fix(hookify): pass warn/block messages to Claude

Warn rules previously returned only `systemMessage`, which is shown to the
user console but never injected into Claude's context. Educational messages
(the entire point of warn rules) never reached the model. Block rules
similarly returned only a generic deny without `permissionDecisionReason`.

Now:
- Warn rules on PreToolUse/PostToolUse also emit `additionalContext` inside
  `hookSpecificOutput`, so Claude sees the message and can self-correct.
- Block rules on PreToolUse/PostToolUse also emit `permissionDecisionReason`,
  so Claude sees why the operation was denied.

PreToolUse `additionalContext` support landed in Claude Code v2.1.9
(see anthropics/claude-code#15664, anthropics/claude-code#15345). PostToolUse
has supported it since earlier. The capability is now available; this patch
makes hookify use it.

Refs:
- anthropics/claude-code#15203 (warn — closed as duplicate)
- anthropics/claude-code#12446 (block — closed)
- anthropics/claude-code#15664, anthropics/claude-code#15345 (SDK additionalContext support)
- Prior PR attempts: anthropics/claude-code#15218, anthropics/claude-code#15219 (closed unmerged, predate v2.1.9)

Author: WingsOfPanda

plugins/hookify/core/rule_engine.py

@@ -73,7 +73,8 @@ def evaluate_rules(self, rules: List[Rule], input_data: Dict[str, Any]) -> Dict[
                 return {
                     "hookSpecificOutput": {
                         "hookEventName": hook_event,
-                        "permissionDecision": "deny"
+                        "permissionDecision": "deny",
+                        "permissionDecisionReason": combined_message,
                     },
                     "systemMessage": combined_message
                 }
@@ -86,9 +87,17 @@ def evaluate_rules(self, rules: List[Rule], input_data: Dict[str, Any]) -> Dict[
         # If only warnings, show them but allow operation
         if warning_rules:
             messages = [f"**[{r.name}]**\n{r.message}" for r in warning_rules]
-            return {
-                "systemMessage": "\n\n".join(messages)
-            }
+            combined = "\n\n".join(messages)
+            response = {"systemMessage": combined}
+            # PreToolUse additionalContext supported since Claude Code v2.1.9.
+            # PostToolUse has supported it since earlier. Inject so Claude (not
+            # just the user console) sees the educational message.
+            if hook_event in ('PreToolUse', 'PostToolUse'):
+                response["hookSpecificOutput"] = {
+                    "hookEventName": hook_event,
+                    "additionalContext": combined,
+                }
+            return response
 
         # No matches - allow operation
         return {}