EvoScientist 0.0.1.dev4__py3-none-any.whl → 0.1.0rc1__py3-none-any.whl

EvoScientist/cli.py

@@ -10,216 +10,1114 @@ Features:
 - Response panel (green) - shows final response
 - Thread ID support for multi-turn conversations
 - Interactive mode with prompt_toolkit
-- Configuration management (onboard, config commands)
 """
 
-import logging
+import argparse
+import asyncio
 import os
 import sys
 import uuid
 from datetime import datetime
-from typing import Any, Optional
+from typing import Any, AsyncIterator
 
-import typer  # type: ignore[import-untyped]
+from dotenv import load_dotenv  # type: ignore[import-untyped]
 from prompt_toolkit import PromptSession  # type: ignore[import-untyped]
 from prompt_toolkit.history import FileHistory  # type: ignore[import-untyped]
 from prompt_toolkit.auto_suggest import AutoSuggestFromHistory  # type: ignore[import-untyped]
 from prompt_toolkit.formatted_text import HTML  # type: ignore[import-untyped]
+from rich.console import Console, Group  # type: ignore[import-untyped]
+from rich.panel import Panel  # type: ignore[import-untyped]
+from rich.markdown import Markdown  # type: ignore[import-untyped]
+from rich.live import Live  # type: ignore[import-untyped]
 from rich.text import Text  # type: ignore[import-untyped]
-from rich.table import Table  # type: ignore[import-untyped]
+from rich.spinner import Spinner  # type: ignore[import-untyped]
+from langchain_core.messages import AIMessage, AIMessageChunk  # type: ignore[import-untyped]
 
-# Backward-compat re-exports (tests import these from EvoScientist.cli)
-from .stream.state import SubAgentState, StreamState, _parse_todo_items, _build_todo_stats  # noqa: F401
-from .stream.display import console, _run_streaming
-from .paths import ensure_dirs, new_run_dir, default_workspace_dir
+from .stream import (
+    StreamEventEmitter,
+    ToolCallTracker,
+    ToolResultFormatter,
+    DisplayLimits,
+    ToolStatus,
+    format_tool_compact,
+    is_success,
+)
 
+load_dotenv(override=True)
 
-def _shorten_path(path: str) -> str:
-    """Shorten absolute path to relative path from current directory."""
-    if not path:
-        return path
-    try:
-        cwd = os.getcwd()
-        if path.startswith(cwd):
-            # Remove cwd prefix, keep the relative part
-            rel = path[len(cwd):].lstrip(os.sep)
-            # Add current dir name for context
-            return os.path.join(os.path.basename(cwd), rel) if rel else os.path.basename(cwd)
-        return path
-    except Exception:
-        return path
+console = Console(
+    legacy_windows=(sys.platform == 'win32'),
+    no_color=os.getenv('NO_COLOR') is not None,
+)
+
+formatter = ToolResultFormatter()
 
 
 # =============================================================================
-# Banner
+# Stream event generator
 # =============================================================================
 
-EVOSCIENTIST_ASCII_LINES = [
-    r" ███████╗ ██╗   ██╗  ██████╗  ███████╗  ██████╗ ██╗ ███████╗ ███╗   ██╗ ████████╗ ██╗ ███████╗ ████████╗",
-    r" ██╔════╝ ██║   ██║ ██╔═══██╗ ██╔════╝ ██╔════╝ ██║ ██╔════╝ ████╗  ██║ ╚══██╔══╝ ██║ ██╔════╝ ╚══██╔══╝",
-    r" █████╗   ██║   ██║ ██║   ██║ ███████╗ ██║      ██║ █████╗   ██╔██╗ ██║    ██║    ██║ ███████╗    ██║   ",
-    r" ██╔══╝   ╚██╗ ██╔╝ ██║   ██║ ╚════██║ ██║      ██║ ██╔══╝   ██║╚██╗██║    ██║    ██║ ╚════██║    ██║   ",
-    r" ███████╗  ╚████╔╝  ╚██████╔╝ ███████║ ╚██████╗ ██║ ███████╗ ██║ ╚████║    ██║    ██║ ███████║    ██║   ",
-    r" ╚══════╝   ╚═══╝    ╚═════╝  ╚══════╝  ╚═════╝ ╚═╝ ╚══════╝ ╚═╝  ╚═══╝    ╚═╝    ╚═╝ ╚══════╝    ╚═╝   ",
-]
+async def stream_agent_events(agent: Any, message: str, thread_id: str) -> AsyncIterator[dict]:
+    """Stream events from the agent graph using async iteration.
 
-# Blue gradient: deep navy -> royal blue -> sky blue -> cyan
-_GRADIENT_COLORS = ["#1a237e", "#1565c0", "#1e88e5", "#42a5f5", "#64b5f6", "#90caf9"]
+    Uses agent.astream() with subgraphs=True to see sub-agent activity.
 
+    Args:
+        agent: Compiled state graph from create_deep_agent()
+        message: User message
+        thread_id: Thread ID for conversation persistence
+
+    Yields:
+        Event dicts: thinking, text, tool_call, tool_result,
+                     subagent_start, subagent_tool_call, subagent_tool_result, subagent_end,
+                     done, error
+    """
+    config = {"configurable": {"thread_id": thread_id}}
+    emitter = StreamEventEmitter()
+    tracker = ToolCallTracker()
+    full_response = ""
+
+    # Track sub-agent names by root namespace element
+    _subagent_names: dict[str, str] = {}  # root_ns_element → display name
+    # Track which task tool_call_ids have been announced
+    _announced_tasks: set[str] = set()
+
+    def _get_subagent_name(namespace: tuple) -> str | None:
+        """Get sub-agent name from namespace, or None if main agent.
+
+        Any non-empty namespace is a sub-agent. Name is resolved by checking
+        all registered names for a prefix match against namespace elements.
+        """
+        if not namespace:
+            return None
+        root = str(namespace[0]) if namespace else ""
+        # Exact match
+        if root in _subagent_names:
+            return _subagent_names[root]
+        # Prefix match: namespace root might be "task:abc123" and we
+        # registered "task:call_xyz" — check if any registered key
+        # appears as a substring of the root or vice versa
+        for key, name in _subagent_names.items():
+            if key in root or root in key:
+                _subagent_names[root] = name  # cache for next lookup
+                return name
+        # Auto-register: infer from namespace string
+        if ":" in root:
+            inferred = root.split(":")[0]
+        else:
+            inferred = root
+        name = inferred or "sub-agent"
+        _subagent_names[root] = name
+        return name
 
-def print_banner(
-    thread_id: str,
-    workspace_dir: str | None = None,
-    memory_dir: str | None = None,
-    mode: str | None = None,
-    model: str | None = None,
-    provider: str | None = None,
-):
-    """Print welcome banner with ASCII art logo, thread ID, workspace path, and mode."""
-    for line, color in zip(EVOSCIENTIST_ASCII_LINES, _GRADIENT_COLORS):
-        console.print(Text(line, style=f"{color} bold"))
-    info = Text()
-    info.append("  Thread: ", style="dim")
-    info.append(thread_id, style="yellow")
-    if workspace_dir:
-        info.append("\n  Workspace: ", style="dim")
-        info.append(_shorten_path(workspace_dir), style="cyan")
-    if memory_dir:
-        trimmed = memory_dir.rstrip("/").rstrip("\\")
-        info.append("\n  Memory dir: ", style="dim")
-        info.append(_shorten_path(trimmed), style="cyan")
-    if model or provider or mode:
-        info.append("\n  ", style="dim")
-        parts = []
-        if model:
-            parts.append(("Model: ", model))
-        if provider:
-            parts.append(("Provider: ", provider))
-        if mode:
-            parts.append(("Mode: ", mode))
-        for i, (label, value) in enumerate(parts):
-            if i > 0:
-                info.append("  ", style="dim")
-            info.append(label, style="dim")
-            info.append(value, style="magenta")
-    info.append("\n  Commands: ", style="dim")
-    info.append("/exit", style="bold")
-    info.append(", ", style="dim")
-    info.append("/new", style="bold")
-    info.append(", ", style="dim")
-    info.append("/thread", style="bold")
-    info.append(", ", style="dim")
-    info.append("/skills", style="bold")
-    info.append(", ", style="dim")
-    info.append("/install-skill", style="bold")
-    info.append(", ", style="dim")
-    info.append("/uninstall-skill", style="bold")
-    console.print(info)
-    console.print()
+    try:
+        async for chunk in agent.astream(
+            {"messages": [{"role": "user", "content": message}]},
+            config=config,
+            stream_mode="messages",
+            subgraphs=True,
+        ):
+            # With subgraphs=True, event is (namespace, (message, metadata))
+            namespace: tuple = ()
+            data: Any = chunk
+
+            if isinstance(chunk, tuple) and len(chunk) >= 2:
+                first = chunk[0]
+                if isinstance(first, tuple):
+                    # (namespace_tuple, (message, metadata))
+                    namespace = first
+                    data = chunk[1]
+                else:
+                    # (message, metadata) — no namespace
+                    data = chunk
+
+            # Unpack message from data
+            msg: Any
+            if isinstance(data, tuple) and len(data) >= 2:
+                msg = data[0]
+            else:
+                msg = data
+
+            subagent = _get_subagent_name(namespace)
+
+            # Process AIMessageChunk / AIMessage
+            if isinstance(msg, (AIMessageChunk, AIMessage)):
+                if subagent:
+                    # Sub-agent content — emit sub-agent events
+                    for ev in _process_chunk_content(msg, emitter, tracker):
+                        if ev.type == "tool_call":
+                            yield emitter.subagent_tool_call(
+                                subagent, ev.data["name"], ev.data["args"], ev.data.get("id", "")
+                            ).data
+                        # Skip text/thinking from sub-agents (too noisy)
+
+                    if hasattr(msg, "tool_calls") and msg.tool_calls:
+                        for tc in msg.tool_calls:
+                            name = tc.get("name", "")
+                            args = tc.get("args", {})
+                            tool_id = tc.get("id", "")
+                            # Skip empty-name chunks (incomplete streaming fragments)
+                            if not name and not tool_id:
+                                continue
+                            yield emitter.subagent_tool_call(
+                                subagent, name, args if isinstance(args, dict) else {}, tool_id
+                            ).data
+                else:
+                    # Main agent content
+                    for ev in _process_chunk_content(msg, emitter, tracker):
+                        if ev.type == "text":
+                            full_response += ev.data.get("content", "")
+                        yield ev.data
+
+                    if hasattr(msg, "tool_calls") and msg.tool_calls:
+                        for ev in _process_tool_calls(msg.tool_calls, emitter, tracker):
+                            yield ev.data
+                            # Detect task tool calls → announce sub-agent
+                            tc_data = ev.data
+                            if tc_data.get("name") == "task":
+                                tool_id = tc_data.get("id", "")
+                                if tool_id and tool_id not in _announced_tasks:
+                                    _announced_tasks.add(tool_id)
+                                    args = tc_data.get("args", {})
+                                    sa_name = args.get("subagent_type", "").strip()
+                                    desc = args.get("description", "").strip()
+                                    # Use subagent_type as name; fall back to description snippet
+                                    if not sa_name:
+                                        sa_name = desc[:30] + "..." if len(desc) > 30 else desc
+                                    if not sa_name:
+                                        sa_name = "sub-agent"
+                                    # Pre-register name so namespace lookup finds it
+                                    _subagent_names[f"task:{tool_id}"] = sa_name
+                                    yield emitter.subagent_start(sa_name, desc).data
+
+            # Process ToolMessage (tool execution result)
+            elif hasattr(msg, "type") and msg.type == "tool":
+                if subagent:
+                    name = getattr(msg, "name", "unknown")
+                    raw_content = str(getattr(msg, "content", ""))
+                    content = raw_content[:DisplayLimits.TOOL_RESULT_MAX]
+                    success = is_success(content)
+                    yield emitter.subagent_tool_result(subagent, name, content, success).data
+                else:
+                    for ev in _process_tool_result(msg, emitter, tracker):
+                        yield ev.data
+                    # Check if this is a task result → sub-agent ended
+                    name = getattr(msg, "name", "")
+                    if name == "task":
+                        tool_call_id = getattr(msg, "tool_call_id", "")
+                        # Find the sub-agent name for this task
+                        sa_key = f"task:{tool_call_id}"
+                        sa_name = _subagent_names.get(sa_key, "sub-agent")
+                        yield emitter.subagent_end(sa_name).data
+
+    except Exception as e:
+        yield emitter.error(str(e)).data
+        raise
+
+    yield emitter.done(full_response).data
+
+
+def _process_chunk_content(chunk, emitter: StreamEventEmitter, tracker: ToolCallTracker):
+    """Process content blocks from an AI message chunk."""
+    content = chunk.content
+
+    if isinstance(content, str):
+        if content:
+            yield emitter.text(content)
+            return
+
+    blocks = None
+    if hasattr(chunk, "content_blocks"):
+        try:
+            blocks = chunk.content_blocks
+        except Exception:
+            blocks = None
+
+    if blocks is None:
+        if isinstance(content, dict):
+            blocks = [content]
+        elif isinstance(content, list):
+            blocks = content
+        else:
+            return
+
+    for raw_block in blocks:
+        block = raw_block
+        if not isinstance(block, dict):
+            if hasattr(block, "model_dump"):
+                block = block.model_dump()
+            elif hasattr(block, "dict"):
+                block = block.dict()
+            else:
+                continue
+
+        block_type = block.get("type")
+
+        if block_type in ("thinking", "reasoning"):
+            thinking_text = block.get("thinking") or block.get("reasoning") or ""
+            if thinking_text:
+                yield emitter.thinking(thinking_text)
+
+        elif block_type == "text":
+            text = block.get("text") or block.get("content") or ""
+            if text:
+                yield emitter.text(text)
+
+        elif block_type in ("tool_use", "tool_call"):
+            tool_id = block.get("id", "")
+            name = block.get("name", "")
+            args = block.get("input") if block_type == "tool_use" else block.get("args")
+            args_payload = args if isinstance(args, dict) else {}
+
+            if tool_id:
+                tracker.update(tool_id, name=name, args=args_payload)
+                if tracker.is_ready(tool_id):
+                    tracker.mark_emitted(tool_id)
+                    yield emitter.tool_call(name, args_payload, tool_id)
+
+        elif block_type == "input_json_delta":
+            partial_json = block.get("partial_json", "")
+            if partial_json:
+                tracker.append_json_delta(partial_json, block.get("index", 0))
+
+        elif block_type == "tool_call_chunk":
+            tool_id = block.get("id", "")
+            name = block.get("name", "")
+            if tool_id:
+                tracker.update(tool_id, name=name)
+            partial_args = block.get("args", "")
+            if isinstance(partial_args, str) and partial_args:
+                tracker.append_json_delta(partial_args, block.get("index", 0))
+
+
+def _process_tool_calls(tool_calls: list, emitter: StreamEventEmitter, tracker: ToolCallTracker):
+    """Process tool_calls from chunk.tool_calls attribute."""
+    for tc in tool_calls:
+        tool_id = tc.get("id", "")
+        if tool_id:
+            name = tc.get("name", "")
+            args = tc.get("args", {})
+            args_payload = args if isinstance(args, dict) else {}
+
+            tracker.update(tool_id, name=name, args=args_payload)
+            if tracker.is_ready(tool_id):
+                tracker.mark_emitted(tool_id)
+                yield emitter.tool_call(name, args_payload, tool_id)
+
+
+def _process_tool_result(chunk, emitter: StreamEventEmitter, tracker: ToolCallTracker):
+    """Process a ToolMessage result."""
+    tracker.finalize_all()
+
+    # Re-emit all tool calls with complete args
+    for info in tracker.get_all():
+        yield emitter.tool_call(info.name, info.args, info.id)
+
+    name = getattr(chunk, "name", "unknown")
+    raw_content = str(getattr(chunk, "content", ""))
+    content = raw_content[:DisplayLimits.TOOL_RESULT_MAX]
+    if len(raw_content) > DisplayLimits.TOOL_RESULT_MAX:
+        content += "\n... (truncated)"
+
+    success = is_success(content)
+    yield emitter.tool_result(name, content, success)
 
 
 # =============================================================================
-# Skill management commands
+# Stream state
 # =============================================================================
 
+class SubAgentState:
+    """Tracks a single sub-agent's activity."""
 
-def _cmd_list_skills() -> None:
-    """List installed user skills."""
-    from .skills_manager import list_skills
-    from .paths import USER_SKILLS_DIR
+    def __init__(self, name: str, description: str = ""):
+        self.name = name
+        self.description = description
+        self.tool_calls: list[dict] = []
+        self.tool_results: list[dict] = []
+        self._result_map: dict[str, dict] = {}  # tool_call_id → result
+        self.is_active = True
 
-    skills = list_skills(include_system=False)
+    def add_tool_call(self, name: str, args: dict, tool_id: str = ""):
+        # Skip empty-name calls without an id (incomplete streaming chunks)
+        if not name and not tool_id:
+            return
+        tc_data = {"id": tool_id, "name": name, "args": args}
+        if tool_id:
+            for i, tc in enumerate(self.tool_calls):
+                if tc.get("id") == tool_id:
+                    # Merge: keep the non-empty name/args
+                    if name:
+                        self.tool_calls[i]["name"] = name
+                    if args:
+                        self.tool_calls[i]["args"] = args
+                    return
+        # Skip if name is empty and we can't deduplicate by id
+        if not name:
+            return
+        self.tool_calls.append(tc_data)
+
+    def add_tool_result(self, name: str, content: str, success: bool = True):
+        result = {"name": name, "content": content, "success": success}
+        self.tool_results.append(result)
+        # Try to match result to the first unmatched tool call with same name
+        for tc in self.tool_calls:
+            tc_id = tc.get("id", "")
+            tc_name = tc.get("name", "")
+            if tc_id and tc_id not in self._result_map and tc_name == name:
+                self._result_map[tc_id] = result
+                return
+        # Fallback: match first unmatched tool call
+        for tc in self.tool_calls:
+            tc_id = tc.get("id", "")
+            if tc_id and tc_id not in self._result_map:
+                self._result_map[tc_id] = result
+                return
+
+    def get_result_for(self, tc: dict) -> dict | None:
+        """Get matched result for a tool call."""
+        tc_id = tc.get("id", "")
+        if tc_id:
+            return self._result_map.get(tc_id)
+        # Fallback: index-based matching
+        try:
+            idx = self.tool_calls.index(tc)
+            if idx < len(self.tool_results):
+                return self.tool_results[idx]
+        except ValueError:
+            pass
+        return None
+
+
+class StreamState:
+    """Accumulates stream state for display updates."""
+
+    def __init__(self):
+        self.thinking_text = ""
+        self.response_text = ""
+        self.tool_calls = []
+        self.tool_results = []
+        self.is_thinking = False
+        self.is_responding = False
+        self.is_processing = False
+        # Sub-agent tracking
+        self.subagents: list[SubAgentState] = []
+        self._subagent_map: dict[str, SubAgentState] = {}  # name → state
+
+    def _get_or_create_subagent(self, name: str, description: str = "") -> SubAgentState:
+        if name not in self._subagent_map:
+            # Check if there's a generic "sub-agent" entry that should be merged
+            # This happens when namespace events arrive before the task tool call
+            # registers the proper name
+            if name != "sub-agent" and "sub-agent" in self._subagent_map:
+                old_sa = self._subagent_map.pop("sub-agent")
+                old_sa.name = name
+                if description:
+                    old_sa.description = description
+                self._subagent_map[name] = old_sa
+                return old_sa
+            sa = SubAgentState(name, description)
+            self.subagents.append(sa)
+            self._subagent_map[name] = sa
+        elif description and not self._subagent_map[name].description:
+            self._subagent_map[name].description = description
+        return self._subagent_map[name]
+
+    def handle_event(self, event: dict) -> str:
+        """Process a single stream event, update internal state, return event type."""
+        event_type: str = event.get("type", "")
+
+        if event_type == "thinking":
+            self.is_thinking = True
+            self.is_responding = False
+            self.is_processing = False
+            self.thinking_text += event.get("content", "")
+
+        elif event_type == "text":
+            self.is_thinking = False
+            self.is_responding = True
+            self.is_processing = False
+            self.response_text += event.get("content", "")
+
+        elif event_type == "tool_call":
+            self.is_thinking = False
+            self.is_responding = False
+            self.is_processing = False
+
+            tool_id = event.get("id", "")
+            tc_data = {
+                "id": tool_id,
+                "name": event.get("name", "unknown"),
+                "args": event.get("args", {}),
+            }
+
+            if tool_id:
+                updated = False
+                for i, tc in enumerate(self.tool_calls):
+                    if tc.get("id") == tool_id:
+                        self.tool_calls[i] = tc_data
+                        updated = True
+                        break
+                if not updated:
+                    self.tool_calls.append(tc_data)
+            else:
+                self.tool_calls.append(tc_data)
+
+        elif event_type == "tool_result":
+            self.is_processing = True
+            self.tool_results.append({
+                "name": event.get("name", "unknown"),
+                "content": event.get("content", ""),
+            })
+
+        elif event_type == "subagent_start":
+            name = event.get("name", "sub-agent")
+            desc = event.get("description", "")
+            sa = self._get_or_create_subagent(name, desc)
+            sa.is_active = True
+
+        elif event_type == "subagent_tool_call":
+            sa_name = event.get("subagent", "sub-agent")
+            sa = self._get_or_create_subagent(sa_name)
+            sa.add_tool_call(
+                event.get("name", "unknown"),
+                event.get("args", {}),
+                event.get("id", ""),
+            )
+
+        elif event_type == "subagent_tool_result":
+            sa_name = event.get("subagent", "sub-agent")
+            sa = self._get_or_create_subagent(sa_name)
+            sa.add_tool_result(
+                event.get("name", "unknown"),
+                event.get("content", ""),
+                event.get("success", True),
+            )
+
+        elif event_type == "subagent_end":
+            name = event.get("name", "sub-agent")
+            if name in self._subagent_map:
+                self._subagent_map[name].is_active = False
+
+        elif event_type == "done":
+            self.is_processing = False
+            if not self.response_text:
+                self.response_text = event.get("response", "")
+
+        elif event_type == "error":
+            self.is_processing = False
+            self.is_thinking = False
+            self.is_responding = False
+            error_msg = event.get("message", "Unknown error")
+            self.response_text += f"\n\n[Error] {error_msg}"
+
+        return event_type
+
+    def get_display_args(self) -> dict:
+        """Get kwargs for create_streaming_display()."""
+        return {
+            "thinking_text": self.thinking_text,
+            "response_text": self.response_text,
+            "tool_calls": self.tool_calls,
+            "tool_results": self.tool_results,
+            "is_thinking": self.is_thinking,
+            "is_responding": self.is_responding,
+            "is_processing": self.is_processing,
+            "subagents": self.subagents,
+        }
 
-    if not skills:
-        console.print("[dim]No user skills installed.[/dim]")
-        console.print("[dim]Install with:[/dim] /install-skill <path-or-url>")
-        console.print(f"[dim]Skills directory:[/dim] [cyan]{_shorten_path(str(USER_SKILLS_DIR))}[/cyan]")
-        console.print()
-        return
 
-    console.print(f"[bold]Installed Skills[/bold] ({len(skills)}):")
-    for skill in skills:
-        console.print(f"  [green]{skill.name}[/green] - {skill.description}")
-    console.print(f"\n[dim]Location:[/dim] [cyan]{_shorten_path(str(USER_SKILLS_DIR))}[/cyan]")
-    console.print()
+# =============================================================================
+# Display functions
+# =============================================================================
 
+def _parse_todo_items(content: str) -> list[dict] | None:
+    """Parse todo items from write_todos output.
 
-def _cmd_install_skill(source: str) -> None:
-    """Install a skill from local path or GitHub URL."""
-    from .skills_manager import install_skill
+    Attempts to extract a list of dicts with 'status' and 'content' keys
+    from the tool result string. Returns None if parsing fails.
+    """
+    import ast
+    import json
 
-    if not source:
-        console.print("[red]Usage:[/red] /install-skill <path-or-url>")
-        console.print("[dim]Examples:[/dim]")
-        console.print("  /install-skill ./my-skill")
-        console.print("  /install-skill https://github.com/user/repo/tree/main/skill-name")
-        console.print("  /install-skill user/repo@skill-name")
-        console.print()
-        return
+    content = content.strip()
+
+    # Try JSON first
+    try:
+        data = json.loads(content)
+        if isinstance(data, list) and data and isinstance(data[0], dict):
+            return data
+    except (json.JSONDecodeError, ValueError):
+        pass
 
-    console.print(f"[dim]Installing skill from:[/dim] {source}")
+    # Try Python literal
+    try:
+        data = ast.literal_eval(content)
+        if isinstance(data, list) and data and isinstance(data[0], dict):
+            return data
+    except (ValueError, SyntaxError):
+        pass
+
+    # Try to find a list embedded in the output
+    for line in content.split("\n"):
+        line = line.strip()
+        if line.startswith("[") and line.endswith("]"):
+            try:
+                data = json.loads(line)
+                if isinstance(data, list):
+                    return data
+            except (json.JSONDecodeError, ValueError):
+                try:
+                    data = ast.literal_eval(line)
+                    if isinstance(data, list):
+                        return data
+                except (ValueError, SyntaxError):
+                    pass
+
+    return None
+
+
+def _build_todo_stats(items: list[dict]) -> str:
+    """Build stats string like '2 active | 1 pending | 3 done'."""
+    counts: dict[str, int] = {}
+    for item in items:
+        status = str(item.get("status", "todo")).lower()
+        # Normalize status names
+        if status in ("done", "completed", "complete"):
+            status = "done"
+        elif status in ("active", "in_progress", "in-progress", "working"):
+            status = "active"
+        else:
+            status = "pending"
+        counts[status] = counts.get(status, 0) + 1
+
+    parts = []
+    for key in ("active", "pending", "done"):
+        if counts.get(key, 0) > 0:
+            parts.append(f"{counts[key]} {key}")
+    return " | ".join(parts) if parts else f"{len(items)} items"
+
+
+def _format_single_todo(item: dict) -> Text:
+    """Format a single todo item with status symbol."""
+    status = str(item.get("status", "todo")).lower()
+    content_text = str(item.get("content", item.get("task", item.get("title", ""))))
+
+    if status in ("done", "completed", "complete"):
+        symbol = "\u2713"
+        label = "done  "
+        style = "green dim"
+    elif status in ("active", "in_progress", "in-progress", "working"):
+        symbol = "\u25cf"
+        label = "active"
+        style = "yellow"
+    else:
+        symbol = "\u25cb"
+        label = "todo  "
+        style = "dim"
 
-    result = install_skill(source)
+    line = Text()
+    line.append(f"    {symbol} ", style=style)
+    line.append(label, style=style)
+    line.append(" ", style="dim")
+    # Truncate long content
+    if len(content_text) > 60:
+        content_text = content_text[:57] + "..."
+    line.append(content_text, style=style)
+    return line
 
-    if result["success"]:
-        console.print(f"[green]Installed:[/green] {result['name']}")
-        console.print(f"[dim]Description:[/dim] {result.get('description', '(none)')}")
-        console.print(f"[dim]Path:[/dim] [cyan]{_shorten_path(result['path'])}[/cyan]")
-        console.print()
-        console.print("[dim]Reload the agent with /new to use the skill.[/dim]")
+
+def format_tool_result_compact(_name: str, content: str, max_lines: int = 5) -> list:
+    """Format tool result as tree output.
+
+    Special handling for write_todos: shows formatted checklist with status symbols.
+    """
+    elements = []
+
+    if not content.strip():
+        elements.append(Text("  \u2514 (empty)", style="dim"))
+        return elements
+
+    # Special handling for write_todos
+    if _name == "write_todos":
+        items = _parse_todo_items(content)
+        if items:
+            stats = _build_todo_stats(items)
+            stats_line = Text()
+            stats_line.append("  \u2514 ", style="dim")
+            stats_line.append(stats, style="dim")
+            elements.append(stats_line)
+            elements.append(Text("", style="dim"))  # blank line
+
+            max_preview = 4
+            for item in items[:max_preview]:
+                elements.append(_format_single_todo(item))
+
+            remaining = len(items) - max_preview
+            if remaining > 0:
+                elements.append(Text(f"    ... {remaining} more", style="dim italic"))
+
+            return elements
+
+    lines = content.strip().split("\n")
+    total_lines = len(lines)
+
+    display_lines = lines[:max_lines]
+    for i, line in enumerate(display_lines):
+        prefix = "\u2514" if i == 0 else " "
+        if len(line) > 80:
+            line = line[:77] + "..."
+        style = "dim" if is_success(content) else "red dim"
+        elements.append(Text(f"  {prefix} {line}", style=style))
+
+    remaining = total_lines - max_lines
+    if remaining > 0:
+        elements.append(Text(f"    ... +{remaining} lines", style="dim italic"))
+
+    return elements
+
+
+def _render_tool_call_line(tc: dict, tr: dict | None) -> Text:
+    """Render a single tool call line with status indicator."""
+    is_task = tc.get('name', '').lower() == 'task'
+
+    if tr is not None:
+        content = tr.get('content', '')
+        if is_success(content):
+            style = "bold green"
+            indicator = "\u2713" if is_task else ToolStatus.SUCCESS.value
+        else:
+            style = "bold red"
+            indicator = "\u2717" if is_task else ToolStatus.ERROR.value
     else:
-        console.print(f"[red]Failed:[/red] {result['error']}")
-    console.print()
+        style = "bold yellow" if not is_task else "bold cyan"
+        indicator = "\u25b6" if is_task else ToolStatus.RUNNING.value
 
+    tool_compact = format_tool_compact(tc['name'], tc.get('args'))
+    tool_text = Text()
+    tool_text.append(f"{indicator} ", style=style)
+    tool_text.append(tool_compact, style=style)
+    return tool_text
 
-def _cmd_uninstall_skill(name: str) -> None:
-    """Uninstall a user-installed skill."""
-    from .skills_manager import uninstall_skill
 
-    if not name:
-        console.print("[red]Usage:[/red] /uninstall-skill <skill-name>")
-        console.print("[dim]Use /skills to see installed skills.[/dim]")
-        console.print()
-        return
+def _render_subagent_section(sa: 'SubAgentState', compact: bool = False) -> list:
+    """Render a sub-agent's activity as a compact indented section.
+
+    Args:
+        sa: Sub-agent state to render
+        compact: If True, render minimal 1-2 line summary (for final display)
+
+    Completed tools are collapsed into a summary line.
+    Only the currently running tool is shown expanded.
+    """
+    elements = []
+    BORDER = "dim cyan" if sa.is_active else "dim"
+
+    # Filter out tool calls with empty names
+    valid_calls = [tc for tc in sa.tool_calls if tc.get("name")]
+
+    # Split into completed and pending
+    completed = []
+    pending = []
+    for tc in valid_calls:
+        tr = sa.get_result_for(tc)
+        if tr is not None:
+            completed.append((tc, tr))
+        else:
+            pending.append(tc)
 
-    result = uninstall_skill(name)
+    succeeded = sum(1 for _, tr in completed if tr.get("success", True))
+    failed = len(completed) - succeeded
 
-    if result["success"]:
-        console.print(f"[green]Uninstalled:[/green] {name}")
-        console.print("[dim]Reload the agent with /new to apply changes.[/dim]")
+    # --- Compact mode: 1-2 line summary for final display ---
+    if compact:
+        line = Text()
+        if not sa.is_active:
+            line.append("  \u2713 ", style="green")
+            line.append(sa.name, style="bold green")
+        else:
+            line.append("  \u25b6 ", style="cyan")
+            line.append(sa.name, style="bold cyan")
+        if sa.description:
+            desc = sa.description[:50] + "..." if len(sa.description) > 50 else sa.description
+            line.append(f" \u2014 {desc}", style="dim")
+        elements.append(line)
+        # Stats line
+        if valid_calls:
+            stats = Text("    ")
+            stats.append(f"{succeeded} completed", style="dim green")
+            if failed > 0:
+                stats.append(f" \u00b7 {failed} failed", style="dim red")
+            if pending:
+                stats.append(f" \u00b7 {len(pending)} running", style="dim yellow")
+            elements.append(stats)
+        return elements
+
+    # --- Full mode: bordered section for Live streaming ---
+    # Shows every tool call individually with status indicators
+
+    # Header
+    header = Text()
+    header.append("  \u250c ", style=BORDER)
+    if sa.is_active:
+        header.append(sa.name, style="bold cyan")
     else:
-        console.print(f"[red]Failed:[/red] {result['error']}")
-    console.print()
+        header.append(sa.name, style="bold green")
+        header.append(" \u2713", style="green")
+    if sa.description:
+        desc = sa.description[:55] + "..." if len(sa.description) > 55 else sa.description
+        header.append(f" \u2014 {desc}", style="dim")
+    elements.append(header)
+
+    # Show every tool call with its status
+    for tc, tr in completed:
+        tc_line = Text("  \u2502 ", style=BORDER)
+        tc_name = format_tool_compact(tc["name"], tc.get("args"))
+        if tr.get("success", True):
+            tc_line.append(f"\u2713 {tc_name}", style="green")
+        else:
+            tc_line.append(f"\u2717 {tc_name}", style="red")
+            # Show first line of error
+            content = tr.get("content", "")
+            first_line = content.strip().split("\n")[0][:70]
+            if first_line:
+                err_line = Text("  \u2502   ", style=BORDER)
+                err_line.append(f"\u2514 {first_line}", style="red dim")
+                elements.append(tc_line)
+                elements.append(err_line)
+                continue
+        elements.append(tc_line)
+
+    # Pending/running tools
+    for tc in pending:
+        tc_line = Text("  \u2502 ", style=BORDER)
+        tc_name = format_tool_compact(tc["name"], tc.get("args"))
+        tc_line.append(f"\u25cf {tc_name}", style="bold yellow")
+        elements.append(tc_line)
+        spinner_line = Text("  \u2502   ", style=BORDER)
+        spinner_line.append("\u21bb running...", style="yellow dim")
+        elements.append(spinner_line)
+
+    # Footer
+    if not sa.is_active:
+        total = len(valid_calls)
+        footer = Text(f"  \u2514 done ({total} tools)", style="dim green")
+        elements.append(footer)
+    elif valid_calls:
+        footer = Text("  \u2514 running...", style="dim cyan")
+        elements.append(footer)
+
+    return elements
+
+
+def create_streaming_display(
+    thinking_text: str = "",
+    response_text: str = "",
+    tool_calls: list | None = None,
+    tool_results: list | None = None,
+    is_thinking: bool = False,
+    is_responding: bool = False,
+    is_waiting: bool = False,
+    is_processing: bool = False,
+    show_thinking: bool = True,
+    subagents: list | None = None,
+) -> Any:
+    """Create Rich display layout for streaming output.
+
+    Returns:
+        Rich Group for Live display
+    """
+    elements = []
+    tool_calls = tool_calls or []
+    tool_results = tool_results or []
+    subagents = subagents or []
+
+    # Initial waiting state
+    if is_waiting and not thinking_text and not response_text and not tool_calls:
+        spinner = Spinner("dots", text=" Thinking...", style="cyan")
+        elements.append(spinner)
+        return Group(*elements)
+
+    # Thinking panel
+    if show_thinking and thinking_text:
+        thinking_title = "Thinking"
+        if is_thinking:
+            thinking_title += " ..."
+        display_thinking = thinking_text
+        if len(display_thinking) > DisplayLimits.THINKING_STREAM:
+            display_thinking = "..." + display_thinking[-DisplayLimits.THINKING_STREAM:]
+        elements.append(Panel(
+            Text(display_thinking, style="dim"),
+            title=thinking_title,
+            border_style="blue",
+            padding=(0, 1),
+        ))
+
+    # Tool calls and results paired display
+    # Collapse older completed tools to prevent overflow in Live mode
+    MAX_VISIBLE_TOOLS = 4
+
+    if tool_calls:
+        # Split into completed and pending/running
+        completed_tools = []
+        recent_tools = []  # last few completed + all pending
+
+        for i, tc in enumerate(tool_calls):
+            has_result = i < len(tool_results)
+            tr = tool_results[i] if has_result else None
+            if has_result:
+                completed_tools.append((tc, tr))
+            else:
+                recent_tools.append((tc, None))
+
+        # Determine how many completed tools to show
+        # Keep the last few completed + all pending within MAX_VISIBLE_TOOLS
+        slots_for_completed = max(0, MAX_VISIBLE_TOOLS - len(recent_tools))
+        hidden_completed = completed_tools[:-slots_for_completed] if slots_for_completed and len(completed_tools) > slots_for_completed else (completed_tools if not slots_for_completed else [])
+        visible_completed = completed_tools[-slots_for_completed:] if slots_for_completed else []
+
+        # Summary line for hidden completed tools
+        if hidden_completed:
+            ok = sum(1 for _, tr in hidden_completed if is_success(tr.get('content', '')))
+            fail = len(hidden_completed) - ok
+            summary = Text()
+            summary.append(f"\u2713 {ok} completed", style="dim green")
+            if fail > 0:
+                summary.append(f" | {fail} failed", style="dim red")
+            elements.append(summary)
+
+        # Render visible completed tools (compact: 1 line each, no result expansion)
+        for tc, tr in visible_completed:
+            elements.append(_render_tool_call_line(tc, tr))
+            # Only expand result for write_todos (useful) or errors
+            content = tr.get('content', '') if tr else ''
+            if tc.get('name') == 'write_todos' or (tr and not is_success(content)):
+                result_elements = format_tool_result_compact(
+                    tr['name'],
+                    content,
+                    max_lines=5,
+                )
+                elements.extend(result_elements)
+
+        # Render pending/running tools (expanded with spinner)
+        for tc, tr in recent_tools:
+            elements.append(_render_tool_call_line(tc, tr))
+            if tc.get('name') != 'task':
+                spinner = Spinner("dots", text=" Running...", style="yellow")
+                elements.append(spinner)
+
+    # Sub-agent activity sections
+    for sa in subagents:
+        if sa.tool_calls or sa.is_active:
+            elements.extend(_render_subagent_section(sa))
+
+    # Processing state after tool execution
+    if is_processing and not is_thinking and not is_responding and not response_text:
+        # Check if any sub-agent is active
+        any_active = any(sa.is_active for sa in subagents)
+        if not any_active:
+            spinner = Spinner("dots", text=" Analyzing results...", style="cyan")
+            elements.append(spinner)
+
+    # Response text display logic
+    has_pending_tools = len(tool_calls) > len(tool_results)
+    any_active_subagent = any(sa.is_active for sa in subagents)
+    has_used_tools = len(tool_calls) > 0
+
+    if response_text and not has_pending_tools and not any_active_subagent:
+        if has_used_tools:
+            # Tools were used — treat all text as intermediate during Live streaming.
+            # Final rendering is handled by display_final_results().
+            preview = response_text
+            if len(preview) > 200:
+                preview = "..." + preview[-197:]
+            for line in preview.strip().split("\n")[-3:]:
+                if line.strip():
+                    elements.append(Text(f"    {line.strip()}", style="dim italic"))
+        else:
+            # Pure text response (no tools used) — render as Markdown
+            elements.append(Text(""))  # blank separator
+            elements.append(Markdown(response_text))
+    elif is_responding and not thinking_text and not has_pending_tools:
+        elements.append(Text("Generating response...", style="dim"))
+
+    return Group(*elements) if elements else Text("Processing...", style="dim")
+
+
+def display_final_results(
+    state: StreamState,
+    thinking_max_length: int = DisplayLimits.THINKING_FINAL,
+    show_thinking: bool = True,
+    show_tools: bool = True,
+) -> None:
+    """Display final results after streaming completes."""
+    if show_thinking and state.thinking_text:
+        display_thinking = state.thinking_text
+        if len(display_thinking) > thinking_max_length:
+            half = thinking_max_length // 2
+            display_thinking = display_thinking[:half] + "\n\n... (truncated) ...\n\n" + display_thinking[-half:]
+        console.print(Panel(
+            Text(display_thinking, style="dim"),
+            title="Thinking",
+            border_style="blue",
+        ))
+
+    if show_tools and state.tool_calls:
+        shown_sa_names: set[str] = set()
+
+        for i, tc in enumerate(state.tool_calls):
+            has_result = i < len(state.tool_results)
+            tr = state.tool_results[i] if has_result else None
+            content = tr.get('content', '') if tr is not None else ''
+            is_task = tc.get('name', '').lower() == 'task'
+
+            # Task tools: show delegation line + compact sub-agent summary
+            if is_task:
+                console.print(_render_tool_call_line(tc, tr))
+                sa_name = tc.get('args', {}).get('subagent_type', '')
+                task_desc = tc.get('args', {}).get('description', '')
+                matched_sa = None
+                for sa in state.subagents:
+                    if sa.name == sa_name or (task_desc and task_desc in (sa.description or '')):
+                        matched_sa = sa
+                        break
+                if matched_sa:
+                    shown_sa_names.add(matched_sa.name)
+                    for elem in _render_subagent_section(matched_sa, compact=True):
+                        console.print(elem)
+                continue
+
+            # Regular tools: show tool call line + result
+            console.print(_render_tool_call_line(tc, tr))
+            if has_result and tr is not None:
+                result_elements = format_tool_result_compact(
+                    tr['name'],
+                    content,
+                    max_lines=10,
+                )
+                for elem in result_elements:
+                    console.print(elem)
+
+        # Render any sub-agents not already shown via task tool calls
+        for sa in state.subagents:
+            if sa.name not in shown_sa_names and (sa.tool_calls or sa.is_active):
+                for elem in _render_subagent_section(sa, compact=True):
+                    console.print(elem)
+
+        console.print()
+
+    if state.response_text:
+        console.print()
+        console.print(Markdown(state.response_text))
+        console.print()
 
 
 # =============================================================================
-# CLI commands
+# Async-to-sync bridge
 # =============================================================================
 
-def cmd_interactive(
+def _run_streaming(
     agent: Any,
-    show_thinking: bool = True,
-    workspace_dir: str | None = None,
-    workspace_fixed: bool = False,
-    mode: str | None = None,
-    model: str | None = None,
-    provider: str | None = None,
+    message: str,
+    thread_id: str,
+    show_thinking: bool,
+    interactive: bool,
 ) -> None:
+    """Run async streaming and render with Rich Live display.
+
+    Bridges the async stream_agent_events() into synchronous Rich Live rendering
+    using asyncio.run().
+
+    Args:
+        agent: Compiled agent graph
+        message: User message
+        thread_id: Thread ID
+        show_thinking: Whether to show thinking panel
+        interactive: If True, use simplified final display (no panel)
+    """
+    state = StreamState()
+
+    async def _consume() -> None:
+        async for event in stream_agent_events(agent, message, thread_id):
+            event_type = state.handle_event(event)
+            live.update(create_streaming_display(
+                **state.get_display_args(),
+                show_thinking=show_thinking,
+            ))
+            if event_type in (
+                "tool_call", "tool_result",
+                "subagent_start", "subagent_tool_call",
+                "subagent_tool_result", "subagent_end",
+            ):
+                live.refresh()
+
+    with Live(console=console, refresh_per_second=10, transient=True) as live:
+        live.update(create_streaming_display(is_waiting=True))
+        asyncio.run(_consume())
+
+    if interactive:
+        display_final_results(
+            state,
+            thinking_max_length=500,
+            show_thinking=False,
+            show_tools=True,
+        )
+    else:
+        console.print()
+        display_final_results(
+            state,
+            show_tools=True,
+        )
+
+
+# =============================================================================
+# CLI commands
+# =============================================================================
+
+EVOSCIENTIST_ASCII_LINES = [
+    r" ███████╗ ██╗   ██╗  ██████╗  ███████╗  ██████╗ ██╗ ███████╗ ███╗   ██╗ ████████╗ ██╗ ███████╗ ████████╗",
+    r" ██╔════╝ ██║   ██║ ██╔═══██╗ ██╔════╝ ██╔════╝ ██║ ██╔════╝ ████╗  ██║ ╚══██╔══╝ ██║ ██╔════╝ ╚══██╔══╝",
+    r" █████╗   ██║   ██║ ██║   ██║ ███████╗ ██║      ██║ █████╗   ██╔██╗ ██║    ██║    ██║ ███████╗    ██║   ",
+    r" ██╔══╝   ╚██╗ ██╔╝ ██║   ██║ ╚════██║ ██║      ██║ ██╔══╝   ██║╚██╗██║    ██║    ██║ ╚════██║    ██║   ",
+    r" ███████╗  ╚████╔╝  ╚██████╔╝ ███████║ ╚██████╗ ██║ ███████╗ ██║ ╚████║    ██║    ██║ ███████║    ██║   ",
+    r" ╚══════╝   ╚═══╝    ╚═════╝  ╚══════╝  ╚═════╝ ╚═╝ ╚══════╝ ╚═╝  ╚═══╝    ╚═╝    ╚═╝ ╚══════╝    ╚═╝   ",
+]
+
+# Blue gradient: deep navy → royal blue → sky blue → cyan
+_GRADIENT_COLORS = ["#1a237e", "#1565c0", "#1e88e5", "#42a5f5", "#64b5f6", "#90caf9"]
+
+
+def print_banner(thread_id: str, workspace_dir: str | None = None):
+    """Print welcome banner with ASCII art logo, thread ID, and workspace path."""
+    for line, color in zip(EVOSCIENTIST_ASCII_LINES, _GRADIENT_COLORS):
+        console.print(Text(line, style=f"{color} bold"))
+    info = Text()
+    info.append("  Thread: ", style="dim")
+    info.append(thread_id, style="yellow")
+    if workspace_dir:
+        info.append("\n  Workspace: ", style="dim")
+        info.append(workspace_dir, style="cyan")
+    info.append("\n  Commands: ", style="dim")
+    info.append("/exit", style="bold")
+    info.append(", ", style="dim")
+    info.append("/new", style="bold")
+    info.append(" (new session), ", style="dim")
+    info.append("/thread", style="bold")
+    info.append(" (show thread ID)", style="dim")
+    console.print(info)
+    console.print()
+
+
+def cmd_interactive(agent: Any, show_thinking: bool = True, workspace_dir: str | None = None) -> None:
     """Interactive conversation mode with streaming output.
 
     Args:
         agent: Compiled agent graph
         show_thinking: Whether to display thinking panels
         workspace_dir: Per-session workspace directory path
-        workspace_fixed: If True, /new keeps the same workspace directory
-        mode: Workspace mode ('daemon' or 'run'), displayed in banner
-        model: Model name to display in banner
-        provider: LLM provider name to display in banner
     """
     thread_id = str(uuid.uuid4())
-    from .EvoScientist import MEMORY_DIR
-    memory_dir = MEMORY_DIR
-    print_banner(thread_id, workspace_dir, memory_dir, mode, model, provider)
+    print_banner(thread_id, workspace_dir)
 
     history_file = str(os.path.expanduser("~/.EvoScientist_history"))
     session = PromptSession(
@@ -228,82 +1126,46 @@ def cmd_interactive(
         enable_history_search=True,
     )
 
-    def _print_separator():
-        """Print a horizontal separator line spanning the terminal width."""
-        width = console.size.width
-        console.print(Text("\u2500" * width, style="dim"))
-
-    _print_separator()
     while True:
         try:
             user_input = session.prompt(
-                HTML('<ansiblue><b>&gt;</b></ansiblue> ')
+                HTML('<ansigreen><b>You:</b></ansigreen> ')
             ).strip()
 
             if not user_input:
-                # Erase the empty prompt line so it looks like nothing happened
-                sys.stdout.write("\033[A\033[2K\r")
-                sys.stdout.flush()
                 continue
 
-            _print_separator()
-
             # Special commands
             if user_input.lower() in ("/exit", "/quit", "/q"):
                 console.print("[dim]Goodbye![/dim]")
                 break
 
             if user_input.lower() == "/new":
-                # New session: new thread; workspace only changes if not fixed
-                if not workspace_fixed:
-                    workspace_dir = _create_session_workspace()
+                # New session: new workspace, new agent, new thread
+                workspace_dir = _create_session_workspace()
                 console.print("[dim]Loading new session...[/dim]")
                 agent = _load_agent(workspace_dir=workspace_dir)
                 thread_id = str(uuid.uuid4())
                 console.print(f"[green]New session:[/green] [yellow]{thread_id}[/yellow]")
-                if workspace_dir:
-                    console.print(f"[dim]Workspace:[/dim] [cyan]{_shorten_path(workspace_dir)}[/cyan]\n")
+                console.print(f"[dim]Workspace:[/dim] [cyan]{workspace_dir}[/cyan]\n")
                 continue
 
             if user_input.lower() == "/thread":
                 console.print(f"[dim]Thread:[/dim] [yellow]{thread_id}[/yellow]")
                 if workspace_dir:
-                    console.print(f"[dim]Workspace:[/dim] [cyan]{_shorten_path(workspace_dir)}[/cyan]")
-                if memory_dir:
-                    console.print(f"[dim]Memory dir:[/dim] [cyan]{_shorten_path(memory_dir)}[/cyan]")
+                    console.print(f"[dim]Workspace:[/dim] [cyan]{workspace_dir}[/cyan]")
                 console.print()
                 continue
 
-            if user_input.lower() == "/skills":
-                _cmd_list_skills()
-                continue
-
-            if user_input.lower().startswith("/install-skill"):
-                source = user_input[len("/install-skill"):].strip()
-                _cmd_install_skill(source)
-                continue
-
-            if user_input.lower().startswith("/uninstall-skill"):
-                name = user_input[len("/uninstall-skill"):].strip()
-                _cmd_uninstall_skill(name)
-                continue
-
             # Stream agent response
             console.print()
             _run_streaming(agent, user_input, thread_id, show_thinking, interactive=True)
-            _print_separator()
 
         except KeyboardInterrupt:
             console.print("\n[dim]Goodbye![/dim]")
             break
         except Exception as e:
-            error_msg = str(e)
-            if "authentication" in error_msg.lower() or "api_key" in error_msg.lower():
-                console.print("[red]Error: API key not configured.[/red]")
-                console.print("[dim]Run [bold]EvoSci onboard[/bold] to set up your API key.[/dim]")
-                break
-            else:
-                console.print(f"[red]Error: {e}[/red]")
+            console.print(f"[red]Error: {e}[/red]")
 
 
 def cmd_run(agent: Any, prompt: str, thread_id: str | None = None, show_thinking: bool = True, workspace_dir: str | None = None) -> None:
@@ -318,37 +1180,27 @@ def cmd_run(agent: Any, prompt: str, thread_id: str | None = None, show_thinking
     """
     thread_id = thread_id or str(uuid.uuid4())
 
-    width = console.size.width
-    sep = Text("\u2500" * width, style="dim")
-    console.print(sep)
-    console.print(Text(f"> {prompt}"))
-    console.print(sep)
+    console.print(Panel(f"[bold cyan]Query:[/bold cyan]\n{prompt}"))
     console.print(f"[dim]Thread: {thread_id}[/dim]")
     if workspace_dir:
-        console.print(f"[dim]Workspace: {_shorten_path(workspace_dir)}[/dim]")
+        console.print(f"[dim]Workspace: {workspace_dir}[/dim]")
     console.print()
 
     try:
         _run_streaming(agent, prompt, thread_id, show_thinking, interactive=False)
     except Exception as e:
-        error_msg = str(e)
-        if "authentication" in error_msg.lower() or "api_key" in error_msg.lower():
-            console.print("[red]Error: API key not configured.[/red]")
-            console.print("[dim]Run [bold]EvoSci onboard[/bold] to set up your API key.[/dim]")
-            raise typer.Exit(1)
-        else:
-            console.print(f"[red]Error: {e}[/red]")
-            raise
+        console.print(f"[red]Error: {e}[/red]")
+        raise
 
 
 # =============================================================================
-# Agent loading helpers
+# Entry point
 # =============================================================================
 
 def _create_session_workspace() -> str:
     """Create a per-session workspace directory and return its path."""
     session_id = datetime.now().strftime("%Y%m%d_%H%M%S")
-    workspace_dir = str(new_run_dir(session_id))
+    workspace_dir = os.path.join(".", "workspace", session_id)
     os.makedirs(workspace_dir, exist_ok=True)
     return workspace_dir
 
@@ -363,305 +1215,66 @@ def _load_agent(workspace_dir: str | None = None):
     return create_cli_agent(workspace_dir=workspace_dir)
 
 
-# =============================================================================
-# Typer app
-# =============================================================================
-
-app = typer.Typer(no_args_is_help=False, add_completion=False)
-
-# Config subcommand group
-config_app = typer.Typer(help="Configuration management commands", invoke_without_command=True)
-app.add_typer(config_app, name="config")
-
-
-# =============================================================================
-# Onboard command
-# =============================================================================
-
-@app.command()
-def onboard(
-    skip_validation: bool = typer.Option(
-        False,
-        "--skip-validation",
-        help="Skip API key validation during setup"
-    ),
-):
-    """Interactive setup wizard for EvoScientist.
-
-    Guides you through configuring API keys, model selection,
-    workspace settings, and agent parameters.
-    """
-    from .onboard import run_onboard
-    run_onboard(skip_validation=skip_validation)
-
-
-# =============================================================================
-# Config commands
-# =============================================================================
-
-@config_app.callback(invoke_without_command=True)
-def config_callback(ctx: typer.Context):
-    """Configuration management commands."""
-    if ctx.invoked_subcommand is None:
-        config_list()
-
-
-@config_app.command("list")
-def config_list():
-    """List all configuration values."""
-    from .config import list_config, get_config_path
-
-    config_data = list_config()
-
-    table = Table(title="EvoScientist Configuration", show_header=True)
-    table.add_column("Setting", style="cyan")
-    table.add_column("Value")
-
-    # Mask API keys
-    def format_value(key: str, value: Any) -> str:
-        if "api_key" in key and value:
-            return "***" + str(value)[-4:] if len(str(value)) > 4 else "***"
-        if value == "":
-            return "[dim](not set)[/dim]"
-        return str(value)
-
-    for key, value in config_data.items():
-        table.add_row(key, format_value(key, value))
-
-    console.print(table)
-    console.print(f"\n[dim]Config file: {get_config_path()}[/dim]")
-
-
-@config_app.command("get")
-def config_get(key: str = typer.Argument(..., help="Configuration key to get")):
-    """Get a single configuration value."""
-    from .config import get_config_value
-
-    value = get_config_value(key)
-    if value is None:
-        console.print(f"[red]Unknown key: {key}[/red]")
-        raise typer.Exit(1)
-
-    # Mask API keys
-    if "api_key" in key and value:
-        display_value = "***" + str(value)[-4:] if len(str(value)) > 4 else "***"
-    elif value == "":
-        display_value = "(not set)"
-    else:
-        display_value = str(value)
-
-    console.print(f"[cyan]{key}[/cyan]: {display_value}")
-
-
-@config_app.command("set")
-def config_set(
-    key: str = typer.Argument(..., help="Configuration key to set"),
-    value: str = typer.Argument(..., help="New value"),
-):
-    """Set a single configuration value."""
-    from .config import set_config_value
-
-    if set_config_value(key, value):
-        console.print(f"[green]Set {key}[/green]")
-    else:
-        console.print(f"[red]Invalid key: {key}[/red]")
-        raise typer.Exit(1)
-
-
-@config_app.command("reset")
-def config_reset(
-    yes: bool = typer.Option(False, "--yes", "-y", help="Skip confirmation prompt"),
-):
-    """Reset configuration to defaults."""
-    from .config import reset_config, get_config_path
-
-    config_path = get_config_path()
-
-    if not config_path.exists():
-        console.print("[yellow]No config file to reset.[/yellow]")
-        return
-
-    if not yes:
-        confirm = typer.confirm("Reset configuration to defaults?")
-        if not confirm:
-            console.print("[dim]Cancelled.[/dim]")
-            return
-
-    reset_config()
-    console.print("[green]Configuration reset to defaults.[/green]")
-
-
-@config_app.command("path")
-def config_path():
-    """Show the configuration file path."""
-    from .config import get_config_path
+def main():
+    """CLI entry point."""
+    parser = argparse.ArgumentParser(
+        description="EvoScientist Agent - AI-powered research & code execution CLI",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+Examples:
+  # Interactive mode (default)
+  python -m EvoScientist --interactive
+
+  # Single-shot query
+  python -m EvoScientist "What is quantum computing?"
+
+  # Resume a conversation thread
+  python -m EvoScientist --thread-id <uuid> "Follow-up question"
+
+  # Disable thinking display
+  python -m EvoScientist --no-thinking "Your query"
+""",
+    )
 
-    path = get_config_path()
-    exists = path.exists()
-    status = "[green]exists[/green]" if exists else "[dim]not created yet[/dim]"
-    console.print(f"{path} ({status})")
+    parser.add_argument(
+        "prompt",
+        nargs="?",
+        help="Query to execute (single-shot mode)",
+    )
+    parser.add_argument(
+        "-i", "--interactive",
+        action="store_true",
+        help="Interactive conversation mode",
+    )
+    parser.add_argument(
+        "--thread-id",
+        type=str,
+        default=None,
+        help="Thread ID for conversation persistence (resume session)",
+    )
+    parser.add_argument(
+        "--no-thinking",
+        action="store_true",
+        help="Disable thinking display",
+    )
 
+    args = parser.parse_args()
+    show_thinking = not args.no_thinking
 
-# =============================================================================
-# Main callback (default behavior)
-# =============================================================================
-
-@app.callback(invoke_without_command=True)
-def _main_callback(
-    ctx: typer.Context,
-    prompt: Optional[str] = typer.Option(None, "-p", "--prompt", help="Query to execute (single-shot mode)"),
-    thread_id: Optional[str] = typer.Option(None, "--thread-id", help="Thread ID for conversation persistence"),
-    no_thinking: bool = typer.Option(False, "--no-thinking", help="Disable thinking display"),
-    workdir: Optional[str] = typer.Option(None, "--workdir", help="Override workspace directory for this session"),
-    use_cwd: bool = typer.Option(False, "--use-cwd", help="Use current working directory as workspace"),
-    mode: Optional[str] = typer.Option(
-        None,
-        "--mode",
-        help="Workspace mode: 'daemon' (persistent, default) or 'run' (isolated per-session)"
-    ),
-):
-    """EvoScientist Agent - AI-powered research & code execution CLI."""
-    # If a subcommand was invoked, don't run the default behavior
-    if ctx.invoked_subcommand is not None:
-        return
-
-    from dotenv import load_dotenv, find_dotenv  # type: ignore[import-untyped]
-    # find_dotenv() traverses up the directory tree to locate .env
-    load_dotenv(find_dotenv(), override=True)
-
-    # Load and apply configuration
-    from .config import get_effective_config, apply_config_to_env
-
-    # Build CLI overrides dict
-    cli_overrides = {}
-    if mode:
-        cli_overrides["default_mode"] = mode
-    if workdir:
-        cli_overrides["default_workdir"] = workdir
-    if no_thinking:
-        cli_overrides["show_thinking"] = False
-
-    config = get_effective_config(cli_overrides)
-    apply_config_to_env(config)
-
-    show_thinking = config.show_thinking if not no_thinking else False
-
-    # Validate mutually exclusive options
-    if workdir and use_cwd:
-        raise typer.BadParameter("Use either --workdir or --use-cwd, not both.")
-
-    if mode and (workdir or use_cwd):
-        raise typer.BadParameter("--mode cannot be combined with --workdir or --use-cwd")
-
-    if mode and mode not in ("run", "daemon"):
-        raise typer.BadParameter("--mode must be 'run' or 'daemon'")
-
-    ensure_dirs()
-
-    # Resolve effective mode from config (CLI mode already applied via overrides)
-    effective_mode: str | None = None  # None means explicit --workdir/--use-cwd was used
-
-    # Resolve workspace directory for this session
-    # Priority: --use-cwd > --workdir > --mode (explicit) > default_workdir > default_mode
-    if use_cwd:
-        workspace_dir = os.getcwd()
-        workspace_fixed = True
-    elif workdir:
-        workspace_dir = os.path.abspath(os.path.expanduser(workdir))
-        os.makedirs(workspace_dir, exist_ok=True)
-        workspace_fixed = True
-    elif mode:
-        # Explicit --mode overrides default_workdir
-        effective_mode = mode
-        workspace_root = config.default_workdir or str(default_workspace_dir())
-        workspace_root = os.path.abspath(os.path.expanduser(workspace_root))
-        if effective_mode == "run":
-            session_id = datetime.now().strftime("%Y%m%d_%H%M%S")
-            workspace_dir = os.path.join(workspace_root, "runs", session_id)
-            os.makedirs(workspace_dir, exist_ok=True)
-            workspace_fixed = False
-        else:  # daemon
-            workspace_dir = workspace_root
-            os.makedirs(workspace_dir, exist_ok=True)
-            workspace_fixed = True
-    elif config.default_workdir:
-        # Use configured default workdir with configured mode
-        workspace_root = os.path.abspath(os.path.expanduser(config.default_workdir))
-        effective_mode = config.default_mode
-        if effective_mode == "run":
-            session_id = datetime.now().strftime("%Y%m%d_%H%M%S")
-            workspace_dir = os.path.join(workspace_root, "runs", session_id)
-            os.makedirs(workspace_dir, exist_ok=True)
-            workspace_fixed = False
-        else:  # daemon
-            workspace_dir = workspace_root
-            os.makedirs(workspace_dir, exist_ok=True)
-            workspace_fixed = True
-    else:
-        effective_mode = config.default_mode
-        if effective_mode == "run":
-            workspace_dir = _create_session_workspace()
-            workspace_fixed = False
-        else:  # daemon mode (default)
-            workspace_dir = str(default_workspace_dir())
-            os.makedirs(workspace_dir, exist_ok=True)
-            workspace_fixed = True
+    # Create per-session workspace
+    workspace_dir = _create_session_workspace()
 
     # Load agent with session workspace
     console.print("[dim]Loading agent...[/dim]")
     agent = _load_agent(workspace_dir=workspace_dir)
 
-    if prompt:
-        # Single-shot mode: execute query and exit
-        cmd_run(agent, prompt, thread_id=thread_id, show_thinking=show_thinking, workspace_dir=workspace_dir)
+    if args.interactive:
+        cmd_interactive(agent, show_thinking=show_thinking, workspace_dir=workspace_dir)
+    elif args.prompt:
+        cmd_run(agent, args.prompt, thread_id=args.thread_id, show_thinking=show_thinking, workspace_dir=workspace_dir)
     else:
-        # Interactive mode (default)
-        cmd_interactive(
-            agent,
-            show_thinking=show_thinking,
-            workspace_dir=workspace_dir,
-            workspace_fixed=workspace_fixed,
-            mode=effective_mode,
-            model=config.model,
-            provider=config.provider,
-        )
-
-
-def _configure_logging():
-    """Configure logging with warning symbols for better visibility."""
-    from rich.logging import RichHandler
-
-    class DimWarningHandler(RichHandler):
-        """Custom handler that renders warnings in dim style."""
-
-        def emit(self, record: logging.LogRecord) -> None:
-            if record.levelno == logging.WARNING:
-                # Use Rich console to print dim warning
-                msg = record.getMessage()
-                console.print(f"[dim yellow]\u26a0\ufe0f  Warning:[/dim yellow] [dim]{msg}[/dim]")
-            else:
-                super().emit(record)
-
-    # Configure root logger to use our handler for WARNING and above
-    handler = DimWarningHandler(console=console, show_time=False, show_path=False, show_level=False)
-    handler.setLevel(logging.WARNING)
-
-    # Apply to root logger (catches all loggers including deepagents)
-    root_logger = logging.getLogger()
-    # Remove existing handlers to avoid duplicate output
-    for h in root_logger.handlers[:]:
-        root_logger.removeHandler(h)
-    root_logger.addHandler(handler)
-    root_logger.setLevel(logging.WARNING)
-
-
-def main():
-    """CLI entry point — delegates to the Typer app."""
-    import warnings
-    warnings.filterwarnings("ignore", message=".*not known to support tools.*")
-    warnings.filterwarnings("ignore", message=".*type is unknown and inference may fail.*")
-    _configure_logging()
-    app()
+        # Default: interactive mode
+        cmd_interactive(agent, show_thinking=show_thinking, workspace_dir=workspace_dir)
 
 
 if __name__ == "__main__":