htmlgraph 0.24.1__py3-none-any.whl → 0.25.0__py3-none-any.whl

htmlgraph/hooks/pretooluse.py

@@ -1,27 +1,38 @@
 """
-Unified PreToolUse Hook - Parallel Orchestrator + Validator
+Unified PreToolUse Hook - Parallel Orchestrator + Validator + Event Tracing
 
-This module provides a unified PreToolUse hook that runs both orchestrator
-enforcement and work validation checks in parallel using asyncio.
+This module provides a unified PreToolUse hook that runs orchestrator
+enforcement, work validation checks, and event tracing in parallel using asyncio.
 
 Architecture:
-- Runs orchestrator check and validator check simultaneously
+- Runs orchestrator check, validator check, and event tracing simultaneously
 - Combines results into Claude Code standard format
 - Returns blocking response only if both checks agree
 - Provides combined guidance from both systems
+- Generates tool_use_id and initiates event tracing for correlation
 
 Performance:
 - ~40-50% faster than sequential subprocess execution
 - Single Python process (no subprocess overhead)
 - Parallel execution via asyncio.gather()
+
+Event Tracing:
+- Generates UUID v4 for tool_use_id
+- Captures tool name, input, start time (ISO8601 UTC), session_id
+- Inserts start event into tool_traces table for PostToolUse correlation
+- Non-blocking - errors gracefully degrade to allow tool execution
 """
 
 import asyncio
 import json
+import logging
 import os
 import sys
+import uuid
+from datetime import datetime, timezone
 from typing import Any
 
+from htmlgraph.db.schema import HtmlGraphDB
 from htmlgraph.hooks.orchestrator import enforce_orchestrator_mode
 from htmlgraph.hooks.task_enforcer import enforce_task_saving
 from htmlgraph.hooks.validator import (
@@ -32,6 +43,450 @@ from htmlgraph.hooks.validator import (
     validate_tool_call,
 )
 
+logger = logging.getLogger(__name__)
+
+
+def generate_tool_use_id() -> str:
+    """
+    Generate UUID v4 for tool_use_id.
+
+    Used for trace correlation between PreToolUse and PostToolUse hooks.
+
+    Returns:
+        UUID v4 string (36 chars)
+    """
+    return str(uuid.uuid4())
+
+
+def get_current_session_id() -> str | None:
+    """
+    Query current session_id from environment or session files.
+
+    Reads from:
+    1. Environment variable HTMLGRAPH_SESSION_ID (set by SessionStart hook)
+    2. Latest session HTML file (fallback if env var not set)
+    3. Session registry file (fallback if HTML file not found)
+
+    Returns:
+        Session ID string or None if not found
+    """
+    # First try environment variable
+    session_id = os.environ.get("HTMLGRAPH_SESSION_ID")
+    if session_id:
+        logger.debug(f"Session ID from environment: {session_id}")
+        return session_id
+
+    # Fallback: Read from latest session HTML file
+    try:
+        import re
+        from pathlib import Path
+
+        graph_dir = Path.cwd() / ".htmlgraph"
+        sessions_dir = graph_dir / "sessions"
+
+        logger.debug(f"Looking for session files in: {sessions_dir}")
+
+        if sessions_dir.exists():
+            # Get the most recent session HTML file
+            session_files = sorted(
+                sessions_dir.glob("sess-*.html"),
+                key=lambda p: p.stat().st_mtime,
+                reverse=True,
+            )
+            logger.debug(f"Found {len(session_files)} session files")
+
+            for session_file in session_files:
+                try:
+                    # Extract session_id from filename (sess-XXXXX.html)
+                    match = re.search(r"sess-([a-f0-9]+)", session_file.name)
+                    if match:
+                        session_id = f"sess-{match.group(1)}"
+                        logger.debug(f"Found session ID from file: {session_id}")
+                        return session_id
+                except Exception as e:
+                    logger.debug(f"Error reading session file {session_file}: {e}")
+                    continue
+            logger.debug("No valid session files found")
+        else:
+            logger.debug(f"Sessions directory not found: {sessions_dir}")
+    except Exception as e:
+        logger.debug(f"Could not read from session files: {e}")
+
+    # Fallback: Read from session registry
+    try:
+        import json
+        from pathlib import Path
+
+        graph_dir = Path.cwd() / ".htmlgraph"
+        registry_dir = graph_dir / "sessions" / "registry" / "active"
+
+        if registry_dir.exists():
+            # Get the most recent session file
+            session_files = sorted(
+                registry_dir.glob("*.json"),
+                key=lambda p: p.stat().st_mtime,
+                reverse=True,
+            )
+
+            for session_file in session_files:
+                try:
+                    with open(session_file) as f:
+                        data = json.load(f)
+                        if data.get("status") == "active":
+                            session_id = data.get("session_id")
+                            if isinstance(session_id, str):
+                                return session_id
+                except Exception:
+                    continue
+    except Exception as e:
+        logger.debug(f"Could not read from session registry: {e}")
+
+    return None
+
+
+def sanitize_tool_input(tool_input: dict[str, Any]) -> dict[str, Any]:
+    """
+    Sanitize tool input to remove sensitive data before storage.
+
+    Removes or truncates:
+    - Passwords and tokens (any field with 'password', 'token', 'secret', 'key')
+    - Large binary data
+    - Deeply nested structures
+
+    Args:
+        tool_input: Raw tool input to sanitize
+
+    Returns:
+        Sanitized copy of tool_input
+    """
+    try:
+        sanitized = {}
+        sensitive_keys = {"password", "token", "secret", "key", "auth", "api_key"}
+
+        for key, value in tool_input.items():
+            # Remove sensitive fields
+            if any(sens in key.lower() for sens in sensitive_keys):
+                sanitized[key] = "[REDACTED]"
+            # Truncate very large values
+            elif isinstance(value, str) and len(value) > 10000:
+                sanitized[key] = f"{value[:10000]}... [TRUNCATED]"
+            # Keep other values
+            else:
+                sanitized[key] = value
+
+        return sanitized
+    except Exception as e:
+        logger.warning(f"Error sanitizing tool input: {e}")
+        return tool_input
+
+
+def extract_subagent_type(tool_input: dict[str, Any]) -> str | None:
+    """
+    Extract subagent_type from Task() tool input.
+
+    Looks for patterns like:
+    - "subagent_type": "gemini-spawner"
+    - Task with specific naming patterns
+
+    Args:
+        tool_input: Task() tool input parameters
+
+    Returns:
+        Subagent type string or None if not found
+    """
+    try:
+        # Check for explicit subagent_type parameter
+        if "subagent_type" in tool_input:
+            return str(tool_input.get("subagent_type"))
+
+        # Check in prompt for agent references
+        prompt = str(tool_input.get("prompt", "")).lower()
+        if "gemini" in prompt:
+            return "gemini-spawner"
+        if "codex" in prompt:
+            return "codex-spawner"
+        if "researcher" in prompt:
+            return "researcher"
+        if "debugger" in prompt:
+            return "debugger"
+
+        return None
+    except Exception:
+        return None
+
+
+def create_task_parent_event(
+    db: HtmlGraphDB,
+    tool_input: dict[str, Any],
+    session_id: str,
+    start_time: str,
+) -> str | None:
+    """
+    Create a parent event for Task() delegations.
+
+    Inserts into agent_events with:
+    - event_type: 'task_delegation'
+    - subagent_type: Extracted from tool input
+    - status: 'started'
+    - parent_event_id: UserQuery event ID (links back to conversation root)
+
+    This event will be linked to child events created by the subagent
+    and updated when SubagentStop fires.
+
+    Args:
+        db: Database connection
+        tool_input: Task() tool input parameters
+        session_id: Current session ID
+        start_time: ISO8601 UTC timestamp
+
+    Returns:
+        Parent event_id if successful, None otherwise
+    """
+    try:
+        from pathlib import Path
+
+        if not db.connection:
+            db.connect()
+
+        parent_event_id = f"evt-{str(uuid.uuid4())[:8]}"
+        subagent_type = extract_subagent_type(tool_input)
+        prompt = str(tool_input.get("prompt", ""))[:200]
+
+        # Load UserQuery event ID for parent-child linking
+        graph_dir = Path.cwd() / ".htmlgraph"
+        user_query_event_id = None
+        try:
+            from htmlgraph.hooks.event_tracker import load_user_query_event
+
+            user_query_event_id = load_user_query_event(graph_dir, session_id)
+        except Exception:
+            pass
+
+        # Build input summary
+        input_summary = json.dumps(
+            {
+                "subagent_type": subagent_type or "general-purpose",
+                "prompt": prompt,
+            }
+        )[:500]
+
+        cursor = db.connection.cursor()  # type: ignore[union-attr]
+
+        # Insert parent event
+        cursor.execute(
+            """
+            INSERT INTO agent_events
+            (event_id, agent_id, event_type, timestamp, tool_name,
+             input_summary, session_id, status, subagent_type, parent_event_id)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        """,
+            (
+                parent_event_id,
+                "claude-code",  # Main orchestrator agent
+                "task_delegation",
+                start_time,
+                "Task",
+                input_summary,
+                session_id,
+                "started",
+                subagent_type or "general-purpose",
+                user_query_event_id,  # Link to UserQuery event
+            ),
+        )
+
+        db.connection.commit()  # type: ignore[union-attr]
+
+        # Export to environment for subagent reference
+        os.environ["HTMLGRAPH_PARENT_EVENT"] = parent_event_id
+        os.environ["HTMLGRAPH_PARENT_QUERY_EVENT"] = (
+            user_query_event_id or ""
+        )  # For spawners to use
+        os.environ["HTMLGRAPH_SUBAGENT_TYPE"] = subagent_type or "general-purpose"
+
+        logger.debug(
+            f"Created parent event for Task delegation: "
+            f"event_id={parent_event_id}, subagent_type={subagent_type}, "
+            f"parent_query_event={user_query_event_id}"
+        )
+
+        return parent_event_id
+
+    except Exception as e:
+        logger.warning(f"Error creating parent event: {e}")
+        return None
+
+
+def create_start_event(
+    tool_name: str, tool_input: dict[str, Any], session_id: str
+) -> str | None:
+    """
+    Capture and store tool execution start event.
+
+    Inserts into tool_traces table with:
+    - tool_use_id: UUID v4 for correlation
+    - trace_id: Parent trace ID (from context)
+    - session_id: Current session
+    - tool_name: Tool being executed
+    - tool_input: Sanitized input parameters
+    - start_time: ISO8601 UTC timestamp
+    - status: 'started'
+
+    For Task() calls, also creates a parent event for event nesting.
+
+    Args:
+        tool_name: Name of tool being executed
+        tool_input: Tool input parameters (will be sanitized)
+        session_id: Current session ID
+
+    Returns:
+        tool_use_id on success, None on error
+    """
+    tool_use_id = None
+    try:
+        from pathlib import Path
+
+        tool_use_id = generate_tool_use_id()
+        trace_id = os.environ.get("HTMLGRAPH_TRACE_ID", tool_use_id)
+        start_time = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M:%S")
+
+        # Sanitize input before storing
+        sanitized_input = sanitize_tool_input(tool_input)
+
+        # Connect to database (use project's .htmlgraph/index.sqlite, not home directory)
+        graph_dir = Path.cwd() / ".htmlgraph"
+        db_path = str(graph_dir / "index.sqlite")
+        db = HtmlGraphDB(db_path)
+
+        # Ensure session exists (create placeholder if needed)
+        if not db._ensure_session_exists(session_id, "system"):
+            logger.warning(f"Could not ensure session {session_id} exists in database")
+
+        # Insert start event into tool_traces
+        if not db.connection:
+            db.connect()
+
+        cursor = db.connection.cursor()  # type: ignore[union-attr]
+
+        # Check if this is a Task() call for parent event creation
+        parent_event_id = None
+        if tool_name == "Task":
+            parent_event_id = create_task_parent_event(
+                db, tool_input, session_id, start_time
+            )
+
+        # Insert into agent_events table (for dashboard display)
+        import uuid
+
+        event_id = f"evt-{str(uuid.uuid4())[:8]}"
+
+        cursor.execute(
+            """
+            INSERT INTO agent_events
+            (event_id, agent_id, event_type, timestamp, tool_name,
+             input_summary, session_id, status, parent_event_id)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)
+        """,
+            (
+                event_id,
+                "claude-code",  # Agent executing the tool
+                "tool_call",
+                start_time,
+                tool_name,
+                json.dumps(sanitized_input)[:500],  # Truncate for summary
+                session_id,
+                "recorded",
+                parent_event_id,  # Link to parent if this is Task()
+            ),
+        )
+
+        # Also insert into tool_traces for correlation (if table exists)
+        try:
+            cursor.execute(
+                """
+                INSERT INTO tool_traces
+                (tool_use_id, trace_id, session_id, tool_name, tool_input,
+                 start_time, status, parent_tool_use_id)
+                VALUES (?, ?, ?, ?, ?, ?, ?, ?)
+            """,
+                (
+                    tool_use_id,
+                    trace_id,
+                    session_id,
+                    tool_name,
+                    json.dumps(sanitized_input),
+                    start_time,
+                    "started",
+                    None,  # Will be set by SubagentStop hook
+                ),
+            )
+        except Exception as e:
+            logger.debug(f"Could not insert into tool_traces: {e}")
+
+        db.connection.commit()  # type: ignore[union-attr]
+        db.disconnect()
+
+        logger.debug(
+            f"Created start event: tool_use_id={tool_use_id}, "
+            f"tool={tool_name}, session={session_id}, parent_event={parent_event_id}"
+        )
+        return tool_use_id
+
+    except Exception as e:
+        logger.warning(f"Error creating start event: {e}")
+        # Graceful degradation - return None but don't block tool
+        return None
+
+
+async def run_event_tracing(
+    tool_input: dict[str, Any],
+) -> dict[str, Any]:
+    """
+    Run event tracing (async wrapper).
+
+    Generates tool_use_id and creates start event in database.
+    Non-blocking - errors don't prevent tool execution.
+
+    Args:
+        tool_input: Hook input with tool name and parameters
+
+    Returns:
+        Event tracing response: {"hookSpecificOutput": {"tool_use_id": "...", ...}}
+    """
+    try:
+        loop = asyncio.get_event_loop()
+        tool_name = tool_input.get("name", "") or tool_input.get("tool_name", "")
+        session_id = get_current_session_id()
+
+        # Skip if no session ID
+        if not session_id:
+            logger.debug("No session ID found, skipping event tracing")
+            return {}
+
+        # Run in thread pool since it involves I/O
+        tool_use_id = await loop.run_in_executor(
+            None,
+            create_start_event,
+            tool_name,
+            tool_input,
+            session_id,
+        )
+
+        if tool_use_id:
+            # Store in environment for PostToolUse correlation
+            os.environ["HTMLGRAPH_TOOL_USE_ID"] = tool_use_id
+
+            return {
+                "hookSpecificOutput": {
+                    "tool_use_id": tool_use_id,
+                    "additionalContext": f"Event tracing started: {tool_use_id}",
+                }
+            }
+
+        return {}
+    except Exception:
+        # Graceful degradation - allow on error
+        return {}
+
 
 async def run_orchestrator_check(tool_input: dict[str, Any]) -> dict[str, Any]:
     """
@@ -183,17 +638,20 @@ async def pretooluse_hook(tool_input: dict[str, Any]) -> dict[str, Any]:
             "hookSpecificOutput": {
                 "hookEventName": "PreToolUse",
                 "updatedInput": {...},  # If task enforcer modified input
-                "additionalContext": "Combined guidance"
+                "additionalContext": "Combined guidance",
+                "tool_use_id": "..."  # For PostToolUse correlation
             }
         }
     """
-    # Run all four checks in parallel using asyncio.gather
+    # Run all five checks in parallel using asyncio.gather
     (
+        event_tracing_response,
         orch_response,
         validate_response,
         task_response,
         debug_guidance,
     ) = await asyncio.gather(
+        run_event_tracing(tool_input),
         run_orchestrator_check(tool_input),
         run_validation_check(tool_input),
         run_task_enforcement(tool_input),
@@ -209,6 +667,12 @@ async def pretooluse_hook(tool_input: dict[str, Any]) -> dict[str, Any]:
     # Collect guidance from all systems
     guidance_parts = []
 
+    # Event tracing guidance
+    if "hookSpecificOutput" in event_tracing_response:
+        ctx = event_tracing_response["hookSpecificOutput"].get("additionalContext", "")
+        if ctx:
+            guidance_parts.append(f"[EventTrace] {ctx}")
+
     # Orchestrator guidance
     if "hookSpecificOutput" in orch_response:
         ctx = orch_response["hookSpecificOutput"].get("additionalContext", "")
@@ -245,6 +709,12 @@ async def pretooluse_hook(tool_input: dict[str, Any]) -> dict[str, Any]:
         }
     }
 
+    # Add tool_use_id for PostToolUse correlation if available
+    if "hookSpecificOutput" in event_tracing_response:
+        tool_use_id = event_tracing_response["hookSpecificOutput"].get("tool_use_id")
+        if tool_use_id:
+            response["hookSpecificOutput"]["tool_use_id"] = tool_use_id
+
     # Check if task enforcer provided updatedInput
     updated_input = None
     if "hookSpecificOutput" in task_response: