Update app.py

app.py

@@ -673,9 +673,13 @@ TEXT_EXTENSIONS = {
 import zipfile
 import shutil
 
+# =============================================================================
+# ROBUST FILE & API HANDLERS (The "Box Cutter" & "Persistent Dialer")
+# =============================================================================
+
 def process_uploaded_file(file) -> str:
     """
-    Read an uploaded file.
+    Read an uploaded file. 
     1. Handles Gradio list inputs (fixes the crash).
     2. Unzips archives so the agent can see inside (fixes the blindness).
     3. Reads text files into context.
@@ -688,7 +692,6 @@ def process_uploaded_file(file) -> str:
         if len(file) == 0: return ""
         file = file[0]
 
-    # Handle both file objects (which have .name) and direct string paths
     file_path = file.name if hasattr(file, 'name') else str(file)
     file_name = os.path.basename(file_path)
     suffix = os.path.splitext(file_name)[1].lower()
@@ -696,9 +699,7 @@ def process_uploaded_file(file) -> str:
     # FIX 2: Handle ZIP files (The "Unpacking Protocol")
     if suffix == '.zip':
         try:
-            # Create a clean directory for this upload
             extract_to = Path(REPO_PATH) / "uploaded_assets" / file_name.replace(".zip", "")
-            # Clean up if it exists from a previous run to avoid clutter
             if extract_to.exists():
                 shutil.rmtree(extract_to)
             extract_to.mkdir(parents=True, exist_ok=True)
@@ -706,11 +707,9 @@ def process_uploaded_file(file) -> str:
             with zipfile.ZipFile(file_path, 'r') as zip_ref:
                 zip_ref.extractall(extract_to)
                 
-            # Get a preview of contents to show the Agent
             file_list = [f.name for f in extract_to.glob('*')]
             preview = ", ".join(file_list[:10])
-            if len(file_list) > 10:
-                preview += f", ... (+{len(file_list)-10} more)"
+            if len(file_list) > 10: preview += f", ... (+{len(file_list)-10} more)"
                 
             return (f"📦 **Unzipped: {file_name}**\n"
                     f"Location: `{extract_to}`\n"
@@ -719,7 +718,7 @@ def process_uploaded_file(file) -> str:
         except Exception as e:
             return f"⚠️ Failed to unzip {file_name}: {e}"
 
-    # EXISTING: Handle Text files
+    # Handle Text files
     if suffix in TEXT_EXTENSIONS or suffix == '':
         try:
             with open(file_path, 'r', encoding='utf-8', errors='ignore') as f:
@@ -730,7 +729,7 @@ def process_uploaded_file(file) -> str:
         except Exception as e:
             return f"📎 **Uploaded: {file_name}** (error reading: {e})"
     
-    # Fallback for actual binary files (images, binaries)
+    # Fallback for binary
     else:
         try:
             size = os.path.getsize(file_path)
@@ -739,134 +738,72 @@ def process_uploaded_file(file) -> str:
             return f"📎 **Uploaded: {file_name}** (error getting size: {e})"
 
 
-    # die Standard file handling remains the same...
-
-# =============================================================================
-# AGENTIC LOOP
-# =============================================================================
-# CHANGELOG [2026-02-01 - Claude/Opus]
-# The core conversation loop. For each user message:
-# 1. Build messages array with system prompt + history + new message
-# 2. Send to Kimi K2.5 via HF Inference API
-# 3. Parse response for tool calls
-# 4. If READ tool calls: execute immediately, inject results, loop back to Kimi
-# 5. If WRITE tool calls: stage in approval queue, notify user
-# 6. If no tool calls: return conversational response
-# 7. Save the turn to ChromaDB for persistent memory
-#
-# The loop runs up to MAX_ITERATIONS times to handle multi-step tool use.
-# Each iteration either executes tools and loops, or returns the final text.
-#
-# IMPORTANT: Gradio 5.0+ chatbot with type="messages" expects history as a
-# list of {"role": str, "content": str} dicts. We maintain that format
-# throughout.
-# =============================================================================
-
-MAX_ITERATIONS = 5
-
-
 def call_model_with_retry(messages, model_id, max_retries=4):
-    """
-    Tries to call the API. If it hits a 504 (Busy/Timeout), it waits and retries.
-    """
+    """Tries to call the API. If it hits a 504/503 (Busy), it waits and retries."""
     for attempt in range(max_retries):
         try:
-            # Attempt the call
             return client.chat_completion(
-                model=model_id,
-                messages=messages,
-                max_tokens=2048,
-                temperature=0.7
+                model=model_id, messages=messages, max_tokens=2048, temperature=0.7
             )
         except Exception as e:
-            # Check if it is a 'busy' error (504 Gateway Timeout or 503 Service Unavailable)
             error_str = str(e)
             if "504" in error_str or "503" in error_str or "timeout" in error_str.lower():
-                # If this was our last attempt, we have to give up.
-                if attempt == max_retries - 1:
-                    print(f"❌ Final API attempt failed: {e}")
-                    raise e
-                
-                # The Backoff: Wait 2 seconds, then 4, then 8...
-                wait_time = 2 * (2 ** attempt) 
+                if attempt == max_retries - 1: raise e
+                wait_time = 2 * (2 ** attempt)
                 print(f"⚠️ API Busy (Attempt {attempt+1}/{max_retries}). Retrying in {wait_time}s...")
                 time.sleep(wait_time)
             else:
-                # If it's a different error (like 401 Unauthorized), don't retry.
                 raise e
 
 
-def agent_loop(message: str, history: list, pending_proposals: list, uploaded_file) -> tuple:
-    """Main agentic conversation loop.
+# =============================================================================
+# AGENTIC LOOP
+# =============================================================================
 
-    Args:
-        message: User's text input
-        history: Chat history as list of {"role": ..., "content": ...} dicts
-        pending_proposals: Current list of staged write proposals (gr.State)
-        uploaded_file: Optional uploaded file from the file input widget
+MAX_ITERATIONS = 5
 
-    Returns:
-        Tuple of (updated_history, cleared_textbox, updated_proposals,
-                  updated_gate_choices, updated_stats_files, updated_stats_convos)
-    """
+def agent_loop(message: str, history: list, pending_proposals: list, uploaded_file) -> tuple:
+    # 1. Handle empty inputs
     if not message.strip() and uploaded_file is None:
-        # Nothing to do
-        return (
-            history, "", pending_proposals,
-            _format_gate_choices(pending_proposals),
-            _stats_label_files(), _stats_label_convos()
-        )
+        return (history, "", pending_proposals, _format_gate_choices(pending_proposals), _stats_label_files(), _stats_label_convos())
 
-    # Inject uploaded file content if present
+    # 2. Inject File Content
     full_message = message.strip()
     if uploaded_file is not None:
-        file_context = process_uploaded_file(uploaded_file)
-        if file_context:
-            full_message = f"{file_context}\n\n{full_message}" if full_message else file_context
+        try:
+            file_context = process_uploaded_file(uploaded_file)
+            if file_context:
+                full_message = f"{file_context}\n\n{full_message}" if full_message else file_context
+        except Exception as e:
+             full_message += f"\n[System Error processing file: {e}]"
 
     if not full_message:
-        return (
-            history, "", pending_proposals,
-            _format_gate_choices(pending_proposals),
-            _stats_label_files(), _stats_label_convos()
-        )
+        return (history, "", pending_proposals, _format_gate_choices(pending_proposals), _stats_label_files(), _stats_label_convos())
 
-    # Add user message to history
+    # 3. Update History
     history = history + [{"role": "user", "content": full_message}]
 
-    # Build messages for the API
+    # 4. Prepare Context
     system_prompt = build_system_prompt()
     api_messages = [{"role": "system", "content": system_prompt}]
-
-    # Include recent history (cap to avoid token overflow)
-    # Keep last 20 turns to stay within Kimi's context window
-    recent_history = history[-40:]  # 40 entries = ~20 turns (user+assistant pairs)
+    recent_history = history[-40:]
     for h in recent_history:
         api_messages.append({"role": h["role"], "content": h["content"]})
 
-    # Agentic loop: tool calls → execution → re-prompt → repeat
     accumulated_text = ""
     staged_this_turn = []
 
+    # 5. The Thinking Loop
     for iteration in range(MAX_ITERATIONS):
         try:
-            response = client.chat_completion(
-                model=MODEL_ID,
-                messages=api_messages,
-                max_tokens=2048,
-                temperature=0.7
-            )
+            # USE RETRY LOGIC HERE
+            response = call_model_with_retry(api_messages, MODEL_ID)
             content = response.choices[0].message.content or ""
         except Exception as e:
             error_msg = f"⚠️ API Error: {e}"
             history = history + [{"role": "assistant", "content": error_msg}]
-            return (
-                history, "", pending_proposals,
-                _format_gate_choices(pending_proposals),
-                _stats_label_files(), _stats_label_convos()
-            )
+            return (history, "", pending_proposals, _format_gate_choices(pending_proposals), _stats_label_files(), _stats_label_convos())
 
-        # Parse for tool calls
         tool_calls = parse_tool_calls(content)
         conversational_text = extract_conversational_text(content)
 
@@ -874,81 +811,60 @@ def agent_loop(message: str, history: list, pending_proposals: list, uploaded_fi
             accumulated_text += ("\n\n" if accumulated_text else "") + conversational_text
 
         if not tool_calls:
-            # No tools — this is the final response
             break
 
-        # Process each tool call
         tool_results_for_context = []
         for tool_name, args in tool_calls:
             result = execute_tool(tool_name, args)
-
             if result["status"] == "executed":
-                # READ tool — executed, feed result back to model
-                tool_results_for_context.append(
-                    f"[Tool Result: {tool_name}]\n{result['result']}"
-                )
+                tool_results_for_context.append(f"[Tool Result: {tool_name}]\n{result['result']}")
             elif result["status"] == "staged":
-                # WRITE tool — staged for approval
                 proposal = {
                     "id": f"proposal_{int(time.time())}_{tool_name}",
-                    "tool": tool_name,
-                    "args": result["args"],
-                    "description": result["description"],
-                    "timestamp": time.strftime("%H:%M:%S")
+                    "tool": tool_name, "args": result["args"],
+                    "description": result["description"], "timestamp": time.strftime("%H:%M:%S")
                 }
                 staged_this_turn.append(proposal)
-                tool_results_for_context.append(
-                    f"[Tool {tool_name}: STAGED for human approval. "
-                    f"Josh will review this in the Build Approval Gate.]"
-                )
+                tool_results_for_context.append(f"[Tool {tool_name}: STAGED for human approval.]")
             elif result["status"] == "error":
-                tool_results_for_context.append(
-                    f"[Tool Error: {tool_name}]\n{result['result']}"
-                )
+                tool_results_for_context.append(f"[Tool Error: {tool_name}]\n{result['result']}")
 
-        # If we only had staged tools and no reads, break the loop
         if tool_results_for_context:
-            # Feed tool results back as a system message for the next iteration
             combined_results = "\n\n".join(tool_results_for_context)
             api_messages.append({"role": "assistant", "content": content})
             api_messages.append({"role": "user", "content": f"[Tool Results]\n{combined_results}"})
         else:
             break
 
-    # Build final response
+    # 6. Finalize Response
     final_response = accumulated_text
-
-    # Append staging notifications if any writes were staged
     if staged_this_turn:
         staging_notice = "\n\n---\n🛡️ **Staged for your approval** (see Build Approval Gate tab):\n"
         for proposal in staged_this_turn:
             staging_notice += f"- {proposal['description']}\n"
         final_response += staging_notice
-        # Add to persistent queue
         pending_proposals = pending_proposals + staged_this_turn
 
     if not final_response:
-        final_response = "🤔 I processed your request but didn't generate a text response. Check the Build Approval Gate if I staged any operations."
+        final_response = "🤔 I processed your request but didn't generate a text response."
 
-    # Add assistant response to history
     history = history + [{"role": "assistant", "content": final_response}]
 
-    # Save conversation turn for persistent memory
     try:
-        # We use the robust retry function we added
-        response = call_model_with_retry(api_messages, MODEL_ID)
-        content = response.choices[0].message.content or ""
-    except Exception as e:
-        # This line is now correctly aligned with 'try'
-        error_msg = f"⚠️ API Error: {e}"
-        history = history + [{"role": "assistant", "content": error_msg}]
-        return (
-            history, "", pending_proposals,
-            _format_gate_choices(pending_proposals),
-            _stats_label_files(), _stats_label_convos()
-        )
-
-            # ... error handling ...
+        turn_count = len([h for h in history if h["role"] == "user"])
+        ctx.save_conversation_turn(full_message, final_response, turn_count)
+    except Exception:
+        pass
+
+    # 7. THE CRITICAL RETURN (Aligned with the Function Definition)
+    return (
+        history,
+        "",  # Clear the textbox
+        pending_proposals,
+        _format_gate_choices(pending_proposals),
+        _stats_label_files(),
+        _stats_label_convos()
+    )
 
 
 # =============================================================================