moai-adk 0.8.0__py3-none-any.whl → 1.1.0__py3-none-any.whl

moai_adk/rank/hook.py

@@ -0,0 +1,1503 @@
+"""Session end hook for automatic token usage submission.
+
+This module provides a hook that is called when a Claude Code session ends,
+automatically submitting the session's token usage to the MoAI Rank service.
+
+The hook is installed globally at ~/.claude/hooks/moai/ by default to collect
+session data from all projects. Users can opt-out specific projects via
+~/.moai/rank/config.yaml configuration.
+
+Token Usage Collection:
+    Claude Code's SessionEnd hook only provides metadata (session_id, transcript_path,
+    cwd, etc.) - NOT token counts. Token usage must be extracted from the transcript
+    JSONL file by parsing message.usage fields.
+
+Cost Calculation:
+    Based on Anthropic's pricing (as of 2024):
+    - Claude 3.5 Sonnet: Input $3/MTok, Output $15/MTok
+    - Claude 3 Opus: Input $15/MTok, Output $75/MTok
+    - Cache Creation: $3.75/MTok (Sonnet), $18.75/MTok (Opus)
+    - Cache Read: $0.30/MTok (Sonnet), $1.50/MTok (Opus)
+"""
+
+import hashlib
+import os
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any, Optional
+
+from moai_adk.rank.client import RankClient, RankClientError, SessionSubmission
+from moai_adk.rank.config import RankConfig
+
+# Token pricing per million tokens (USD)
+# Source: https://platform.claude.com/docs/en/about-claude/pricing
+# Last updated: 2025-01
+MODEL_PRICING: dict[str, dict[str, float]] = {
+    # Claude Opus 4.5 - Latest flagship model
+    "claude-opus-4-5-20251101": {
+        "input": 5.00,
+        "output": 25.00,
+        "cache_creation": 6.25,  # 5m cache writes
+        "cache_read": 0.50,
+    },
+    # Claude Opus 4.1
+    "claude-opus-4-1-20250414": {
+        "input": 15.00,
+        "output": 75.00,
+        "cache_creation": 18.75,
+        "cache_read": 1.50,
+    },
+    # Claude Opus 4
+    "claude-opus-4-20250514": {
+        "input": 15.00,
+        "output": 75.00,
+        "cache_creation": 18.75,
+        "cache_read": 1.50,
+    },
+    # Claude Sonnet 4.5
+    "claude-sonnet-4-5-20251022": {
+        "input": 3.00,
+        "output": 15.00,
+        "cache_creation": 3.75,
+        "cache_read": 0.30,
+    },
+    # Claude Sonnet 4
+    "claude-sonnet-4-20250514": {
+        "input": 3.00,
+        "output": 15.00,
+        "cache_creation": 3.75,
+        "cache_read": 0.30,
+    },
+    # Claude Sonnet 3.7 (deprecated but still supported)
+    "claude-3-7-sonnet-20250219": {
+        "input": 3.00,
+        "output": 15.00,
+        "cache_creation": 3.75,
+        "cache_read": 0.30,
+    },
+    # Claude Haiku 4.5
+    "claude-haiku-4-5-20251022": {
+        "input": 1.00,
+        "output": 5.00,
+        "cache_creation": 1.25,
+        "cache_read": 0.10,
+    },
+    # Claude Haiku 3.5
+    "claude-3-5-haiku-20241022": {
+        "input": 0.80,
+        "output": 4.00,
+        "cache_creation": 1.00,
+        "cache_read": 0.08,
+    },
+    # Claude Opus 3 (deprecated)
+    "claude-3-opus-20240229": {
+        "input": 15.00,
+        "output": 75.00,
+        "cache_creation": 18.75,
+        "cache_read": 1.50,
+    },
+    # Claude Haiku 3
+    "claude-3-haiku-20240307": {
+        "input": 0.25,
+        "output": 1.25,
+        "cache_creation": 0.30,
+        "cache_read": 0.03,
+    },
+    # Legacy Claude 3.5 Sonnet versions
+    "claude-3-5-sonnet-20241022": {
+        "input": 3.00,
+        "output": 15.00,
+        "cache_creation": 3.75,
+        "cache_read": 0.30,
+    },
+    "claude-3-5-sonnet-20240620": {
+        "input": 3.00,
+        "output": 15.00,
+        "cache_creation": 3.75,
+        "cache_read": 0.30,
+    },
+    # Default fallback (Sonnet 4 pricing)
+    "default": {
+        "input": 3.00,
+        "output": 15.00,
+        "cache_creation": 3.75,
+        "cache_read": 0.30,
+    },
+}
+
+
+@dataclass
+class TokenUsage:
+    """Token usage extracted from a session transcript.
+
+    Attributes:
+        input_tokens: Total input tokens consumed
+        output_tokens: Total output tokens generated
+        cache_creation_tokens: Tokens used for cache creation
+        cache_read_tokens: Tokens read from cache
+        model_name: Primary model used in the session
+        cost_usd: Calculated cost in USD
+
+        # Dashboard fields (for activity visualization)
+        started_at: Session start timestamp (UTC ISO format)
+        duration_seconds: Total session duration in seconds
+        turn_count: Number of user turns (messages)
+        tool_usage: Tool usage counts (e.g., {"Read": 5, "Write": 3})
+        model_usage: Per-model token usage (e.g., {"claude-opus-4-5": {"input": 5000, "output": 2000}})
+        code_metrics: Code change metrics (linesAdded, linesDeleted, filesModified, filesCreated)
+    """
+
+    input_tokens: int = 0
+    output_tokens: int = 0
+    cache_creation_tokens: int = 0
+    cache_read_tokens: int = 0
+    model_name: Optional[str] = None
+    cost_usd: float = 0.0
+
+    # Dashboard fields for activity visualization
+    started_at: Optional[str] = None
+    duration_seconds: int = 0
+    turn_count: int = 0
+    tool_usage: Optional[dict[str, int]] = None
+    model_usage: Optional[dict[str, dict[str, int]]] = None
+    code_metrics: Optional[dict[str, int]] = None
+
+
+def compute_anonymous_project_id(project_path: str) -> str:
+    """Compute an anonymized project identifier.
+
+    Uses a hash of the project path to track sessions by project
+    without revealing the actual project name or path.
+
+    Args:
+        project_path: Full path to the project directory
+
+    Returns:
+        First 16 characters of the SHA-256 hash
+    """
+    # Normalize the path
+    normalized = os.path.normpath(os.path.expanduser(project_path))
+    # Hash it
+    hash_value = hashlib.sha256(normalized.encode()).hexdigest()
+    return hash_value[:16]
+
+
+def get_model_pricing(model_name: Optional[str]) -> dict[str, float]:
+    """Get pricing for a specific model.
+
+    Uses exact match first, then falls back to pattern matching
+    for model families (opus, sonnet, haiku).
+
+    Args:
+        model_name: Model identifier string (e.g., 'claude-sonnet-4-20250514')
+
+    Returns:
+        Dictionary with pricing per million tokens:
+        - input: Base input token price
+        - output: Output token price
+        - cache_creation: 5-minute cache write price
+        - cache_read: Cache hit/refresh price
+    """
+    if not model_name:
+        return MODEL_PRICING["default"]
+
+    # Try exact match first
+    if model_name in MODEL_PRICING:
+        return MODEL_PRICING[model_name]
+
+    # Try pattern matching for model families
+    model_lower = model_name.lower()
+
+    # Opus family - check version specifics
+    if "opus" in model_lower:
+        if "4-5" in model_lower or "4.5" in model_lower:
+            return MODEL_PRICING["claude-opus-4-5-20251101"]
+        elif "4-1" in model_lower or "4.1" in model_lower:
+            return MODEL_PRICING["claude-opus-4-1-20250414"]
+        elif "opus-4" in model_lower or "opus4" in model_lower:
+            return MODEL_PRICING["claude-opus-4-20250514"]
+        elif "opus-3" in model_lower or "opus3" in model_lower:
+            return MODEL_PRICING["claude-3-opus-20240229"]
+        # Default to Opus 4 pricing for unknown opus versions
+        return MODEL_PRICING["claude-opus-4-20250514"]
+
+    # Haiku family
+    elif "haiku" in model_lower:
+        if "4-5" in model_lower or "4.5" in model_lower:
+            return MODEL_PRICING["claude-haiku-4-5-20251022"]
+        elif "3-5" in model_lower or "3.5" in model_lower:
+            return MODEL_PRICING["claude-3-5-haiku-20241022"]
+        elif "haiku-3" in model_lower or "haiku3" in model_lower:
+            return MODEL_PRICING["claude-3-haiku-20240307"]
+        # Default to Haiku 3.5 pricing
+        return MODEL_PRICING["claude-3-5-haiku-20241022"]
+
+    # Sonnet family
+    elif "sonnet" in model_lower:
+        if "4-5" in model_lower or "4.5" in model_lower:
+            return MODEL_PRICING["claude-sonnet-4-5-20251022"]
+        elif "sonnet-4" in model_lower or "sonnet4" in model_lower:
+            return MODEL_PRICING["claude-sonnet-4-20250514"]
+        elif "3-7" in model_lower or "3.7" in model_lower:
+            return MODEL_PRICING["claude-3-7-sonnet-20250219"]
+        elif "3-5" in model_lower or "3.5" in model_lower:
+            return MODEL_PRICING["claude-3-5-sonnet-20241022"]
+        # Default to Sonnet 4 pricing
+        return MODEL_PRICING["claude-sonnet-4-20250514"]
+
+    return MODEL_PRICING["default"]
+
+
+def calculate_cost(usage: TokenUsage) -> float:
+    """Calculate the cost in USD for token usage.
+
+    Args:
+        usage: TokenUsage dataclass with token counts and model
+
+    Returns:
+        Cost in USD
+    """
+    pricing = get_model_pricing(usage.model_name)
+
+    # Cost per token (pricing is per million tokens)
+    input_cost = (usage.input_tokens / 1_000_000) * pricing["input"]
+    output_cost = (usage.output_tokens / 1_000_000) * pricing["output"]
+    cache_creation_cost = (usage.cache_creation_tokens / 1_000_000) * pricing["cache_creation"]
+    cache_read_cost = (usage.cache_read_tokens / 1_000_000) * pricing["cache_read"]
+
+    return input_cost + output_cost + cache_creation_cost + cache_read_cost
+
+
+def parse_transcript_for_usage(transcript_path: str) -> Optional[TokenUsage]:
+    """Parse a JSONL transcript file to extract total token usage and session metadata.
+
+    Claude Code stores conversation transcripts as JSONL files where each line
+    contains a message object. Token usage is in the message.usage field.
+
+    Extracts:
+        - Token counts: input, output, cache creation, cache read
+        - Model name: Primary model used
+        - Session timing: started_at, duration_seconds
+        - User engagement: turn_count (number of user messages)
+        - Tool usage: Count of each tool used (e.g., {"Read": 5, "Write": 3})
+
+    Args:
+        transcript_path: Path to the transcript JSONL file
+
+    Returns:
+        TokenUsage dataclass with aggregated data, or None if parsing fails
+    """
+    import json
+    from datetime import datetime as dt
+
+    usage = TokenUsage()
+    transcript_file = Path(transcript_path)
+
+    if not transcript_file.exists():
+        return None
+
+    first_timestamp: Optional[str] = None
+    last_timestamp: Optional[str] = None
+    turn_count = 0
+    tool_counts: dict[str, int] = {}
+    model_counts: dict[str, dict[str, int]] = {}
+    current_model: Optional[str] = None
+
+    # Code metrics tracking
+    lines_added = 0
+    lines_deleted = 0
+    files_modified: set[str] = set()
+    files_created: set[str] = set()
+
+    def count_lines(text: str) -> int:
+        """Count non-empty lines in text."""
+        if not text:
+            return 0
+        return len([line for line in text.split("\n") if line.strip()])
+
+    try:
+        with open(transcript_file, encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if not line:
+                    continue
+
+                try:
+                    data = json.loads(line)
+                except json.JSONDecodeError:
+                    continue
+
+                # Extract timestamp for session timing
+                timestamp = data.get("timestamp")
+                if timestamp:
+                    if first_timestamp is None:
+                        first_timestamp = timestamp
+                    last_timestamp = timestamp
+
+                # Count user turns
+                msg_type = data.get("type")
+                if msg_type == "user":
+                    turn_count += 1
+
+                # Extract usage from message.usage field
+                message = data.get("message", {})
+                msg_usage = message.get("usage", {})
+
+                # Extract model name first (needed for per-model tracking)
+                model = message.get("model") or data.get("model")
+                if model:
+                    current_model = model
+                    usage.model_name = model
+
+                if msg_usage:
+                    input_toks = msg_usage.get("input_tokens", 0)
+                    output_toks = msg_usage.get("output_tokens", 0)
+
+                    # Aggregate total tokens
+                    usage.input_tokens += input_toks
+                    usage.output_tokens += output_toks
+                    usage.cache_creation_tokens += msg_usage.get("cache_creation_input_tokens", 0)
+                    usage.cache_read_tokens += msg_usage.get("cache_read_input_tokens", 0)
+
+                    # Track per-model token usage
+                    if current_model:
+                        if current_model not in model_counts:
+                            model_counts[current_model] = {"input": 0, "output": 0}
+                        model_counts[current_model]["input"] += input_toks
+                        model_counts[current_model]["output"] += output_toks
+
+                # Count tool usage and extract code metrics from message.content
+                content = message.get("content", [])
+                if isinstance(content, list):
+                    for block in content:
+                        if isinstance(block, dict) and block.get("type") == "tool_use":
+                            tool_name = block.get("name", "unknown")
+                            tool_counts[tool_name] = tool_counts.get(tool_name, 0) + 1
+
+                            # Extract code metrics from Edit/Write tools
+                            tool_input = block.get("input", {})
+                            if isinstance(tool_input, dict):
+                                file_path = tool_input.get("file_path", "")
+
+                                if tool_name == "Edit":
+                                    # Edit: count old_string (deleted) and new_string (added)
+                                    old_str = tool_input.get("old_string", "")
+                                    new_str = tool_input.get("new_string", "")
+                                    lines_deleted += count_lines(old_str)
+                                    lines_added += count_lines(new_str)
+                                    if file_path:
+                                        files_modified.add(file_path)
+
+                                elif tool_name == "Write":
+                                    # Write: count content lines as added
+                                    content_str = tool_input.get("content", "")
+                                    lines_added += count_lines(content_str)
+                                    if file_path:
+                                        files_created.add(file_path)
+
+                                elif tool_name == "MultiEdit":
+                                    # MultiEdit: process multiple edits
+                                    edits = tool_input.get("edits", [])
+                                    if isinstance(edits, list):
+                                        for edit in edits:
+                                            if isinstance(edit, dict):
+                                                old_str = edit.get("old_string", "")
+                                                new_str = edit.get("new_string", "")
+                                                lines_deleted += count_lines(old_str)
+                                                lines_added += count_lines(new_str)
+                                    if file_path:
+                                        files_modified.add(file_path)
+
+        # Set dashboard fields
+        usage.started_at = first_timestamp
+        usage.turn_count = turn_count
+        usage.tool_usage = tool_counts if tool_counts else None
+        usage.model_usage = model_counts if model_counts else None
+
+        # Set code metrics (only if there was any code activity)
+        if lines_added > 0 or lines_deleted > 0 or files_modified or files_created:
+            usage.code_metrics = {
+                "linesAdded": lines_added,
+                "linesDeleted": lines_deleted,
+                "filesModified": len(files_modified),
+                "filesCreated": len(files_created),
+            }
+
+        # Calculate duration in seconds
+        if first_timestamp and last_timestamp:
+            try:
+                # Parse ISO format timestamps
+                start_dt = dt.fromisoformat(first_timestamp.replace("Z", "+00:00"))
+                end_dt = dt.fromisoformat(last_timestamp.replace("Z", "+00:00"))
+                usage.duration_seconds = int((end_dt - start_dt).total_seconds())
+            except (ValueError, TypeError):
+                usage.duration_seconds = 0
+
+        # Calculate cost
+        if usage.input_tokens > 0 or usage.output_tokens > 0:
+            usage.cost_usd = calculate_cost(usage)
+            return usage
+
+        return None
+
+    except (OSError, IOError):
+        return None
+
+
+def parse_session_data(session_data: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Parse and validate session data from Claude Code SessionEnd hook.
+
+    IMPORTANT: Claude Code's SessionEnd hook does NOT provide token counts directly.
+    Token usage must be extracted from the transcript_path JSONL file.
+
+    Args:
+        session_data: Raw session data from Claude Code containing:
+            - session_id: Unique session identifier
+            - transcript_path: Path to the JSONL transcript file
+            - cwd: Current working directory (project path)
+            - permission_mode: Permission mode used
+            - hook_event_name: Name of the hook event
+            - reason: Reason for session end
+
+    Returns:
+        Parsed session data dictionary with token usage, or None if invalid
+    """
+    try:
+        # Get transcript path - this is where token data lives
+        transcript_path = session_data.get("transcript_path")
+        if not transcript_path:
+            return None
+
+        # Parse the transcript file for token usage
+        usage = parse_transcript_for_usage(transcript_path)
+        if not usage:
+            return None
+
+        # Skip sessions with no token usage
+        if usage.input_tokens == 0 and usage.output_tokens == 0:
+            return None
+
+        # Get project path from cwd
+        project_path = session_data.get("cwd")
+
+        # Generate ended_at timestamp
+        from datetime import datetime as dt
+        from datetime import timezone
+
+        ended_at = dt.now(timezone.utc).isoformat().replace("+00:00", "Z")
+
+        return {
+            "input_tokens": usage.input_tokens,
+            "output_tokens": usage.output_tokens,
+            "cache_creation_tokens": usage.cache_creation_tokens,
+            "cache_read_tokens": usage.cache_read_tokens,
+            "model_name": usage.model_name,
+            "project_path": project_path,
+            "ended_at": ended_at,
+            "cost_usd": usage.cost_usd,
+            "session_id": session_data.get("session_id"),
+            # Dashboard fields
+            "started_at": usage.started_at,
+            "duration_seconds": usage.duration_seconds,
+            "turn_count": usage.turn_count,
+            "tool_usage": usage.tool_usage,
+            "model_usage": usage.model_usage,
+            "code_metrics": usage.code_metrics,
+        }
+    except (KeyError, ValueError, TypeError):
+        return None
+
+
+def parse_transcript_to_submission(
+    transcript_path: Path,
+) -> Optional[SessionSubmission]:
+    """Parse a transcript file and return a SessionSubmission object.
+
+    This function is used for batch processing - it parses the transcript
+    without submitting, allowing multiple sessions to be batched together.
+
+    Args:
+        transcript_path: Path to the JSONL transcript file
+
+    Returns:
+        SessionSubmission object if valid, None if no token usage or error
+    """
+    try:
+        session_data = {
+            "session_id": transcript_path.stem,
+            "transcript_path": str(transcript_path),
+            "cwd": str(transcript_path.parent.parent),
+        }
+
+        parsed = parse_session_data(session_data)
+        if not parsed:
+            return None
+
+        client = RankClient()
+
+        # Compute session hash
+        session_hash = client.compute_session_hash(
+            input_tokens=parsed["input_tokens"],
+            output_tokens=parsed["output_tokens"],
+            cache_creation_tokens=parsed["cache_creation_tokens"],
+            cache_read_tokens=parsed["cache_read_tokens"],
+            model_name=parsed["model_name"],
+            ended_at=parsed["ended_at"],
+        )
+
+        return SessionSubmission(
+            session_hash=session_hash,
+            ended_at=parsed["ended_at"],
+            input_tokens=parsed["input_tokens"],
+            output_tokens=parsed["output_tokens"],
+            cache_creation_tokens=parsed["cache_creation_tokens"],
+            cache_read_tokens=parsed["cache_read_tokens"],
+            model_name=parsed["model_name"],
+            started_at=parsed.get("started_at"),
+            duration_seconds=parsed.get("duration_seconds", 0),
+            turn_count=parsed.get("turn_count", 0),
+            tool_usage=parsed.get("tool_usage"),
+            model_usage=parsed.get("model_usage"),
+            code_metrics=parsed.get("code_metrics"),
+        )
+    except Exception:
+        return None
+
+
+def submit_session_hook(session_data: dict[str, Any]) -> dict[str, Any]:
+    """Hook function to submit session data to MoAI Rank.
+
+    This function is called by the Claude Code hook system when a session ends.
+    It parses the transcript JSONL file to extract token usage and submits it
+    to the rank service.
+
+    Args:
+        session_data: Session data dictionary from Claude Code SessionEnd hook:
+            - session_id: Unique session identifier
+            - transcript_path: Path to the JSONL transcript file (contains token data)
+            - cwd: Current working directory (project path)
+            - permission_mode: Permission mode used
+            - hook_event_name: Name of the hook event
+            - reason: Reason for session end
+
+    Returns:
+        Result dictionary with:
+            - success: Whether submission succeeded
+            - message: Status message
+            - session_id: Session ID from server (if successful)
+            - cost_usd: Calculated cost in USD (if successful)
+    """
+    result = {
+        "success": False,
+        "message": "",
+        "session_id": None,
+        "cost_usd": 0.0,
+    }
+
+    # Check if user has registered
+    if not RankConfig.has_credentials():
+        result["message"] = "Not registered with MoAI Rank"
+        return result
+
+    # Parse session data
+    parsed = parse_session_data(session_data)
+    if not parsed:
+        result["message"] = "No token usage to submit"
+        return result
+
+    try:
+        client = RankClient()
+
+        # Compute session hash
+        session_hash = client.compute_session_hash(
+            input_tokens=parsed["input_tokens"],
+            output_tokens=parsed["output_tokens"],
+            cache_creation_tokens=parsed["cache_creation_tokens"],
+            cache_read_tokens=parsed["cache_read_tokens"],
+            model_name=parsed["model_name"],
+            ended_at=parsed["ended_at"],
+        )
+
+        # Create submission
+        submission = SessionSubmission(
+            session_hash=session_hash,
+            ended_at=parsed["ended_at"],
+            input_tokens=parsed["input_tokens"],
+            output_tokens=parsed["output_tokens"],
+            cache_creation_tokens=parsed["cache_creation_tokens"],
+            cache_read_tokens=parsed["cache_read_tokens"],
+            model_name=parsed["model_name"],
+            # Dashboard fields
+            started_at=parsed.get("started_at"),
+            duration_seconds=parsed.get("duration_seconds", 0),
+            turn_count=parsed.get("turn_count", 0),
+            tool_usage=parsed.get("tool_usage"),
+            model_usage=parsed.get("model_usage"),
+            code_metrics=parsed.get("code_metrics"),
+        )
+
+        # Submit to API
+        response = client.submit_session(submission)
+
+        result["success"] = True
+        result["message"] = response.get("message", "Session submitted")
+        result["session_id"] = response.get("sessionId")
+        result["cost_usd"] = parsed.get("cost_usd", 0.0)
+
+    except RankClientError as e:
+        result["message"] = f"Submission failed: {e}"
+
+    return result
+
+
+def create_hook_script() -> str:
+    """Generate a hook script for Claude Code session end.
+
+    Returns:
+        Python script content for the hook
+    """
+    return '''#!/usr/bin/env python3
+"""MoAI Rank Session Hook
+
+This hook submits Claude Code session token usage to the MoAI Rank service.
+It is triggered automatically when a session ends.
+"""
+
+import json
+import sys
+
+def main():
+    # Read session data from stdin
+    try:
+        input_data = sys.stdin.read()
+        if not input_data:
+            return
+
+        session_data = json.loads(input_data)
+
+        # Lazy import to avoid startup delay
+        from moai_adk.rank.hook import submit_session_hook
+
+        result = submit_session_hook(session_data)
+
+        if result["success"]:
+            print(f"Session submitted to MoAI Rank", file=sys.stderr)
+        elif result["message"] != "Not registered with MoAI Rank":
+            print(f"MoAI Rank: {result['message']}", file=sys.stderr)
+
+    except json.JSONDecodeError:
+        pass
+    except ImportError:
+        # moai-adk not installed, silently skip
+        pass
+    except Exception as e:
+        print(f"MoAI Rank hook error: {e}", file=sys.stderr)
+
+if __name__ == "__main__":
+    main()
+'''
+
+
+def load_rank_config() -> dict[str, Any]:
+    """Load rank configuration from ~/.moai/rank/config.yaml.
+
+    Returns:
+        Configuration dictionary with rank settings
+    """
+    import yaml
+
+    config_file = Path.home() / ".moai" / "rank" / "config.yaml"
+    if not config_file.exists():
+        return {"enabled": True, "exclude_projects": []}
+
+    try:
+        with open(config_file) as f:
+            data = yaml.safe_load(f) or {}
+        rank_config = data.get("rank", {})
+        return {
+            "enabled": rank_config.get("enabled", True),
+            "exclude_projects": rank_config.get("exclude_projects", []),
+        }
+    except (OSError, yaml.YAMLError):
+        return {"enabled": True, "exclude_projects": []}
+
+
+def is_project_excluded(project_path: Optional[str]) -> bool:
+    """Check if a project is excluded from rank submission.
+
+    Args:
+        project_path: Path to the project directory
+
+    Returns:
+        True if project should be excluded from submission
+    """
+    import fnmatch
+
+    if not project_path:
+        return False
+
+    config = load_rank_config()
+
+    # Check if rank is globally disabled
+    if not config.get("enabled", True):
+        return True
+
+    # Normalize the project path
+    normalized_path = os.path.normpath(os.path.expanduser(project_path))
+
+    # Check against exclusion patterns
+    for pattern in config.get("exclude_projects", []):
+        pattern = os.path.expanduser(pattern)
+        # Support both exact paths and wildcard patterns
+        if fnmatch.fnmatch(normalized_path, pattern):
+            return True
+        # Also check if pattern is a prefix (directory match)
+        if normalized_path.startswith(os.path.normpath(pattern)):
+            return True
+
+    return False
+
+
+def save_rank_config(config: dict[str, Any]) -> bool:
+    """Save rank configuration to ~/.moai/rank/config.yaml.
+
+    Args:
+        config: Configuration dictionary with rank settings
+
+    Returns:
+        True if saved successfully
+    """
+    import yaml
+
+    config_dir = Path.home() / ".moai" / "rank"
+    config_file = config_dir / "config.yaml"
+
+    try:
+        config_dir.mkdir(parents=True, exist_ok=True)
+
+        # Load existing config to preserve other settings
+        existing: dict[str, Any] = {}
+        if config_file.exists():
+            with open(config_file) as f:
+                existing = yaml.safe_load(f) or {}
+
+        existing["rank"] = config
+        with open(config_file, "w") as f:
+            yaml.safe_dump(existing, f, default_flow_style=False)
+        return True
+    except (OSError, yaml.YAMLError):
+        return False
+
+
+def add_project_exclusion(project_path: str) -> bool:
+    """Add a project to the exclusion list.
+
+    Args:
+        project_path: Path or pattern to exclude
+
+    Returns:
+        True if added successfully
+    """
+    config = load_rank_config()
+    exclude_list = config.get("exclude_projects", [])
+
+    # Normalize the path
+    normalized = os.path.normpath(os.path.expanduser(project_path))
+
+    if normalized not in exclude_list:
+        exclude_list.append(normalized)
+        config["exclude_projects"] = exclude_list
+        return save_rank_config(config)
+
+    return True  # Already excluded
+
+
+def remove_project_exclusion(project_path: str) -> bool:
+    """Remove a project from the exclusion list.
+
+    Args:
+        project_path: Path or pattern to remove from exclusion
+
+    Returns:
+        True if removed successfully
+    """
+    config = load_rank_config()
+    exclude_list = config.get("exclude_projects", [])
+
+    normalized = os.path.normpath(os.path.expanduser(project_path))
+
+    if normalized in exclude_list:
+        exclude_list.remove(normalized)
+        config["exclude_projects"] = exclude_list
+        return save_rank_config(config)
+
+    return True  # Not in list anyway
+
+
+def create_global_hook_script() -> str:
+    """Generate a global hook script with opt-out support.
+
+    Returns:
+        Python script content for the global hook
+    """
+    return '''#!/usr/bin/env python3
+"""MoAI Rank Session Hook (Global)
+
+This hook submits Claude Code session token usage to the MoAI Rank service.
+It is installed globally at ~/.claude/hooks/moai/ and runs for all projects.
+
+Opt-out: Configure ~/.moai/rank/config.yaml to exclude specific projects:
+    rank:
+      enabled: true
+      exclude_projects:
+        - "/path/to/private-project"
+        - "*/confidential/*"
+"""
+
+import json
+import sys
+
+def main():
+    try:
+        input_data = sys.stdin.read()
+        if not input_data:
+            return
+
+        session_data = json.loads(input_data)
+
+        # Lazy import to avoid startup delay
+        from moai_adk.rank.hook import is_project_excluded, submit_session_hook
+
+        # Check if this project is excluded
+        project_path = session_data.get("projectPath") or session_data.get("cwd")
+        if is_project_excluded(project_path):
+            return  # Silently skip excluded projects
+
+        result = submit_session_hook(session_data)
+
+        if result["success"]:
+            print("Session submitted to MoAI Rank", file=sys.stderr)
+        elif result["message"] != "Not registered with MoAI Rank":
+            print(f"MoAI Rank: {result['message']}", file=sys.stderr)
+
+    except json.JSONDecodeError:
+        pass
+    except ImportError:
+        # moai-adk not installed, silently skip
+        pass
+    except Exception as e:
+        print(f"MoAI Rank hook error: {e}", file=sys.stderr)
+
+if __name__ == "__main__":
+    main()
+'''
+
+
+def _register_hook_in_settings() -> bool:
+    """Register SessionEnd hook in ~/.claude/settings.json.
+
+    Claude Code requires hooks to be explicitly registered in settings.json
+    to be executed. This function adds the SessionEnd hook configuration.
+
+    Returns:
+        True if successfully registered, False otherwise
+    """
+    import json
+
+    settings_file = Path.home() / ".claude" / "settings.json"
+    hook_command = f"python3 {Path.home()}/.claude/hooks/moai/session_end__rank_submit.py"
+
+    try:
+        # Load existing settings or create new
+        if settings_file.exists():
+            with open(settings_file, encoding="utf-8") as f:
+                settings = json.load(f)
+        else:
+            settings = {}
+
+        # Ensure hooks section exists
+        if "hooks" not in settings:
+            settings["hooks"] = {}
+
+        # Check if SessionEnd hook already exists
+        if "SessionEnd" in settings["hooks"]:
+            # Check if our hook is already registered
+            for hook_config in settings["hooks"]["SessionEnd"]:
+                for hook in hook_config.get("hooks", []):
+                    if "session_end__rank_submit.py" in hook.get("command", ""):
+                        return True  # Already registered
+
+        # Add SessionEnd hook
+        session_end_config = {
+            "matcher": ".*",
+            "hooks": [{"type": "command", "command": hook_command}],
+        }
+
+        if "SessionEnd" not in settings["hooks"]:
+            settings["hooks"]["SessionEnd"] = []
+
+        settings["hooks"]["SessionEnd"].append(session_end_config)
+
+        # Write back
+        with open(settings_file, "w", encoding="utf-8") as f:
+            json.dump(settings, f, indent=2)
+
+        return True
+
+    except (OSError, json.JSONDecodeError):
+        return False
+
+
+def install_hook(project_path: Optional[Path] = None) -> bool:
+    """Install the session end hook globally to ~/.claude/hooks/moai/.
+
+    This installs the hook at the user level so it runs for all projects.
+    Users can opt-out specific projects via ~/.moai/rank/config.yaml.
+
+    Also registers the hook in ~/.claude/settings.json so Claude Code
+    knows to execute it on session end.
+
+    Args:
+        project_path: Deprecated, ignored. Hook is always installed globally.
+
+    Returns:
+        True if hook was installed successfully
+    """
+    _ = project_path  # Deprecated parameter, kept for backwards compatibility
+
+    # Install to global user hooks directory
+    hooks_dir = Path.home() / ".claude" / "hooks" / "moai"
+    hooks_dir.mkdir(parents=True, exist_ok=True)
+
+    hook_file = hooks_dir / "session_end__rank_submit.py"
+
+    try:
+        hook_file.write_text(create_global_hook_script())
+        hook_file.chmod(0o755)  # Make executable
+
+        # Register hook in settings.json
+        if not _register_hook_in_settings():
+            return False
+
+        return True
+    except OSError:
+        return False
+
+
+def _unregister_hook_from_settings() -> bool:
+    """Remove SessionEnd hook from ~/.claude/settings.json.
+
+    Returns:
+        True if successfully unregistered, False otherwise
+    """
+    import json
+
+    settings_file = Path.home() / ".claude" / "settings.json"
+
+    try:
+        if not settings_file.exists():
+            return True  # Nothing to remove
+
+        with open(settings_file, encoding="utf-8") as f:
+            settings = json.load(f)
+
+        if "hooks" not in settings or "SessionEnd" not in settings["hooks"]:
+            return True  # Nothing to remove
+
+        # Filter out our hook
+        settings["hooks"]["SessionEnd"] = [
+            hook_config
+            for hook_config in settings["hooks"]["SessionEnd"]
+            if not any(
+                "session_end__rank_submit.py" in hook.get("command", "") for hook in hook_config.get("hooks", [])
+            )
+        ]
+
+        # Remove empty SessionEnd array
+        if not settings["hooks"]["SessionEnd"]:
+            del settings["hooks"]["SessionEnd"]
+
+        # Write back
+        with open(settings_file, "w", encoding="utf-8") as f:
+            json.dump(settings, f, indent=2)
+
+        return True
+
+    except (OSError, json.JSONDecodeError):
+        return False
+
+
+def uninstall_hook() -> bool:
+    """Uninstall the global session end hook.
+
+    Removes both the hook file and the registration in settings.json.
+
+    Returns:
+        True if hook was uninstalled successfully
+    """
+    hook_file = Path.home() / ".claude" / "hooks" / "moai" / "session_end__rank_submit.py"
+
+    try:
+        # Remove hook file
+        if hook_file.exists():
+            hook_file.unlink()
+
+        # Unregister from settings.json
+        _unregister_hook_from_settings()
+
+        return True
+    except OSError:
+        return False
+
+
+def is_hook_installed() -> bool:
+    """Check if the global hook is installed.
+
+    Checks both the hook file existence and registration in settings.json.
+
+    Returns:
+        True if the hook file exists and is registered
+    """
+    import json
+
+    hook_file = Path.home() / ".claude" / "hooks" / "moai" / "session_end__rank_submit.py"
+    if not hook_file.exists():
+        return False
+
+    # Also check if registered in settings.json
+    settings_file = Path.home() / ".claude" / "settings.json"
+    try:
+        if not settings_file.exists():
+            return False
+
+        with open(settings_file, encoding="utf-8") as f:
+            settings = json.load(f)
+
+        if "hooks" not in settings or "SessionEnd" not in settings["hooks"]:
+            return False
+
+        # Check if our hook is registered
+        for hook_config in settings["hooks"]["SessionEnd"]:
+            for hook in hook_config.get("hooks", []):
+                if "session_end__rank_submit.py" in hook.get("command", ""):
+                    return True
+
+        return False
+
+    except (OSError, json.JSONDecodeError):
+        return False
+
+
+def prompt_hook_installation(console: Any = None, confirm_func: Any = None) -> bool:
+    """Prompt user to install MoAI Rank hook if eligible.
+
+    This function checks if the user is registered with MoAI Rank
+    and if the global hook is not yet installed. If both conditions
+    are met, it prompts the user to install the hook.
+
+    Args:
+        console: Rich console instance for output (optional)
+        confirm_func: Confirmation function like click.confirm (optional)
+
+    Returns:
+        True if hook was installed, False otherwise
+    """
+    # Check if user is registered with MoAI Rank
+    if not RankConfig.has_credentials():
+        return False
+
+    # Check if hook is already installed
+    if is_hook_installed():
+        return False
+
+    # Lazy import to avoid circular dependencies
+    if console is None:
+        from rich.console import Console
+
+        console = Console()
+
+    if confirm_func is None:
+        import click
+
+        confirm_func = click.confirm
+
+    # Prompt user for hook installation
+    console.print()
+    console.print("[cyan]🏆 MoAI Rank Hook Installation[/cyan]")
+    console.print("[dim]You are registered with MoAI Rank but the session tracking hook is not installed.[/dim]")
+    console.print("[dim]The hook automatically submits your Claude Code session stats to the leaderboard.[/dim]")
+    console.print()
+
+    if confirm_func("Would you like to install the MoAI Rank session hook?", default=True):
+        if install_hook():
+            console.print()
+            console.print("[green]✅ MoAI Rank hook installed successfully![/green]")
+            console.print("[dim]Location: ~/.claude/hooks/moai/session_end__rank_submit.py[/dim]")
+            console.print("[dim]To exclude specific projects: moai rank exclude /path/to/project[/dim]")
+            console.print()
+            return True
+        else:
+            console.print("[yellow]⚠ Failed to install hook. You can try later with: moai rank register[/yellow]")
+            return False
+    else:
+        console.print("[dim]Skipped. You can install later with: moai rank register[/dim]")
+        return False
+
+
+def find_all_transcripts() -> list[Path]:
+    """Find all Claude Code transcript files.
+
+    Scans ~/.claude/projects/ for all session transcript JSONL files.
+    Excludes subagent transcripts.
+
+    Returns:
+        List of paths to transcript files
+    """
+    claude_dir = Path.home() / ".claude" / "projects"
+    if not claude_dir.exists():
+        return []
+
+    transcripts = []
+    for jsonl_file in claude_dir.rglob("*.jsonl"):
+        # Skip subagent transcripts
+        if "subagents" in str(jsonl_file):
+            continue
+        transcripts.append(jsonl_file)
+
+    return transcripts
+
+
+def sync_all_sessions(
+    console: Any = None,
+    max_workers: int = 20,
+    batch_size: int = 100,
+) -> dict[str, int]:
+    """Sync all existing Claude Code sessions to MoAI Rank.
+
+    Uses a 2-phase approach:
+    1. Parse all transcripts in parallel (local I/O)
+    2. Submit in batches to minimize network requests (falls back to individual if batch not available)
+
+    Args:
+        console: Rich console instance for output (optional)
+        max_workers: Maximum number of parallel workers for parsing (default: 20)
+        batch_size: Number of sessions per batch request (default: 100, max: 100)
+
+    Returns:
+        Dictionary with sync statistics:
+        - total: Total transcripts found
+        - submitted: Successfully submitted
+        - skipped: Already submitted or no token usage
+        - failed: Failed to submit
+    """
+    from concurrent.futures import ThreadPoolExecutor, as_completed
+    from threading import Lock
+
+    from rich.progress import BarColumn, Progress, SpinnerColumn, TextColumn
+
+    if console is None:
+        from rich.console import Console
+
+        console = Console()
+
+    # Check credentials
+    if not RankConfig.has_credentials():
+        console.print("[yellow]Not registered with MoAI Rank.[/yellow]")
+        return {"total": 0, "submitted": 0, "skipped": 0, "failed": 0}
+
+    # Find all transcripts
+    transcripts = find_all_transcripts()
+    if not transcripts:
+        console.print("[dim]No session transcripts found.[/dim]")
+        return {"total": 0, "submitted": 0, "skipped": 0, "failed": 0}
+
+    stats = {"total": len(transcripts), "submitted": 0, "skipped": 0, "failed": 0}
+    stats_lock = Lock()
+
+    console.print()
+    console.print(f"[cyan]Syncing {len(transcripts)} session(s) to MoAI Rank[/cyan]")
+    console.print(f"[dim]Phase 1: Parsing transcripts (parallel: {max_workers} workers)[/dim]")
+    console.print()
+
+    # Phase 1: Parse all transcripts in parallel
+    submissions: list[SessionSubmission] = []
+    skipped_count = 0
+
+    with Progress(
+        SpinnerColumn(),
+        TextColumn("[progress.description]{task.description}"),
+        BarColumn(),
+        TextColumn("[progress.percentage]{task.percentage:>3.0f}%"),
+        TextColumn("[dim]({task.completed}/{task.total})[/dim]"),
+        console=console,
+    ) as progress:
+        task = progress.add_task("Parsing transcripts", total=len(transcripts))
+
+        with ThreadPoolExecutor(max_workers=max_workers) as executor:
+            futures = {executor.submit(parse_transcript_to_submission, path): path for path in transcripts}
+
+            for future in as_completed(futures):
+                submission = future.result()
+                if submission is not None:
+                    submissions.append(submission)
+                    with stats_lock:
+                        stats["submitted"] += 1
+                else:
+                    with stats_lock:
+                        stats["skipped"] += 1
+                    skipped_count += 1
+
+                progress.update(task, advance=1)
+
+    # Phase 2: Submit
+    if not submissions:
+        console.print()
+        console.print("[dim]No valid sessions to submit.[/dim]")
+        return {
+            "total": len(transcripts),
+            "submitted": 0,
+            "skipped": len(transcripts),
+            "failed": 0,
+        }
+
+    # Reset submitted count for submission tracking
+    stats["submitted"] = 0
+    stats["skipped"] = skipped_count
+
+    client = RankClient()
+
+    # First, try to detect if batch API is available
+    batch_available = False
+    if submissions:
+        try:
+            # Try a small batch first
+            test_batch = submissions[:1]
+            client.submit_sessions_batch(test_batch)
+            batch_available = True
+        except Exception:
+            batch_available = False
+
+    if batch_available:
+        # Use batch submission
+        batch_count = (len(submissions) + batch_size - 1) // batch_size
+        console.print()
+        console.print(f"[cyan]Phase 2: Submitting {len(submissions)} session(s) [dim](batch mode)[/dim][/cyan]")
+        console.print(f"[dim]Batch size: {batch_size} | Batches: {batch_count}[/dim]")
+        console.print()
+
+        batch_submitted = 0
+        batch_skipped = 0
+        batch_failed = 0
+
+        with Progress(
+            SpinnerColumn(),
+            TextColumn("[progress.description]{task.description}"),
+            BarColumn(),
+            TextColumn("[progress.percentage]{task.percentage:>3.0f}%"),
+            TextColumn("[dim]({task.completed}/{task.total})[/dim]"),
+            console=console,
+        ) as progress:
+            task = progress.add_task("Submitting batches", total=batch_count)
+
+            for i in range(0, len(submissions), batch_size):
+                batch = submissions[i : i + batch_size]
+
+                try:
+                    response = client.submit_sessions_batch(batch)
+
+                    if response.get("success"):
+                        batch_submitted += response.get("succeeded", 0)
+                        batch_failed += response.get("failed", 0)
+
+                        # Check individual results for duplicates/skipped
+                        for result in response.get("results", []):
+                            if not result.get("success"):
+                                error_msg = result.get("error", "").lower()
+                                if "duplicate" in error_msg or "already" in error_msg:
+                                    batch_skipped += 1
+                                    batch_failed -= 1
+                    else:
+                        batch_failed += len(batch)
+
+                except Exception:
+                    batch_failed += len(batch)
+
+                progress.update(task, advance=1)
+
+        stats["submitted"] = batch_submitted
+        stats["skipped"] += batch_skipped
+        stats["failed"] = batch_failed
+    else:
+        # Fall back to individual submissions (sequential)
+        console.print()
+        console.print(f"[cyan]Phase 2: Submitting {len(submissions)} session(s) [dim](individual mode)[/dim][/cyan]")
+        console.print()
+
+        from time import sleep
+
+        from moai_adk.rank.client import ApiError, RankClientError
+
+        # Track errors for debugging
+        error_samples: list[str] = []
+        error_type_counts: dict[str, int] = {}
+        error_lock = Lock()
+
+        # Ensure log directory exists and set up logging
+        RankConfig.ensure_config_dir()
+        log_file = RankConfig.CONFIG_DIR / "sync_errors.log"
+
+        # Clear previous log
+        if log_file.exists():
+            log_file.unlink()
+
+        import logging
+
+        logging.basicConfig(
+            filename=str(log_file),
+            level=logging.ERROR,
+            format="%(asctime)s - %(message)s",
+            force=True,
+        )
+
+        # Use sequential submission with rate limiting to avoid server limits
+        submission_delay = 0.1  # 100ms between submissions
+
+        console.print(f"[dim]Rate limit: {submission_delay}s between submissions[/dim]")
+        console.print()
+
+        # Estimate time
+        estimated_seconds = len(submissions) * submission_delay
+        console.print(f"[dim]Estimated time: ~{int(estimated_seconds / 60)}m {int(estimated_seconds % 60)}s[/dim]")
+        console.print()
+
+        with Progress(
+            SpinnerColumn(),
+            TextColumn("[progress.description]{task.description}"),
+            BarColumn(),
+            TextColumn("[progress.percentage]{task.percentage:>3.0f}%"),
+            TextColumn("[dim]({task.completed}/{task.total})[/dim]"),
+            TextColumn("[dim]|[/dim]"),
+            TextColumn("[green]✓ {task.fields[submitted]}[/green]"),
+            TextColumn("[dim]○ {task.fields[skipped]}[/dim]"),
+            TextColumn("[red]✗ {task.fields[failed]}[/red]"),
+            console=console,
+        ) as progress:
+            task = progress.add_task(
+                "Submitting sessions",
+                total=len(submissions),
+                submitted=0,
+                skipped=0,
+                failed=0,
+            )
+
+            for idx, submission in enumerate(submissions):
+                try:
+                    response = client.submit_session(submission)
+
+                    # Handle nested response structure
+                    session_id: str | None = None
+                    if isinstance(response, dict):
+                        # Try direct access
+                        session_id = response.get("sessionId")
+                        # If not found, check nested data structure
+                        if not session_id and isinstance(response.get("data"), dict):
+                            session_id = response["data"].get("sessionId")
+
+                    if session_id:
+                        stats["submitted"] += 1
+                    else:
+                        # Check for duplicate/already recorded
+                        msg = str(response.get("message", "")).lower()
+                        if "duplicate" in msg or "already" in msg:
+                            stats["skipped"] += 1
+                        else:
+                            # Log unexpected failure
+                            with error_lock:
+                                if len(error_samples) < 5:
+                                    error_samples.append(f"No sessionId: {response}")
+                            stats["failed"] += 1
+
+                except (ApiError, RankClientError) as e:
+                    error_msg = str(e).lower()
+                    if "duplicate" in error_msg or "already" in error_msg:
+                        stats["skipped"] += 1
+                    elif "too soon" in error_msg or "rate limit" in error_msg:
+                        # Rate limit hit - wait and retry
+                        with error_lock:
+                            if len(error_samples) < 5:
+                                error_samples.append(f"Rate limited: {e}")
+                        sleep(2)  # Wait before retry
+                        try:
+                            response = client.submit_session(submission)
+                            session_id = None
+                            if isinstance(response, dict):
+                                session_id = response.get("sessionId")
+                                if not session_id and isinstance(response.get("data"), dict):
+                                    session_id = response["data"].get("sessionId")
+                            if session_id:
+                                stats["submitted"] += 1
+                            else:
+                                stats["failed"] += 1
+                        except Exception:
+                            stats["failed"] += 1
+                    else:
+                        # Log API errors
+                        with error_lock:
+                            if len(error_samples) < 20:
+                                error_samples.append(f"API Error: {e}")
+                            # Categorize error type
+                            err_str = str(e)
+                            if "too soon" in err_str.lower():
+                                error_type_counts["rate_limited"] = error_type_counts.get("rate_limited", 0) + 1
+                            elif "validation" in err_str.lower():
+                                error_type_counts["validation"] = error_type_counts.get("validation", 0) + 1
+                            else:
+                                error_type_counts["other_api"] = error_type_counts.get("other_api", 0) + 1
+                        logging.error(f"API Error: {e}")
+                        stats["failed"] += 1
+                except Exception as e:
+                    # Log unexpected errors
+                    with error_lock:
+                        if len(error_samples) < 20:
+                            error_samples.append(f"Unexpected: {e}")
+                        error_type_counts["unexpected"] = error_type_counts.get("unexpected", 0) + 1
+                    logging.error(f"Unexpected error: {e}")
+                    stats["failed"] += 1
+
+                # Update progress bar
+                progress.update(
+                    task,
+                    advance=1,
+                    submitted=stats["submitted"],
+                    skipped=stats["skipped"],
+                    failed=stats["failed"],
+                )
+
+                # Small delay to avoid rate limiting
+                if idx < len(submissions) - 1:
+                    sleep(submission_delay)
+
+        # Show error samples if there were failures
+        if error_samples:
+            console.print()
+            console.print("[yellow]Sample errors:[/yellow]")
+            for i, error in enumerate(error_samples, 1):
+                console.print(f"  [dim]{i}. {error}[/dim]")
+
+        # Show error type breakdown
+        if error_type_counts:
+            console.print()
+            console.print("[yellow]Error breakdown:[/yellow]")
+            for error_type, count in sorted(error_type_counts.items(), key=lambda x: -x[1]):
+                console.print(f"  [dim]{error_type}: {count}[/dim]")
+
+    # Print summary
+    console.print()
+    console.print("[bold]Sync Complete[/bold]")
+    console.print(f"  [green]✓ Submitted:[/green] {stats['submitted']}")
+    console.print(f"  [dim]○ Skipped:[/dim]   {stats['skipped']} [dim](no usage or duplicate)[/dim]")
+    if stats["failed"] > 0:
+        console.print(f"  [red]✗ Failed:[/red]    {stats['failed']}")
+    console.print()
+
+    return stats