stravinsky 0.2.40__py3-none-any.whl → 0.3.4__py3-none-any.whl

mcp_bridge/cli/session_report.py

@@ -0,0 +1,585 @@
+#!/usr/bin/env python3
+"""
+Session Report CLI
+
+A rich CLI tool for viewing Claude Code sessions with tool/agent/model summaries.
+Optionally uses Gemini for session summarization.
+"""
+
+import argparse
+import json
+import os
+import re
+import sys
+from collections import Counter
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+
+from rich.console import Console
+from rich.panel import Panel
+from rich.progress import Progress, SpinnerColumn, TextColumn
+from rich.table import Table
+from rich.text import Text
+from rich.tree import Tree
+
+
+console = Console()
+
+
+def get_sessions_directory() -> Path:
+    """Get the Claude sessions directory."""
+    return Path.home() / ".claude" / "projects"
+
+
+def get_sessions(limit: int = 10, search_id: str | None = None) -> list[dict]:
+    """Get recent sessions sorted by modification time.
+
+    If search_id is provided, searches ALL sessions (ignores limit).
+    """
+    sessions_dir = get_sessions_directory()
+    if not sessions_dir.exists():
+        return []
+
+    sessions = []
+    for project_dir in sessions_dir.iterdir():
+        if not project_dir.is_dir():
+            continue
+
+        for session_file in project_dir.glob("*.jsonl"):
+            try:
+                stat = session_file.stat()
+                mtime = datetime.fromtimestamp(stat.st_mtime)
+                sessions.append({
+                    "id": session_file.stem,
+                    "path": str(session_file),
+                    "project": project_dir.name,
+                    "modified": mtime,
+                    "size": stat.st_size,
+                })
+            except Exception:
+                continue
+
+    sessions.sort(key=lambda s: s["modified"], reverse=True)
+
+    # If searching by ID, don't limit
+    if search_id:
+        return sessions
+
+    return sessions[:limit]
+
+
+def read_session_messages(session_path: str) -> list[dict]:
+    """Read all messages from a session file."""
+    messages = []
+    try:
+        with open(session_path) as f:
+            for line in f:
+                if line.strip():
+                    try:
+                        msg = json.loads(line)
+                        messages.append(msg)
+                    except json.JSONDecodeError:
+                        continue
+    except Exception:
+        pass
+    return messages
+
+
+def extract_tool_usage(messages: list[dict]) -> dict[str, Any]:
+    """Extract tool, agent, and model usage from session messages.
+
+    Claude Code session format:
+    - Top-level: {type: "user"|"assistant", message: {role, content, model?}}
+    - content can be string or list of {type: "text"|"tool_use"|"tool_result", ...}
+    - tool_use has: {type: "tool_use", name: "ToolName", input: {...}}
+
+    Captures:
+    1. Subagents spawned (Task tool with subagent_type, MCP agent_spawn)
+    2. External models invoked (invoke_gemini, invoke_openai with model param)
+    3. All tools used (native and MCP)
+    4. LSP tools specifically
+    """
+    # Native Claude tools (Read, Write, Edit, Bash, etc.)
+    native_tools = Counter()
+    # MCP tools by server (stravinsky, github, grep-app, etc.)
+    mcp_tools_by_server: dict[str, Counter] = {}
+    # Subagents from Task tool
+    subagents = Counter()
+    # MCP agents from agent_spawn
+    mcp_agents = Counter()
+    # Claude model used for responses
+    claude_models = Counter()
+    # External models invoked (gemini, openai)
+    external_models = Counter()
+    # LSP tools specifically
+    lsp_tools = Counter()
+
+    for msg in messages:
+        msg_type = msg.get("type", "")
+        inner_msg = msg.get("message", {})
+
+        # Skip non-message types (snapshots, etc.)
+        if msg_type not in ("user", "assistant"):
+            continue
+
+        # Extract Claude model from assistant messages
+        model = inner_msg.get("model", "")
+        if model and msg_type == "assistant":
+            claude_models[model] += 1
+
+        content = inner_msg.get("content", "")
+
+        # Handle content as list (tool_use, text blocks, etc.)
+        if isinstance(content, list):
+            for block in content:
+                if not isinstance(block, dict):
+                    continue
+
+                block_type = block.get("type", "")
+
+                # Tool use blocks
+                if block_type == "tool_use":
+                    tool_name = block.get("name", "unknown")
+                    tool_input = block.get("input", {})
+
+                    # MCP tools (format: mcp__server__tool)
+                    if tool_name.startswith("mcp__"):
+                        parts = tool_name.split("__")
+                        if len(parts) >= 3:
+                            server = parts[1]
+                            tool = parts[2]
+
+                            # Track by server
+                            if server not in mcp_tools_by_server:
+                                mcp_tools_by_server[server] = Counter()
+                            mcp_tools_by_server[server][tool] += 1
+
+                            # Stravinsky-specific: agent_spawn
+                            if server == "stravinsky" and tool == "agent_spawn":
+                                agent_type = tool_input.get("agent_type", "explore")
+                                model_used = tool_input.get("model", "gemini-3-flash")
+                                mcp_agents[f"{agent_type} ({model_used})"] += 1
+
+                            # Stravinsky-specific: invoke_gemini
+                            if server == "stravinsky" and tool == "invoke_gemini":
+                                model_used = tool_input.get("model", "gemini-3-flash")
+                                external_models[f"gemini:{model_used}"] += 1
+
+                            # Stravinsky-specific: invoke_openai
+                            if server == "stravinsky" and tool == "invoke_openai":
+                                model_used = tool_input.get("model", "gpt-5.2-codex")
+                                external_models[f"openai:{model_used}"] += 1
+
+                            # LSP tools
+                            if server == "stravinsky" and tool.startswith("lsp_"):
+                                lsp_tools[tool] += 1
+
+                    # Native Task tool (spawns subagents)
+                    elif tool_name == "Task":
+                        subagent = tool_input.get("subagent_type", "")
+                        if subagent:
+                            subagents[subagent] += 1
+
+                    # Other native tools
+                    else:
+                        native_tools[tool_name] += 1
+
+    # Build categorized MCP tools dict
+    mcp_tools_flat = {}
+    for server, tools in mcp_tools_by_server.items():
+        for tool, count in tools.items():
+            mcp_tools_flat[f"{server}:{tool}"] = count
+
+    return {
+        "native_tools": dict(native_tools),
+        "mcp_tools": mcp_tools_flat,
+        "mcp_tools_by_server": {s: dict(t) for s, t in mcp_tools_by_server.items()},
+        "subagents": dict(subagents),
+        "mcp_agents": dict(mcp_agents),
+        "claude_models": dict(claude_models),
+        "external_models": dict(external_models),
+        "lsp_tools": dict(lsp_tools),
+    }
+
+
+def format_size(size_bytes: int) -> str:
+    """Format bytes to human readable size."""
+    if size_bytes < 1024:
+        return f"{size_bytes} B"
+    elif size_bytes < 1024 * 1024:
+        return f"{size_bytes / 1024:.1f} KB"
+    else:
+        return f"{size_bytes / (1024 * 1024):.1f} MB"
+
+
+def display_session_list(sessions: list[dict]) -> None:
+    """Display sessions in a rich table."""
+    table = Table(title="Recent Claude Code Sessions", show_header=True, header_style="bold magenta")
+    table.add_column("#", style="dim", width=3)
+    table.add_column("Session ID", style="cyan", width=15)
+    table.add_column("Modified", style="green", width=20)
+    table.add_column("Size", justify="right", style="yellow", width=10)
+    table.add_column("Project Hash", style="dim", width=12)
+
+    for i, session in enumerate(sessions, 1):
+        table.add_row(
+            str(i),
+            session["id"][:12] + "...",
+            session["modified"].strftime("%Y-%m-%d %H:%M"),
+            format_size(session["size"]),
+            session["project"][:10] + "...",
+        )
+
+    console.print(table)
+
+
+def extract_hooks(messages: list[dict]) -> dict[str, int]:
+    """Extract hooks triggered from session messages.
+
+    Hooks appear in:
+    - system-reminder tags in user messages
+    - tool_result content with hook output
+    """
+    hooks = Counter()
+
+    for msg in messages:
+        inner_msg = msg.get("message", {})
+        content = inner_msg.get("content", "")
+
+        # Check all content - could be string or list
+        texts_to_check = []
+
+        if isinstance(content, str):
+            texts_to_check.append(content)
+        elif isinstance(content, list):
+            for block in content:
+                if isinstance(block, dict):
+                    # Text blocks
+                    if block.get("type") == "text":
+                        texts_to_check.append(block.get("text", ""))
+                    # Tool result blocks (hooks often appear here)
+                    elif block.get("type") == "tool_result":
+                        result_content = block.get("content", "")
+                        if isinstance(result_content, str):
+                            texts_to_check.append(result_content)
+
+        # Check all collected texts for hook patterns
+        for text in texts_to_check:
+            if "UserPromptSubmit hook" in text:
+                hooks["UserPromptSubmit"] += 1
+            if "PreToolUse hook" in text:
+                hooks["PreToolUse"] += 1
+            if "PostToolUse hook" in text:
+                hooks["PostToolUse"] += 1
+            if "<system-reminder>" in text:
+                # Count system-reminder injections
+                hooks["system-reminder"] += text.count("<system-reminder>")
+
+    return dict(hooks)
+
+
+def display_session_details(session: dict, usage: dict[str, Any], messages: list[dict]) -> None:
+    """Display detailed session information with rich formatting."""
+    # Session header
+    header = Text()
+    header.append("Session: ", style="bold")
+    header.append(session["id"], style="cyan")
+    console.print(Panel(header, title="Session Details", border_style="blue"))
+
+    # Statistics
+    stats_table = Table(show_header=False, box=None)
+    stats_table.add_column("Key", style="bold")
+    stats_table.add_column("Value")
+    stats_table.add_row("Path", session["path"])
+    stats_table.add_row("Modified", session["modified"].strftime("%Y-%m-%d %H:%M:%S"))
+    stats_table.add_row("Size", format_size(session["size"]))
+    stats_table.add_row("Messages", str(len(messages)))
+
+    # Count roles (Claude Code format: type field at top level)
+    user_msgs = sum(1 for m in messages if m.get("type") == "user")
+    assistant_msgs = sum(1 for m in messages if m.get("type") == "assistant")
+    stats_table.add_row("User Messages", str(user_msgs))
+    stats_table.add_row("Assistant Messages", str(assistant_msgs))
+
+    console.print(Panel(stats_table, title="Statistics", border_style="green"))
+
+    # Claude Models (the model Claude Code uses)
+    if usage.get("claude_models"):
+        models_tree = Tree("[bold yellow]Claude Models")
+        for model, count in sorted(usage["claude_models"].items(), key=lambda x: -x[1]):
+            models_tree.add(f"{model}: [cyan]{count}[/cyan] responses")
+        console.print(Panel(models_tree, title="Claude Model", border_style="yellow"))
+
+    # External Models (gemini, openai invoked via MCP)
+    if usage.get("external_models"):
+        ext_tree = Tree("[bold green]External Models Invoked")
+        for model, count in sorted(usage["external_models"].items(), key=lambda x: -x[1]):
+            ext_tree.add(f"{model}: [cyan]{count}[/cyan] calls")
+        console.print(Panel(ext_tree, title="External Models (Gemini/OpenAI)", border_style="green"))
+
+    # Native tools
+    if usage.get("native_tools"):
+        tools_tree = Tree("[bold magenta]Native Tools")
+        for tool, count in sorted(usage["native_tools"].items(), key=lambda x: -x[1]):
+            tools_tree.add(f"{tool}: [yellow]{count}[/yellow] calls")
+        console.print(Panel(tools_tree, title="Native Tool Usage", border_style="magenta"))
+
+    # MCP tools by server
+    if usage.get("mcp_tools_by_server"):
+        for server, tools in sorted(usage["mcp_tools_by_server"].items()):
+            server_tree = Tree(f"[bold cyan]{server}")
+            for tool, count in sorted(tools.items(), key=lambda x: -x[1]):
+                server_tree.add(f"{tool}: [yellow]{count}[/yellow] calls")
+            console.print(Panel(server_tree, title=f"MCP: {server}", border_style="cyan"))
+
+    # Subagents (Task tool)
+    if usage.get("subagents"):
+        subagents_tree = Tree("[bold blue]Subagents (Task tool)")
+        for agent, count in sorted(usage["subagents"].items(), key=lambda x: -x[1]):
+            subagents_tree.add(f"{agent}: [yellow]{count}[/yellow] spawned")
+        console.print(Panel(subagents_tree, title="Subagent Usage", border_style="blue"))
+
+    # MCP Agents (agent_spawn)
+    if usage.get("mcp_agents"):
+        agents_tree = Tree("[bold cyan]MCP Agents (agent_spawn)")
+        for agent, count in sorted(usage["mcp_agents"].items(), key=lambda x: -x[1]):
+            agents_tree.add(f"{agent}: [yellow]{count}[/yellow] spawned")
+        console.print(Panel(agents_tree, title="MCP Agent Usage", border_style="cyan"))
+
+    # LSP Tools
+    if usage.get("lsp_tools"):
+        lsp_tree = Tree("[bold red]LSP Tools")
+        for tool, count in sorted(usage["lsp_tools"].items(), key=lambda x: -x[1]):
+            lsp_tree.add(f"{tool}: [yellow]{count}[/yellow] calls")
+        console.print(Panel(lsp_tree, title="LSP Usage", border_style="red"))
+
+    # Hooks
+    hooks = extract_hooks(messages)
+    if hooks:
+        hooks_tree = Tree("[bold white]Hooks Triggered")
+        for hook, count in sorted(hooks.items(), key=lambda x: -x[1]):
+            hooks_tree.add(f"{hook}: [yellow]{count}[/yellow] times")
+        console.print(Panel(hooks_tree, title="Hooks", border_style="white"))
+
+
+def summarize_with_gemini(session: dict, messages: list[dict], usage: dict[str, Any]) -> str | None:
+    """Use Gemini to summarize the session."""
+    try:
+        from mcp_bridge.tools.model_invoke import invoke_gemini
+    except ImportError:
+        console.print("[red]Error: Could not import invoke_gemini[/red]")
+        return None
+
+    # Build context for Gemini
+    user_msgs = sum(1 for m in messages if m.get("type") == "user")
+    assistant_msgs = sum(1 for m in messages if m.get("type") == "assistant")
+
+    context_parts = [
+        "# Claude Code Session Analysis Request\n\n",
+        f"**Session ID:** {session['id']}\n",
+        f"**Modified:** {session['modified']}\n",
+        f"**Total Messages:** {len(messages)} ({user_msgs} user, {assistant_msgs} assistant)\n\n",
+        "## Tool/Agent/Model Usage Summary\n\n",
+        f"**Tools Used:** {json.dumps(usage['tools'], indent=2)}\n\n",
+        f"**MCP Tools:** {json.dumps(usage['mcp_tools'], indent=2)}\n\n",
+        f"**Subagents (Task):** {usage.get('subagents', {})}\n\n",
+        f"**MCP Agents:** {usage['agents']}\n\n",
+        f"**Models:** {usage['models']}\n\n",
+        "## Session Transcript\n\n",
+    ]
+
+    # Add messages (Gemini has 1M context window)
+    # Parse Claude Code session format
+    for i, msg in enumerate(messages):
+        msg_type = msg.get("type", "")
+        if msg_type not in ("user", "assistant"):
+            continue
+
+        inner_msg = msg.get("message", {})
+        role = inner_msg.get("role", msg_type)
+        content = inner_msg.get("content", "")
+
+        # Extract text from content blocks
+        if isinstance(content, list):
+            text_parts = []
+            for block in content:
+                if isinstance(block, dict):
+                    if block.get("type") == "text":
+                        text_parts.append(block.get("text", ""))
+                    elif block.get("type") == "tool_use":
+                        tool_name = block.get("name", "unknown")
+                        text_parts.append(f"[TOOL: {tool_name}]")
+                    elif block.get("type") == "thinking":
+                        text_parts.append("[THINKING]")
+            content = " ".join(text_parts)
+
+        # Truncate very long messages
+        if len(content) > 2000:
+            content = content[:2000] + "... [truncated]"
+
+        context_parts.append(f"**[{i+1}] {role}:** {content}\n\n")
+
+    context = "".join(context_parts)
+
+    prompt = f"""Analyze this Claude Code session and provide a concise summary:
+
+{context}
+
+Please provide:
+1. **Session Purpose**: What was the user trying to accomplish?
+2. **Key Actions**: Main tools, agents, and operations used
+3. **Outcome**: Was the task successful? Any notable issues?
+4. **Recommendations**: Any suggestions for improvement?
+
+Keep the summary concise but informative."""
+
+    with Progress(
+        SpinnerColumn(),
+        TextColumn("[progress.description]{task.description}"),
+        console=console,
+    ) as progress:
+        progress.add_task(description="Analyzing session with Gemini...", total=None)
+        try:
+            # Run synchronously
+            import asyncio
+            result = asyncio.run(invoke_gemini(
+                prompt=prompt,
+                model="gemini-3-flash",
+                max_tokens=2048,
+            ))
+            return result
+        except Exception as e:
+            console.print(f"[red]Gemini error: {e}[/red]")
+            return None
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Analyze Claude Code sessions with rich formatting",
+        prog="stravinsky-sessions",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+Examples:
+  stravinsky-sessions                    # List 10 recent sessions
+  stravinsky-sessions --limit 20         # List 20 recent sessions
+  stravinsky-sessions --select 1         # Show details for session #1
+  stravinsky-sessions --select 1 --summarize  # Summarize with Gemini
+  stravinsky-sessions --id abc123        # Show session by ID prefix
+""",
+    )
+
+    parser.add_argument(
+        "--limit", "-n",
+        type=int,
+        default=10,
+        help="Number of sessions to list (default: 10)",
+    )
+    parser.add_argument(
+        "--select", "-s",
+        type=int,
+        help="Select session by number from list (1-indexed)",
+    )
+    parser.add_argument(
+        "--id",
+        type=str,
+        help="Select session by ID or ID prefix",
+    )
+    parser.add_argument(
+        "--summarize",
+        action="store_true",
+        help="Use Gemini to summarize the session",
+    )
+    parser.add_argument(
+        "--json",
+        action="store_true",
+        help="Output as JSON instead of rich formatting",
+    )
+
+    args = parser.parse_args()
+
+    # Get sessions (search all if --id is specified)
+    sessions = get_sessions(limit=args.limit, search_id=args.id)
+
+    if not sessions:
+        console.print("[yellow]No sessions found[/yellow]")
+        return 1
+
+    # Select session by number or ID
+    selected_session = None
+
+    if args.select:
+        # For --select, re-get with limit
+        display_sessions = get_sessions(limit=args.limit)
+        if 1 <= args.select <= len(display_sessions):
+            selected_session = display_sessions[args.select - 1]
+        else:
+            console.print(f"[red]Invalid selection: {args.select}. Must be 1-{len(display_sessions)}[/red]")
+            return 1
+    elif args.id:
+        for s in sessions:
+            if s["id"].startswith(args.id):
+                selected_session = s
+                break
+        if not selected_session:
+            console.print(f"[red]Session not found: {args.id}[/red]")
+            return 1
+
+    if selected_session:
+        # Read and analyze session
+        messages = read_session_messages(selected_session["path"])
+        usage = extract_tool_usage(messages)
+
+        if args.json:
+            user_msgs = sum(1 for m in messages if m.get("type") == "user")
+            assistant_msgs = sum(1 for m in messages if m.get("type") == "assistant")
+            hooks = extract_hooks(messages)
+            output = {
+                "session": {
+                    "id": selected_session["id"],
+                    "path": selected_session["path"],
+                    "modified": selected_session["modified"].isoformat(),
+                    "size": selected_session["size"],
+                    "message_count": len(messages),
+                    "user_messages": user_msgs,
+                    "assistant_messages": assistant_msgs,
+                },
+                "claude_models": usage.get("claude_models", {}),
+                "external_models": usage.get("external_models", {}),
+                "native_tools": usage.get("native_tools", {}),
+                "mcp_tools_by_server": usage.get("mcp_tools_by_server", {}),
+                "subagents": usage.get("subagents", {}),
+                "mcp_agents": usage.get("mcp_agents", {}),
+                "lsp_tools": usage.get("lsp_tools", {}),
+                "hooks": hooks,
+            }
+            print(json.dumps(output, indent=2))
+        else:
+            display_session_details(selected_session, usage, messages)
+
+        if args.summarize:
+            console.print()
+            summary = summarize_with_gemini(selected_session, messages, usage)
+            if summary:
+                console.print(Panel(summary, title="Gemini Summary", border_style="green"))
+    else:
+        # List sessions
+        if args.json:
+            output = [
+                {
+                    "id": s["id"],
+                    "modified": s["modified"].isoformat(),
+                    "size": s["size"],
+                }
+                for s in sessions
+            ]
+            print(json.dumps(output, indent=2))
+        else:
+            display_session_list(sessions)
+            console.print("\n[dim]Use --select N or --id PREFIX to view session details[/dim]")
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())