v3.3.2: Full paths in plan, minimal Coder tools (write_file only)

Dockerfile

@@ -1,5 +1,5 @@
 # HuggingFace Spaces Dockerfile for CodePilot
-# BUILD_VERSION: 10 (v3.3.1 coder strict - no sandbox exploration)
+# BUILD_VERSION: 11 (v3.3.2 full paths in plan, minimal coder tools)
 FROM python:3.11-slim
 
 # Set working directory

chainlit_app.py

@@ -20,8 +20,8 @@ from concurrent.futures import ThreadPoolExecutor
 # ============================================================
 # STARTUP VERSION CHECK - Change this to detect if rebuild worked
 # ============================================================
-APP_VERSION = "3.3.1-coder-strict"
-BUILD_ID = "2024-12-19-v9"
+APP_VERSION = "3.3.2-full-paths"
+BUILD_ID = "2024-12-19-v10"
 print("=" * 60)
 print(f"[STARTUP] CodePilot Chainlit App")
 print(f"[STARTUP] APP_VERSION: {APP_VERSION}")

codepilot/agents/coder_agent.py

@@ -21,34 +21,35 @@ from typing import Dict, Any, Optional
 import json
 
 
-# Coder's specialized system prompt (v3.3 - no exploration, write code immediately)
+# Coder's specialized system prompt (v3.3.2 - use full paths from plan)
 CODER_SYSTEM_PROMPT = """You are an expert software engineer. Your ONLY job is to WRITE CODE.
 
-=== CRITICAL RULES ===
-1. START WRITING CODE IMMEDIATELY with write_file
-2. DO NOT run ls, find, pwd, or any exploration commands
-3. DO NOT try to access /tmp/codepilot_repos/ from sandbox - IT WILL FAIL
-4. Use file paths EXACTLY as shown in the exploration results
-5. The sandbox is ISOLATED - it cannot see the cloned repo
+=== READ THE PLAN CAREFULLY ===
+The plan contains FULL ABSOLUTE PATHS like:
+  /tmp/codepilot_repos/flask_abc123/examples/health_check/app.py
 
-=== YOUR WORKFLOW ===
-1. Look at the EXPLORATION RESULTS - they have all the file paths you need
-2. Look at the PLAN - it tells you exactly what to create
-3. Use write_file to create/modify files in the repo
-4. (Optional) Test your code by uploading to sandbox with relative paths
+Use these paths EXACTLY with write_file. Do not modify or verify them.
+
+=== YOUR ONLY WORKFLOW ===
+1. Read the PLAN - it has the exact file paths to create
+2. Call write_file with the FULL PATH from the plan
+3. Done!
 
 === TOOLS ===
-- write_file: CREATE OR MODIFY FILES - use this immediately!
-- get_code_chunk: Read a specific function if you need to see existing code
-- upload_to_sandbox: Upload YOUR code for testing (use relative paths like "app.py")
-- run_command_in_sandbox: Run YOUR uploaded code (like "python app.py")
+- write_file(path, content): Create/modify files - USE THE FULL PATH FROM PLAN
+- get_code_chunk(path, name): Read existing code if you need to copy a pattern
+- read_file(path): Read entire file only if rewriting it
 
 === EXAMPLE ===
-If plan says "create examples/health_check/app.py":
-✅ CORRECT: write_file(path="/tmp/codepilot_repos/flask_xxx/examples/health_check/app.py", content="...")
-❌ WRONG: run_command_in_sandbox("ls /tmp/codepilot_repos/...")  <-- sandbox can't access this!
+Plan says: "Create file: /tmp/codepilot_repos/flask_abc123/examples/health_check/app.py"
+You do: write_file(path="/tmp/codepilot_repos/flask_abc123/examples/health_check/app.py", content="from flask import...")
+
+=== DO NOT ===
+❌ Do NOT explore or verify paths
+❌ Do NOT use sandbox to run ls/find/pwd
+❌ Do NOT ask questions - just write the code
 
-DO NOT EXPLORE. WRITE CODE NOW.
+START WITH write_file IMMEDIATELY.
 """
 
 
@@ -77,16 +78,11 @@ class CoderAgent:
 
         self.conversation = ConversationManager()
 
-        # v3.3: Minimal tools - focus on writing code
-        # Removed sandbox commands that were being misused for exploration
+        # v3.3.2: Only file tools - no sandbox (was being misused)
         self.allowed_tools = [
             "write_file",           # PRIMARY TOOL - create or modify files
             "get_code_chunk",       # Read specific function/class if needed
             "read_file",            # Read entire file (only when rewriting)
-            "upload_to_sandbox",    # Upload code for testing
-            "execute_in_sandbox"    # Execute Python snippets for testing
-            # REMOVED: run_command_in_sandbox - was being misused for ls/find
-            # REMOVED: get_file_outline - Coder should use exploration context
         ]
 
     def run(

codepilot/agents/orchestrator.py

@@ -9,7 +9,7 @@ The orchestrator is the "brain" that:
 """
 
 # VERSION CHECK - If you see this, new code is running!
-ORCHESTRATOR_VERSION = "3.3.1-coder-strict"
+ORCHESTRATOR_VERSION = "3.3.2-full-paths"
 print(f"[ORCHESTRATOR] ========== LOADING VERSION {ORCHESTRATOR_VERSION} ==========")
 
 from enum import Enum

codepilot/agents/planner_agent.py

@@ -56,10 +56,15 @@ Use the EXPLORATION RESULTS to understand the codebase structure and create your
 4. TESTING: How to verify the changes work
 
 === PLAN QUALITY REQUIREMENTS ===
+- Use FULL ABSOLUTE PATHS: Always use the complete path from exploration (e.g., /tmp/codepilot_repos/flask_xxx/examples/file.py)
 - Be SPECIFIC: Include exact file names, function names, line numbers
 - Be ORDERED: Steps should build on each other logically
 - Be COMPLETE: Cover all aspects of the task
-- Be CONCISE: Don't repeat information from exploration
+
+=== CRITICAL FOR CODER ===
+The Coder will use your paths EXACTLY as written. Include the FULL path from the exploration context.
+Example: "Create file: /tmp/codepilot_repos/flask_abc123/examples/health_check/app.py"
+NOT: "Create file: examples/health_check/app.py"
 
 You do NOT write code - just create the plan for the Coder agent to follow.
 """