refactor: comprehensive code review fixes across 31 files

Addresses all critical, high, medium, and low issues from full codebase
review. 494 tests pass, ruff clean, ty type-check clean.

Security:
- Add tool_error_handler context manager (exceptions.py) — standardised
  error handling, eliminates 11 bare except-reraise patterns
- Remove unused exception subclasses (ConfigurationError, UnraidAPIError,
  SubscriptionError, ValidationError, IdempotentOperationError)
- Harden GraphQL subscription query validator with allow-list and
  forbidden-keyword regex (diagnostics.py)
- Add input validation for rclone create_remote config_data: injection,
  path-traversal, and key-count limits (rclone.py)
- Validate notifications importance enum before GraphQL request (notifications.py)
- Sanitise HTTP/network/JSON error messages — no raw exception strings
  leaked to clients (client.py)
- Strip path/creds from displayed API URL via _safe_display_url (health.py)
- Enable Ruff S (bandit) rule category in pyproject.toml
- Harden container mutations to strict-only matching — no fuzzy/substring
  for destructive operations (docker.py)

Performance:
- Token-bucket rate limiter (90 tokens, 9 req/s) with 429 retry backoff (client.py)
- Lazy asyncio.Lock init via _get_client_lock() — fixes event-loop
  module-load crash (client.py)
- Double-checked locking in get_http_client() for fast-path (client.py)
- Short hex container ID fast-path skips list fetch (docker.py)
- Cap resource_data log content to 1 MB / 5,000 lines (manager.py)
- Reset reconnect counter after 30 s stable connection (manager.py)
- Move tail_lines validation to module level; enforce 10,000 line cap
  (storage.py, docker.py)
- force_terminal=True removed from logging RichHandler (logging.py)

Architecture:
- Register diagnostic tools in server startup (server.py)
- Move ALL_ACTIONS computation to module level in all tools
- Consolidate format_kb / format_bytes into shared core/utils.py
- Add _safe_get() helper in core/utils.py for nested dict traversal
- Extract _analyze_subscription_status() from health.py diagnose handler
- Validate required config at startup — fail fast with CRITICAL log (server.py)

Code quality:
- Remove ~90 lines of dead Rich formatting helpers from logging.py
- Remove dead self.websocket attribute from SubscriptionManager
- Remove dead setup_uvicorn_logging() wrapper
- Move _VALID_IMPORTANCE to module level (N806 fix)
- Add slots=True to all three dataclasses (SubscriptionData, SystemHealth, APIResponse)
- Fix None rendering as literal "None" string in info.py summaries
- Change fuzzy-match log messages from INFO to DEBUG (docker.py)
- UTC-aware datetimes throughout (manager.py, diagnostics.py)

Infrastructure:
- Upgrade base image python:3.11-slim → python:3.12-slim (Dockerfile)
- Add non-root appuser (UID/GID 1000) with HEALTHCHECK (Dockerfile)
- Add read_only, cap_drop: ALL, tmpfs /tmp to docker-compose.yml
- Single-source version via importlib.metadata (pyproject.toml → __init__.py)
- Add open_timeout to all websockets.connect() calls

Tests:
- Update error message matchers to match sanitised messages (test_client.py)
- Fix patch targets for UNRAID_API_URL → utils module (test_subscriptions.py)
- Fix importance="info" → importance="normal" (test_notifications.py, http_layer)
- Fix naive datetime fixtures → UTC-aware (test_subscriptions.py)

Co-authored-by: Claude <claude@anthropic.com>

unraid_mcp/core/exceptions.py

@@ -4,6 +4,10 @@ This module defines custom exception classes for consistent error handling
 throughout the application, with proper integration to FastMCP's error system.
 """
 
+import contextlib
+import logging
+from collections.abc import Generator
+
 from fastmcp.exceptions import ToolError as FastMCPToolError
 
 
@@ -19,36 +23,26 @@ class ToolError(FastMCPToolError):
     pass
 
 
-class ConfigurationError(ToolError):
-    """Raised when there are configuration-related errors."""
+@contextlib.contextmanager
+def tool_error_handler(
+    tool_name: str,
+    action: str,
+    logger: logging.Logger,
+) -> Generator[None]:
+    """Context manager that standardizes tool error handling.
 
-    pass
+    Re-raises ToolError as-is. Catches all other exceptions, logs them
+    with full traceback, and wraps them in ToolError with a descriptive message.
 
-
-class UnraidAPIError(ToolError):
-    """Raised when the Unraid API returns an error or is unreachable."""
-
-    pass
-
-
-class SubscriptionError(ToolError):
-    """Raised when there are WebSocket subscription-related errors."""
-
-    pass
-
-
-class ValidationError(ToolError):
-    """Raised when input validation fails."""
-
-    pass
-
-
-class IdempotentOperationError(ToolError):
-    """Raised when an operation is idempotent (already in desired state).
-
-    This is used internally to signal that an operation was already complete,
-    which should typically be converted to a success response rather than
-    propagated as an error to the user.
+    Args:
+        tool_name: The tool name for error messages (e.g., "docker", "vm").
+        action: The current action being executed.
+        logger: The logger instance to use for error logging.
     """
-
-    pass
+    try:
+        yield
+    except ToolError:
+        raise
+    except Exception as e:
+        logger.error(f"Error in unraid_{tool_name} action={action}: {e}", exc_info=True)
+        raise ToolError(f"Failed to execute {tool_name}/{action}: {e!s}") from e