")
+    wt_sub.add_parser("show", help="Print token from file (environment CODEAGENT_WEBUI_TOKEN is not printed)")
+
+    if len(sys.argv) < 2:
+        parser.print_help()
+        sys.exit(0)
+
+    args = parser.parse_args()
+
+    if not args.command:
+        parser.print_help()
+        sys.exit(0)
+
+    cmd = getattr(args, 'command', None)
+    if cmd == 'run':
+        cmd_run(args)
+    elif cmd == 'route':
+        cmd_route(args)
+    elif cmd == 'commands':
+        cmd_commands(args)
+    elif cmd == 'tools':
+        cmd_tools(args)
+    elif cmd == 'summary':
+        cmd_summary(args)
+    elif cmd == 'session':
+        cmd_session(args)
+    elif cmd == 'chat':
+        cmd_chat(args)
+    elif cmd == 'serve':
+        cmd_serve(args)
+    elif cmd == 'restart':
+        cmd_restart(args)
+    elif cmd == 'config':
+        cmd_config(args)
+    elif cmd == "webui-token":
+        cmd_webui_token(args)
+    else:
+        parser.print_help()
+
+
+
+
+import asyncio  # noqa: E402
+import contextlib  # noqa: E402
+import os  # noqa: E402
+import sys  # noqa: E402
+
+
+def cmd_chat(args):
+    """Interactive loop: routing-only or LLM + tools."""
+    if getattr(args, 'llm', False):
+        _cmd_chat_llm(args)
+        return
+
+    limit = getattr(args, 'limit', 8)
+    sess = getattr(args, 'session', None)
+    print("CodeAgent chat — 每行输入会做命令匹配（无 --llm）；exit / quit 退出。")
+    if sess:
+        print(f"Session label: {sess}")
+    while True:
+        try:
+            line = input("> ").strip()
+        except (EOFError, KeyboardInterrupt):
+            print()
+            break
+        if not line:
+            continue
+        if line.lower() in ('exit', 'quit', 'q'):
+            break
+        matched = find_commands(line, limit=limit)
+        if not matched:
+            print("  (无匹配命令)")
+            continue
+        for c in matched:
+            desc = c.description if getattr(c, 'description', None) else ''
+            print(f"  - {c.name}: {desc}")
+
+
+def _cli_audio_preset_id(args) -> str:
+    from codeagent.core import env as ca_env
+
+    pid = (getattr(args, 'audio_llm', None) or '').strip()
+    if pid:
+        return pid
+    return ca_env.pick_default('', 'CODEAGENT_AUDIO_PRESET_ID').strip()
+
+
+def _cli_image_gen_preset_id(args) -> str:
+    from codeagent.core import env as ca_env
+
+    pid = (getattr(args, 'image_gen_llm', None) or '').strip()
+    if pid:
+        return pid
+    return ca_env.pick_default('', 'CODEAGENT_IMAGE_GEN_PRESET_ID').strip()
+
+
+def _cli_music_preset_id(args) -> str:
+    from codeagent.core import env as ca_env
+
+    pid = (getattr(args, 'music_llm', None) or '').strip()
+    if pid:
+        return pid
+    return ca_env.pick_default('', 'CODEAGENT_MUSIC_GEN_PRESET_ID').strip()
+
+
+def _cli_video_gen_preset_id(args) -> str:
+    from codeagent.core import env as ca_env
+
+    pid = (getattr(args, 'video_gen_llm', None) or '').strip()
+    if pid:
+        return pid
+    return ca_env.pick_default('', 'CODEAGENT_VIDEO_GEN_PRESET_ID').strip()
+
+
+def _cli_vision_preset_id(args) -> str:
+    from codeagent.core import env as ca_env
+
+    vid = (getattr(args, 'vision_llm', None) or '').strip()
+    if vid:
+        return vid
+    return ca_env.pick_default('', 'CODEAGENT_VISION_PRESET_ID').strip()
+
+
+def _cli_stage_path_attachment(path: str, *, agent_id: str, session_id: str):
+    from pathlib import Path
+
+    import mimetypes
+
+    from codeagent.core.attachments import save_attachment
+
+    p = Path(path).expanduser()
+    if not p.is_file():
+        print(f'  [warn] not a file: {path}')
+        return None
+    try:
+        raw = p.read_bytes()
+        mime = mimetypes.guess_type(str(p))[0] or ''
+        return save_attachment(
+            agent_id=agent_id,
+            session_id=session_id,
+            raw_bytes=raw,
+            filename=p.name,
+            mime=mime,
+        )
+    except Exception as e:
+        print(f'  [warn] attach failed {path}: {e}')
+        return None
+
+
+def _cli_startup_attachments(args, *, agent_id: str, session_id: str, project_root):
+    from pathlib import Path
+
+    from codeagent.core import env as ca_env
+    from codeagent.core.attachments import scan_image_directory
+
+    metas = []
+    image_dir_tag = ''
+    for path in getattr(args, 'image', None) or []:
+        meta = _cli_stage_path_attachment(path, agent_id=agent_id, session_id=session_id)
+        if meta:
+            metas.append(meta)
+            print(f'  [attach] {meta.filename} → {meta.id}')
+
+    image_dir = (getattr(args, 'image_dir', None) or '').strip()
+    if image_dir:
+        if getattr(args, 'image_dir_ref', False):
+            rel = image_dir.replace('\\', '/').lstrip('/')
+            mx = ca_env.pick_int(32, 'CODEAGENT_ATTACHMENTS_DIR_MAX_FILES')
+            image_dir_tag = f'[image_dir:{rel} max={mx}]'
+            print(f'  [image-dir-ref] {image_dir_tag}')
+        else:
+            pattern = (getattr(args, 'image_pattern', None) or '').strip() or None
+            try:
+                paths, truncated = scan_image_directory(
+                    Path(project_root),
+                    image_dir,
+                    pattern=pattern,
+                )
+            except ValueError as e:
+                print(f'  [warn] image-dir: {e}')
+                paths, truncated = [], False
+            for p in paths:
+                meta = _cli_stage_path_attachment(str(p), agent_id=agent_id, session_id=session_id)
+                if meta:
+                    metas.append(meta)
+            if metas:
+                print(f'  [image-dir] staged {len(metas)} file(s)' + (' (truncated)' if truncated else ''))
+            elif not truncated:
+                print(f'  [warn] no images found in {image_dir}')
+    return metas, image_dir_tag
+
+
+def _cli_handle_slash_attach(line: str, *, agent_id: str, session_id: str, project_root):
+    """Return (handled, attachments, image_dir_tag, message_text)."""
+    from pathlib import Path
+
+    from codeagent.core import env as ca_env
+    from codeagent.core.attachments import scan_image_directory
+
+    low = line.strip().lower()
+    if low in ('/clear-vision', '/clear_vision'):
+        return 'clear_vision', [], '', ''
+    if line.startswith('/attach-dir '):
+        path = line[len('/attach-dir '):].strip()
+        if not path:
+            print('  usage: /attach-dir <path>')
+            return True, [], '', ''
+        mx = ca_env.pick_int(32, 'CODEAGENT_ATTACHMENTS_DIR_MAX_FILES')
+        tag = f'[image_dir:{path.replace(chr(92), "/").lstrip("/")} max={mx}]'
+        print(f'  [image-dir-ref] {tag}')
+        return True, [], tag, ''
+    if line.startswith('/attach '):
+        path = line[len('/attach '):].strip()
+        if not path:
+            print('  usage: /attach <path>')
+            return True, [], '', ''
+        meta = _cli_stage_path_attachment(path, agent_id=agent_id, session_id=session_id)
+        if meta:
+            print(f'  [attach] {meta.filename} → {meta.id}')
+            return True, [meta], '', ''
+        return True, [], '', ''
+    return False, [], '', line
+
+
+def _cmd_chat_llm(args):
+
+    from seed_tools import setup_builtin_tools
+
+    from seed.core.agent_context import (
+        set_active_agent_id,
+        set_active_audio_preset,
+        set_active_image_gen_preset,
+        set_active_llm_session,
+        set_active_vision_preset,
+        set_active_music_preset,
+        set_active_video_gen_preset,
+    )
+    from seed.core.agent_runtime import (
+        build_api_projection_messages,
+        maybe_compact_context_messages,
+        merge_llm_tail_into_full,
+        persist_compact_summary,
+        run_llm_tool_loop,
+        strip_ephemeral_message_fields,
+    )
+    from codeagent.runtime.prompt_enrichment import (
+        build_skills_suffix,
+        fresh_system_prompt,
+        get_cached_system_prompt,
+        record_chat_turn_diary,
+    )
+    from seed.core.llm_exec import LLMError
+    from seed.core.llm_presets import llm_executor_from_resolved, resolve_preset
+    from seed.core.llm_sess import (
+        load_or_create_chat_session,
+        merge_fresh_system,
+        persist_chat_session,
+    )
+    from seed.core.mem_bridge import finalize_episodic_for_llm
+    from seed.integrations.session_title import maybe_llm_refresh_session_title
+
+    name = getattr(args, 'session', None) or 'cli-chat'
+    max_rounds = getattr(args, 'max_tool_rounds', 16) or 16
+
+    registry, executor = setup_builtin_tools()
+    llm = llm_executor_from_resolved(resolve_preset(None))
+    from seed.core.config_plane import project_root as _project_root
+
+    project_root = _project_root()
+    from codeagent.core import env as ca_env
+
+    agent_id = ca_env.default_agent_id()
+    vision_preset_id = _cli_vision_preset_id(args)
+    image_gen_preset_id = _cli_image_gen_preset_id(args)
+    audio_preset_id = _cli_audio_preset_id(args)
+    music_preset_id = _cli_music_preset_id(args)
+    video_gen_preset_id = _cli_video_gen_preset_id(args)
+    chat_sess = load_or_create_chat_session(name, agent_id)
+    fresh_sys = get_cached_system_prompt(chat_sess, agent_id=agent_id)
+    chat_sess.messages[:] = merge_fresh_system(chat_sess.messages, fresh_sys)
+
+    pending_attachments, pending_dir_tag = _cli_startup_attachments(
+        args,
+        agent_id=agent_id,
+        session_id=name,
+        project_root=project_root,
+    )
+
+    print(f"CodeAgent LLM chat — session «{name}» — {len(registry.list_all())} tools — exit / quit 结束。")
+    print(f"  API: {llm.baseURL}  model: {llm.model}")
+    if vision_preset_id:
+        print(f"  Vision preset: {vision_preset_id}")
+    if image_gen_preset_id:
+        print(f"  Image gen preset: {image_gen_preset_id}")
+    if audio_preset_id:
+        print(f"  Audio preset: {audio_preset_id}")
+    if music_preset_id:
+        print(f"  Music preset: {music_preset_id}")
+    if video_gen_preset_id:
+        print(f"  Video gen preset: {video_gen_preset_id}")
+    print("  命令: /attach <path>  /attach-dir <path>  /clear-vision")
+    while True:
+        try:
+            line = input("> ").strip()
+        except (EOFError, KeyboardInterrupt):
+            print()
+            break
+        if not line and not pending_attachments and not pending_dir_tag:
+            continue
+        if line.lower() in ('exit', 'quit', 'q'):
+            break
+
+        slash_atts: list = []
+        slash_dir_tag = ''
+        message_text = line
+        if line.startswith('/'):
+            handled, slash_atts, slash_dir_tag, message_text = _cli_handle_slash_attach(
+                line,
+                agent_id=agent_id,
+                session_id=name,
+                project_root=project_root,
+            )
+            if handled == 'clear_vision':
+                if isinstance(chat_sess.metadata, dict):
+                    chat_sess.metadata.pop('vision_context', None)
+                print('  视觉上下文已清除')
+                continue
+            if handled is True and not slash_atts and not slash_dir_tag:
+                continue
+        else:
+            handled = False
+
+        if slash_atts:
+            pending_attachments.extend(slash_atts)
+        if slash_dir_tag:
+            pending_dir_tag = slash_dir_tag
+
+        from codeagent.core.attachments import (
+            build_user_message,
+            message_has_audio_attachments,
+            message_has_image_attachments,
+            message_has_video_attachments,
+        )
+        from codeagent.core.audio_models import preset_supports_audio_id
+        from codeagent.core.vision_models import preset_supports_vision_id
+
+        user_msg = build_user_message(
+            message_text,
+            pending_attachments,
+            image_dir_tag=pending_dir_tag,
+        )
+        from codeagent.core.image_understanding import image_attachment_allowed, video_attachment_allowed
+
+        if message_has_image_attachments(user_msg) or pending_dir_tag:
+            if not image_attachment_allowed(vision_preset_id or ""):
+                print(
+                    '[error] 发送图片需要 --vision-llm（supports_vision）'
+                    ' 或已配置 MiniMax MCP（understand_image）'
+                )
+                continue
+        if message_has_video_attachments(user_msg):
+            if not video_attachment_allowed(vision_preset_id or ""):
+                print('[error] 发送视频需要有效的 --vision-llm 或 CODEAGENT_VISION_PRESET_ID')
+                continue
+        if message_has_audio_attachments(user_msg):
+            if not audio_preset_id or not preset_supports_audio_id(audio_preset_id):
+                print('[error] 发送音频需要有效的 --audio-llm 或 CODEAGENT_AUDIO_PRESET_ID')
+                continue
+
+        pending_attachments = []
+        pending_dir_tag = ''
+
+        chat_sess.messages.append(user_msg)
+        user_text = str(user_msg.get('content') or '')
+        from codeagent.core.attachments import content_text_for_skills
+
+        _skills_suffix = build_skills_suffix(agent_id, user_text=content_text_for_skills(user_text))
+        api_msgs = build_api_projection_messages(
+            chat_sess.messages,
+            skills_suffix=_skills_suffix,
+        )
+        compact_result = maybe_compact_context_messages(api_msgs, llm)
+        persist_compact_summary(chat_sess.messages, compact_result)
+        strip_ephemeral_message_fields(api_msgs)
+        if not isinstance(chat_sess.metadata, dict):
+            chat_sess.metadata = {}
+        finalize_episodic_for_llm(
+            api_msgs,
+            chat_sess.metadata,
+            agent_id=agent_id,
+            session_id=name,
+            project_id=None,
+            compact_happened=compact_result is not None,
+        )
+        set_active_llm_session(f'{agent_id}::{name}')
+        set_active_agent_id(agent_id)
+        set_active_vision_preset(vision_preset_id or None)
+        set_active_image_gen_preset(image_gen_preset_id or None)
+        set_active_audio_preset(audio_preset_id or None)
+        set_active_music_preset(music_preset_id or None)
+        set_active_video_gen_preset(video_gen_preset_id or None)
+        try:
+            n_before = len(api_msgs)
+            reply, _, tools_used, _tool_trace, _loop_meta = asyncio.run(
+                run_llm_tool_loop(
+                    llm,
+                    executor,
+                    messages=api_msgs,
+                    registry=registry,
+                    max_tool_rounds=max_rounds,
+                )
+            )
+            tail = merge_llm_tail_into_full(chat_sess.messages, api_msgs, n_before)
+            print(reply)
+            if tools_used:
+                print(f"  [tools] {', '.join(tools_used)}")
+            with contextlib.suppress(Exception):
+                persist_chat_session(chat_sess)
+            with contextlib.suppress(Exception):
+                maybe_llm_refresh_session_title(llm, chat_sess)
+            with contextlib.suppress(Exception):
+                record_chat_turn_diary(
+                    agent_id,
+                    user_text=user_text,
+                    reply=reply or "",
+                    tools_used=tools_used,
+                )
+            if os.environ.get("CODEAGENT_MEMORY_LOG", "1").lower() not in ("0", "false", "no"):
+                try:
+                    from seed.core.mem_sys import MemorySystem
+
+                    mem = MemorySystem(base_path=project_root)
+                    mem.log_experience(
+                        task_id=f"chat-{datetime.now().isoformat()}",
+                        outcome=(reply or "")[:2000],
+                        tools_used=tools_used,
+                        session_id=name,
+                    )
+                except Exception:
+                    pass
+        except LLMError as e:
+            print(f"[LLM error] {e}")
+            chat_sess.messages.pop()
+        finally:
+            set_active_llm_session(None)
+            set_active_vision_preset(None)
+            set_active_image_gen_preset(None)
+            set_active_audio_preset(None)
+            set_active_agent_id(None)
+
+
+def cmd_webui_token(args):
+    import secrets
+
+    from codeagent.web.auth_impl import TOKEN_FILENAME
+
+    # cmd_webui_token can be invoked directly; ensure env file is loaded.
+    from codeagent.core.bootstrap import bootstrap_codeagent_runtime
+
+    bootstrap_codeagent_runtime()
+    from seed.core.config_plane import project_root as _project_root
+
+    root = _project_root()
+    cfg = root / "config"
+    cfg.mkdir(parents=True, exist_ok=True)
+    path = cfg / TOKEN_FILENAME
+    act = args.wt_action
+
+    if act == "show":
+        if not path.is_file():
+            print(f"(no token file) {path}")
+            if os.environ.get("CODEAGENT_WEBUI_TOKEN", "").strip():
+                print("CODEAGENT_WEBUI_TOKEN is set in the environment (value not shown).")
+            return
+        print(path)
+        print(path.read_text(encoding="utf-8").strip())
+        return
+
+    if act == "init":
+        if path.exists():
+            print(f"Already exists: {path}")
+            print("Use: codeagent webui-token reset")
+            sys.exit(1)
+        token = secrets.token_urlsafe(32)
+        path.write_text(token + "\n", encoding="utf-8")
+        with contextlib.suppress(OSError, NotImplementedError, AttributeError):
+            os.chmod(path, 0o600)
+        print(f"Written: {path}")
+        print(f"Token (save it): {token}")
+        return
+
+    if act == "reset":
+        token = secrets.token_urlsafe(32)
+        path.write_text(token + "\n", encoding="utf-8")
+        with contextlib.suppress(OSError, NotImplementedError, AttributeError):
+            os.chmod(path, 0o600)
+        print(f"Written: {path}")
+        print(f"New token: {token}")
+        return
+
+
+
+
+import sys  # noqa: E402
+
+
+def cmd_serve(args):
+    try:
+        from codeagent.server import main as server_main
+    except ImportError as e:
+        print(f"无法加载 HTTP 服务模块: {e}", file=sys.stderr)
+        print(
+            "请确认已安装 server 依赖: pip install 'codeagent'（默认已包含 starlette + uvicorn）",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+    try:
+        server_main(host=args.host, port=args.port)
+    except OSError as e:
+        print(f"服务启动失败（端口或地址不可用）: {e}", file=sys.stderr)
+        print(
+            f"可尝试更换端口: codeagent serve --host {args.host} --port {(args.port or 8765) + 1}",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+
+def cmd_serve_tray(args):
+    print("serve-tray has been removed.", file=sys.stderr)
+    sys.exit(1)
+
+
+def cmd_config(args):
+    from seed.core.config_plane import ensure_default_config_files, project_root
+
+    sub = getattr(args, 'cfg_cmd', None)
+    if sub == 'init':
+        root = project_root()
+        ensure_default_config_files(root)
+        ex = root / "config" / "env.example"
+        print(f"Default Markdown written under: {root / 'config'}")
+        print("Tip: optional plugin skills live in config/skills/ directory.")
+        if ex.is_file():
+            print(f"Context/LLM env template: {ex}  -> copy to config/env if you want repo-local settings.")
+        return
+    print("Usage: codeagent config init")
+
+
+def cmd_session(args):
+    """Handle session subcommand (Seed chat sessions on disk)."""
+    from codeagent.core import env as ca_env
+
+    action = getattr(args, 'action', 'list')
+    session_id = getattr(args, 'session_id', None)
+    agent_id = (getattr(args, 'agent_id', None) or ca_env.default_agent_id()).strip() or 'default'
+
+    if action == 'list':
+        from seed.core.llm_sess import list_stored_sessions_meta
+
+        rows = list_stored_sessions_meta(limit=200, agent_id=agent_id)
+        print(f"=== Chat sessions (agent={agent_id})")
+        print(f"Total: {len(rows)}")
+        for row in rows:
+            title = row.get('display_title') or row.get('session_id')
+            print(f" - {row.get('session_id')}: {title}")
+    elif action == 'migrate':
+        from seed.core.llm_sess import migrate_legacy_agent_sessions
+
+        stats = migrate_legacy_agent_sessions(
+            agent_id,
+            dry_run=bool(getattr(args, 'dry_run', False)),
+        )
+        print(json.dumps(stats, ensure_ascii=False, indent=2))
+    elif action == 'audit-list':
+        if not session_id:
+            print("Usage: codeagent session audit-list <session_id> [--agent-id ...] [--project-id ...]")
+            return
+        from seed.core.projection_audit import list_projection_audit_index
+
+        rows = list_projection_audit_index(
+            session_id,
+            agent_id=agent_id,
+            project_id=getattr(args, 'project_id', None),
+        )
+        if not rows:
+            print("No LLM projection audit snapshots (enable SEED_LLM_PROJECTION_AUDIT=1).")
+            return
+        print(f"=== LLM projection audit (agent={agent_id}, session={session_id})")
+        for row in rows:
+            print(
+                f" seq={row.get('seq')} kind={row.get('kind')} round={row.get('round')} "
+                f"bytes={row.get('body_bytes')} file={row.get('file')}"
+            )
+    elif action == 'audit-show':
+        if not session_id:
+            print("Usage: codeagent session audit-show <session_id> --seq N")
+            return
+        seq = getattr(args, 'seq', None)
+        if seq is None:
+            print("--seq is required for audit-show")
+            return
+        from seed.core.projection_audit import load_projection_audit_snapshot
+
+        snap = load_projection_audit_snapshot(
+            session_id,
+            seq=seq,
+            agent_id=agent_id,
+            project_id=getattr(args, 'project_id', None),
+        )
+        if not snap:
+            print(f"Audit snapshot not found: seq={seq}")
+            return
+        print(json.dumps(snap, ensure_ascii=False, indent=2))
+    elif action == 'help':
+        print("Actions: list, migrate, delete, audit-list, audit-show, help")
+        print("  codeagent session migrate [--agent-id default] [--dry-run]")
+        print("  codeagent session audit-list <session_id>")
+        print("  codeagent session audit-show <session_id> --seq N")
+        print("  Enable snapshots: SEED_LLM_PROJECTION_AUDIT=1 in config/env")
+    elif action == 'delete' and session_id:
+        from seed.core.llm_sess import delete_stored_session
+
+        if delete_stored_session(session_id, agent_id):
+            print(f"Session deleted: {session_id}")
+        else:
+            print(f"Session not found: {session_id}")
+    else:
+        print(f"Session action '{action}': invalid or missing arguments.")
+
+# Note: entry points are __main__.py / main.py / cli_pkg, not this file directly