Improved token counting for OpenAI endpoints (llama.cpp token count -> cached prompt_tokens from last output -> naive string estimate). Logger refactoring with JSON format debugging to log file. Various bugfixes. Removed unused templates.py

utils.py

@@ -4,7 +4,6 @@ import time
 import logging
 import json
 import ast
-import contextlib
 from rich.console import Console
 from rich.panel import Panel
 
@@ -12,27 +11,28 @@ logger = logging.getLogger(__name__)
 console = Console()
 
 
-def init_trace_file(debug, log_dir="logs"):
+def init_trace_file(log_dir="logs"):
     if not os.path.exists(log_dir):
         os.makedirs(log_dir)
 
     timestamp = time.strftime("%Y%m%d-%H%M%S")
     filename = os.path.join(log_dir, f"trace_{timestamp}.json")
-    if debug: logger.debug(f"Trace logging initialized: {filename}")
+    logger.debug("Trace logging initialized: %s", filename)
     return filename
 
-def save_agent_trace(filepath, messages, full_history=None):
+def save_agent_trace(filepath, messages, step=0):
     try:
         data_to_save = {
             "timestamp": time.time(),
+            "step": step,
             "context_window": messages
         }
 
-        with open(filepath, 'w', encoding='utf-8') as f:
-            json.dump(data_to_save, f, indent=2, ensure_ascii=False)
+        with open(filepath, 'a', encoding='utf-8') as f:
+            f.write(json.dumps(data_to_save, ensure_ascii=False) + "\n")
 
     except Exception as e:
-        logger.error(f"Failed to save trace file: {e}")
+        logger.error("Failed to save trace file: %s", e)
 
 
 def _analyze_code_safety(code_str):
@@ -101,10 +101,9 @@ def safeguard_and_repair(debug, client, messages, schema, original_code):
     if is_safe:
         return original_code
 
+    logger.warning("Safeguard triggered: %s (Line: %s)", error_msg, line_no)
     if debug:
-        logger.warning(f"Safeguard triggered: {error_msg} (Line: {line_no})")
         console.print(Panel(f"{error_msg}", title="Safeguard Interrupt", style="bold red"))
-
         console.print(Panel(
                 f"[italic]{error_msg}[/italic]",
                 title="Unsafe Code",
@@ -146,8 +145,7 @@ def safeguard_and_repair(debug, client, messages, schema, original_code):
             return full_fixed_code
 
         except json.JSONDecodeError:
-            if debug: logger.error("Snippet repair failed to parse. Falling back to full repair.")
-            pass
+            logger.error("Snippet repair failed to parse. Falling back to full repair.")
 
     repair_messages = messages + [
         {"role": "assistant", "content": json.dumps({
@@ -171,11 +169,11 @@ def safeguard_and_repair(debug, client, messages, schema, original_code):
     except json.JSONDecodeError:
         return ""
 
-def compress_history(debug, client, messages, keep_last_pairs=2):
+def compress_history(client, messages, keep_last_pairs=2):
     keep_count = keep_last_pairs * 2
 
     if len(messages) < (2 + 2 + keep_count):
-        if debug: logger.warning("History too short to compress, but context is full. Crashing safely.")
+        logger.warning("History too short to compress, but context is full. Crashing safely.")
         return messages
 
     to_compress = messages[2:-keep_count]
@@ -194,7 +192,7 @@ def compress_history(debug, client, messages, keep_last_pairs=2):
         f"--- HISTORY START ---\n{history_text}\n--- HISTORY END ---"
     )
 
-    if debug: logger.debug(f"Compressing {len(to_compress)} messages...")
+    logger.debug("Compressing %d messages...", len(to_compress))
 
     summary_text = client.completion([{"role": "user", "content": summary_prompt}])
 
@@ -205,10 +203,10 @@ def compress_history(debug, client, messages, keep_last_pairs=2):
 
     new_messages = [messages[0], messages[1]] + [summary_message] + messages[-keep_count:]
 
-    if debug: logger.info(f"Compression complete. Reduced {len(messages)} msgs to {len(new_messages)}.")
+    logger.info("Compression complete. Reduced %d msgs to %d.", len(messages), len(new_messages))
     return new_messages
 
-def generate_final_report(debug, client, task_text, raw_answer):
+def generate_final_report(client, task_text, raw_answer):
     system_prompt = (
         "You are a professional report writer. "
         "Your goal is to convert the provided Raw Data into a clear, concise, "
@@ -226,7 +224,7 @@ def generate_final_report(debug, client, task_text, raw_answer):
 Write the final response in natural language (Markdown).
     """
 
-    if debug: logger.debug("Generating natural language report...")
+    logger.debug("Generating natural language report...")
     return client.completion([
         {"role": "system", "content": system_prompt},
         {"role": "user", "content": user_prompt}
@@ -237,5 +235,5 @@ def load_file(filepath):
         with open(filepath, 'r', encoding='utf-8') as f:
             return f.read()
     except FileNotFoundError:
-        logger.error(f"File not found: {filepath}")
+        logger.error("File not found: %s", filepath)
         sys.exit(1)