cc120beb · samitouri/QOSami_Agents_Zero-2

@@ -635,66 +635,104 @@ SESSION WORKFLOW:

635 - Use the input tool for interactive programs

636

637 ERROR HANDLING STRATEGY:

638 -- If you encounter the same error twice when trying the same approach, switch to an alternative, more reliable method (such as reading the entire file, editing in memory, and writing back as a single multi-line string, or using EOF CAT-style edits). Document the fallback method used in your response.

638 +- If you encounter the same error twice when trying the same approach, switch to an alternative, more reliable method (such as using Python file I/O as described below). Document the fallback method used in your response.

639 - For library/package issues, try a different library or implement a minimal solution from scratch

640 - When installation fails despite multiple attempts, use a fallback implementation as specified in your task

641 - Document environment issues and your workaround strategy in your final response

642

643 PROJECT CREATION PATTERN:

644 1. Create directories and verify structure

645 -2. Create all required files with explicit paths

645 +2. Create all required files with explicit paths (using reliable methods below)

646 3. Verify all files exist before execution

647 4. Run code in separate sessions from creation

648 5. ALWAYS install packages AND run scripts with terminal runtime to maintain environment consistency

649

650 +FILE EDITING STRATEGIES (Use Terminal First):

651 +

652 +Reading Files:

653 +- Use `cat /path/to/file` in the terminal runtime.

654 +

655 +Writing/Overwriting Files (Preferred Method for Reliability):

656 +- Use `cat > /path/to/file << 'EOF' ... EOF` in the terminal runtime. This is the MOST RELIABLE way to write or overwrite entire files, especially multi-line content or code.

657 +```json

658 +{{

659 + "thoughts": ["Overwriting file with new content using heredoc"],

660 + "tool_name": "code_execution_tool",

661 + "tool_args": {{

662 + "runtime": "terminal",

663 + "session": 0, // Use session 0 for file ops

664 + "code": "cat > /path/to/your/file.py << 'EOF'\\n# Your full new file content here\\nprint(\\'Hello Overwritten World!\\')\\nEOF"

665 + }}

666 +}}

667 +```

668 +

669 +Creating Empty Files or Simple Overwrites (Less Reliable for complex content):

670 +- `echo "single line" > /path/to/file` or `touch /path/to/file`

671 +

672 +Python Fallback (If Terminal Methods Fail Repeatedly):

673 +- ONLY if terminal methods fail, use the `python` runtime with file I/O.

674 +```json

675 +{{

676 + "thoughts": ["Terminal file write failed, falling back to Python file I/O"],

677 + "tool_name": "code_execution_tool",

678 + "tool_args": {{

679 + "runtime": "python", // Python runtime specifically for this fallback

680 + "session": 0, // Still use session 0

681 + "code": "with open('/path/to/your/file.py', 'w') as f:\\n f.write(\\'\\'\\'# Your full new file content here\\nprint(\\\\\\'Hello Python Fallback!\\\\\\')\\n\\'\\'\\')"

682 + }}

683 +}}

684 +```

685 +

686 +*NEVER* use naive string/line replacements or partial edits, especially for code or structured files. Always read the necessary context, modify the content appropriately, and write back the *entire* corrected content using the `cat > ... << EOF` method or the Python fallback.

687 +

688 AVAILABLE TOOLS:

689 - knowledge_tool: For research and information gathering

652 -- code_execution_tool: For computation, data processing, file operations (prioritize terminal runtime for both package installation AND script execution to maintain environment consistency)

690 +- code_execution_tool: For computation, data processing, file operations (use 'terminal' runtime for commands and file ops; use 'python' runtime ONLY for executing Python logic or the file I/O fallback)

691 - input: For providing input to interactive programs

692 - response_tool: REQUIRED for your final output

693

656 -TOOL USAGE:

694 +TOOL USAGE EXAMPLES (Focus on Runtimes):

695

658 -For file creation ONLY:

696 +Checking File Content (Terminal):

697 ```json

698 {{

661 - "thoughts": ["Creating project files"],

699 + "thoughts": ["Checking the content of main.py"],

700 "tool_name": "code_execution_tool",

701 "tool_args": {{

664 - "runtime": "python", // Python runtime ONLY for file creation

665 - "session": 0, // ALWAYS use session 0 for file operations

666 - "code": "import os\n\n# Create directories\nos.makedirs('project/src', exist_ok=True)\n\n# Create files\nwith open('project/src/main.py', 'w') as f:\n f.write(\"print('Hello world')\")"

702 + "runtime": "terminal", // Use terminal to run cat

703 + "session": 0,

704 + "code": "cat project/src/main.py"

705 }}

706 }}

707 ```

708

671 -For executing ANY Python code (imports, tests, etc):

709 +Executing Python Code (Terminal):

710 ```json

711 {{

712 "thoughts": ["Testing code/imports"],

675 - "tool_name": "code_execution_tool",

713 + "tool_name": "code_execution_tool",

714 "tool_args": {{

677 - "runtime": "terminal", // ALWAYS use terminal for running ANY Python code

715 + "runtime": "terminal", // ALWAYS use terminal for running ANY Python code/scripts

716 "session": 1,

717 "code": "python -c 'import pandas; print(pandas.__version__)'"

718 }}

719 }}

720 ```

721

684 -For installing packages:

722 +Installing Packages (Terminal):

723 ```json

724 {{

725 "thoughts": ["Installing required packages"],

726 "tool_name": "code_execution_tool",

727 "tool_args": {{

690 - "runtime": "terminal",

728 + "runtime": "terminal", // Terminal for pip

729 "session": 1,

730 "code": "pip install pandas matplotlib"

731 }}

732 }}

733 ```

734

697 -For running and testing code:

735 +Running Scripts (Terminal):

736 ```json

737 {{

738 "thoughts": ["Reset session before running"],

@@ -710,27 +748,27 @@ For running and testing code:

748 "thoughts": ["Running the created file"],

749 "tool_name": "code_execution_tool",

750 "tool_args": {{

713 - "runtime": "terminal",

751 + "runtime": "terminal", // Terminal to execute the python script

752 "session": 1,

753 "code": "python project/src/main.py"

754 }}

755 }}

756 ```

757

720 -For checking environment:

758 +Checking Environment (Terminal):

759 ```json

760 {{

761 "thoughts": ["Verifying Python environment"],

762 "tool_name": "code_execution_tool",

763 "tool_args": {{

726 - "runtime": "terminal",

764 + "runtime": "terminal", // Terminal for shell commands

765 "session": 1,

766 "code": "which python && python --version && pip list | grep pandas"

767 }}

768 }}

769 ```

770

733 -For research:

771 +Research (Knowledge Tool):

772 ```json

773 {{

774 "thoughts": ["Need information about X"],

@@ -741,7 +779,7 @@ For research:

779 }}

780 ```

781

744 -For your final response (REQUIRED):

782 +Final Response (Response Tool - REQUIRED):

783 ```json

784 {{

785 "thoughts": ["Task complete, delivering results"],

@@ -752,23 +790,17 @@ For your final response (REQUIRED):

790 }}

791 ```

792

755 -IMPORTANT: When executing terminal commands, monitor the output carefully for errors, especially ModuleNotFoundError or ImportError. If library imports fail after installation, verify that your terminal commands and Python code are using the same environment.

793 +IMPORTANT: When executing terminal commands, monitor the output carefully for errors, especially `command not found`, `ModuleNotFoundError` or `ImportError`. If library imports fail after installation, verify that your terminal commands and Python code are using the same environment (`which python`).

794

795 EXECUTION STRATEGY:

796 1. UNDERSTAND the task requirements

797 2. PLAN your approach before writing any code

760 -3. CREATE complete project with all necessary files

761 -4. TEST your implementation thoroughly

762 -5. PIVOT quickly if you encounter repeated errors with the same approach

798 +3. CREATE/EDIT files using reliable terminal methods (`cat`, `cat > EOF`)

799 +4. TEST your implementation thoroughly (using `terminal` runtime for execution)

800 +5. PIVOT quickly if you encounter repeated errors (try Python file I/O fallback for edits)

801 6. DELIVER using the response tool

802

765 -Remember: The response_tool is REQUIRED for your final output.

766 -

767 -CODE EDITING BEST PRACTICE:

768 -- Strongly discourage naive string or line replacements for code edits, especially in Python or structured files, as this can easily break indentation, structure, or introduce subtle bugs.

769 -- Read the file to identify the exact issues. After reviewing the content, implement the necessary fixes to ensure proper syntax throughout the file.

770 -- The most reliable and robust approach is to always read the entire file into memory, make your edits there, and write back the full, updated content as a single multi-line string (or using a 'cat' style overwrite). This ensures file integrity and preserves formatting.

771 -- For all file edits, especially in Python or markdown, always prefer this full overwrite method over partial or regex-based replacements.

803 +Remember: The response_tool is REQUIRED for your final output. Avoid complex escaping; use heredocs or the Python fallback for multi-line strings.

804 """

805

806 # Execute task using call_subordinate pattern

@@ -1350,7 +1382,7 @@ CODE EDITING BEST PRACTICE:

1382 return json.dumps(formatted_response, indent=2)

1383

1384 async def _integrate_results(self, team_id, step=None, review_summary=None, **kwargs):

1353 - """Integrate results from all team members into a final product, now as a two-step process: review, then edit/update.\\n\\nBEST PRACTICE: For all markdown or code section edits, always read the entire file into memory, construct the full updated content, and overwrite the file in one write operation (single multi-line string). After writing, read the file back to confirm the update. Do NOT use regex, partial, or line-by-line replacements for in-place section updates—these are unreliable and discouraged."""

1385 + """Integrate results from all team members into a final product, now as a two-step process: review, then edit/update.\\n\\nBEST PRACTICE: For all markdown or code section edits, always read the necessary context, construct the full updated content, and overwrite the file in one write operation using the terminal `cat > /path/to/file << 'EOF' ... EOF` method. After writing, read the file back (`cat /path/to/file`) to confirm the update. Do NOT use regex, partial, or line-by-line replacements—these are unreliable."""

1386 # Enhanced progress tracking

1387 progress_prefix = f"Team {team_id} Integration"

1388 self.log.update(progress=f"{progress_prefix}: Starting integration process... Step: {step if step else 'default (review -> edit)'}")

@@ -1634,10 +1666,10 @@ CODE EDITING BEST PRACTICE:

1666 doc_check_prompt = (

1667 f"START OF TEAM PLANNING PHASE.\\n"

1668 f"Step 1: Initial Document Check (Perform ONCE)\\n"

1637 - f"This is the very first step. Check if the file {doc_path} exists.\\n"

1638 - f"- If it exists: Read the entire file and provide a structured summary of its sections and key content.\\n"

1639 - f"- If it does NOT exist: State that clearly, create it from the template at {template_path}, and then summarize the new file.\\n"

1640 - f"CRITICAL: After summarizing or creating the file THIS ONE TIME, you MUST immediately proceed to Step 2 (Project File Review). Do NOT repeat Step 1. Do NOT re-initiate team creation. Your output for Step 1 is just the summary. Now, continue to Step 2."

1669 + f"This is the very first step. Check if the file {doc_path} exists using `ls {doc_path}`.\\n"

1670 + f"- If it exists: Read the entire file using `cat {doc_path}` and provide a structured summary of its sections and key content.\\n"

1671 + f"- If it does NOT exist: State that clearly. Create it by copying the template at {template_path} using the reliable `cat > {doc_path} << \\'EOF\\' ... EOF` method, reading the template content first with `cat {template_path}`. Then, read the *newly created* file back using `cat {doc_path}` and summarize it.\\n"

1672 + f"CRITICAL: Use the specified terminal commands (`ls`, `cat`, `cat > EOF`). After summarizing or creating the file THIS ONE TIME, you MUST immediately proceed to Step 2 (Project File Review). Do NOT repeat Step 1. Your output for Step 1 is just the summary. Now, continue to Step 2."

1673 )

1674 # The agent should summarize and then continue, not loop.

1675 team_leader.hist_add_user_message(UserMessage(message=doc_check_prompt, attachments=[]))

@@ -1725,23 +1757,28 @@ CODE EDITING BEST PRACTICE:

1757 self.log.update(progress=f"Planning Step 7/7: Updating planning document at {doc_path}...") # Add log here

1758 # Construct the final planning summary string first

1759 planning_summary_content = (

1728 - f"{doc_check_summary}\n\n" # Use the summary from step 1

1729 - f"## Project File Review\n{file_review}\n\n"

1730 - f"## Project Goal and Challenges\n{goal_summary}\n\n"

1731 - f"## Team Roles and Skills\n{roles_suggestion}\n\n"

1732 - f"## Role-Specific Task Assignment Guidance\n{role_task_guidance}\n\n"

1733 - f"## High-Level Task Breakdown\n{task_breakdown}\n\n"

1734 - f"## Clarifications or Questions\n{clarifications}\n"

1760 + f"{doc_check_summary}\\n\\n" # Use the summary from step 1 # Escaped newline

1761 + f"## Project File Review\\n{file_review}\\n\\n" # Escaped newline

1762 + f"## Project Goal and Challenges\\n{goal_summary}\\n\\n" # Escaped newline

1763 + f"## Team Roles and Skills\\n{roles_suggestion}\\n\\n" # Escaped newline

1764 + f"## Role-Specific Task Assignment Guidance\\n{role_task_guidance}\\n\\n" # Escaped newline

1765 + f"## High-Level Task Breakdown\\n{task_breakdown}\\n\\n" # Escaped newline

1766 + f"## Clarifications or Questions\\n{clarifications}\\n" # Escaped newline

1767 )

1736 -

1768 +

1769 # Use the reliable full-overwrite method to update the document's overview section

1770 + # Escape backticks and dollar signs within the heredoc content if they were potentially present

1771 + escaped_planning_summary_content = planning_summary_content.replace('`', '\\`').replace('$', '\\$')

1772 +

1773 doc_update_prompt = (

1774 f"Step 7: Update Planning Document ({doc_path})\\n"

1740 - f"Task: Read the entire document at '{doc_path}'. Find the '## Project Overview' section. Replace the content *under* this header with the new planning summary provided below. If the section doesn't exist, add it. Construct the full, updated markdown content in memory. Overwrite the file at '{doc_path}' in one single write operation using the full updated content. After writing, read the file back to confirm the '## Project Overview' section contains the new summary.\\n"

1741 - f"CRITICAL: Use the full overwrite method (read all -> modify in memory -> write all). Do NOT use partial replacements.\\n"

1742 - f"\n--- New Project Overview Content ---\n{planning_summary_content}\n--- End New Content ---"

1775 + f"Task: Read the entire document at '{doc_path}' using `cat {doc_path}`. Find the '## Project Overview' section. Replace the content *under* this header with the new planning summary provided below. If the section doesn't exist, add it at an appropriate place (e.g., after the guide). Construct the full, updated markdown content *in memory* first.\\n"

1776 + f"Then, overwrite the file at '{doc_path}' in ONE single write operation using the terminal heredoc method: `cat > {doc_path} << \\'EOF\\'\\n[Your FULL reconstructed file content]\\nEOF`.\\n"

1777 + f"After writing, read the file back using `cat {doc_path}` to confirm the '## Project Overview' section contains the new summary.\\n"

1778 + f"CRITICAL: Use the specified `cat` and `cat > EOF` commands. Do NOT use partial replacements or other runtimes.\\n"

1779 + f"\\n--- New Project Overview Content (to be inserted) ---\\n{escaped_planning_summary_content}\\n--- End New Content ---" # Use the escaped content

1780 )

1744 -

1781 +

1782 team_leader.hist_add_user_message(UserMessage(message=doc_update_prompt, attachments=[]))

1783 doc_update_confirmation = await team_leader.monologue() # This should contain confirmation from the agent

1784 planning_context["doc_update_confirmation"] = doc_update_confirmation

@@ -1753,15 +1790,15 @@ CODE EDITING BEST PRACTICE:

1790 # Compose the complete planning summary for return

1791 # This includes the initial doc check summary and the confirmation from the update step

1792 final_planning_summary = (

1756 - f"{doc_check_summary}\n\n" # Include initial doc check summary

1757 - f"--- Planning Summary ---\n"

1758 - f"Project File Review:\n{file_review}\n\n"

1759 - f"Project Goal and Challenges:\n{goal_summary}\n\n"

1760 - f"Team Roles and Skills:\n{roles_suggestion}\n\n"

1761 - f"Role-Specific Task Assignment Guidance:\n{role_task_guidance}\n\n"

1762 - f"High-Level Task Breakdown:\n{task_breakdown}\n\n"

1763 - f"Clarifications or Questions:\n{clarifications}\n\n"

1764 - f"--- Document Update Confirmation ---\n{doc_update_confirmation}\n" # Include update confirmation

1793 + f"{doc_check_summary}\\n\\n" # Include initial doc check summary

1794 + f"--- Planning Summary ---\\n"

1795 + f"Project File Review:\\n{file_review}\\n\\n"

1796 + f"Project Goal and Challenges:\\n{goal_summary}\\n\\n"

1797 + f"Team Roles and Skills:\\n{roles_suggestion}\\n\\n"

1798 + f"Role-Specific Task Assignment Guidance:\\n{role_task_guidance}\\n\\n"

1799 + f"High-Level Task Breakdown:\\n{task_breakdown}\\n\\n"

1800 + f"Clarifications or Questions:\\n{clarifications}\\n\\n"

1801 + f"--- Document Update Confirmation ---\\n{doc_update_confirmation}\\n" # Include update confirmation

1802 )

1803

1804 # Return planning summary along with confirmed doc_path and status

@@ -1775,14 +1812,14 @@ CODE EDITING BEST PRACTICE:

1812 """Handles the 'review' step of the integration process."""

1813 self.log.update(progress=f"{progress_prefix} - STEP 1/2: Starting document review...")

1814 review_doc_prompt = (

1778 - f"Integration Step 1: Review Planning/Results Document\\\\n" # Escaped newline

1779 - f"Read the current planning/results document at {doc_path}. Summarize its contents, especially the Project Overview and any previous Integration Review sections.\\\\n" # Escaped newline

1780 - f"Then review the current project directory and files. Validate that all deliverables are present and organized. Summarize results, check for gaps, and provide recommendations.\\\\n" # Escaped newline

1781 - f"Your response should include:\\\\n" # Escaped newline

1782 - f"- Directory/file review (list and describe key files and structure)\\\\n" # Escaped newline

1783 - f"- Checklist of deliverables (what was expected, what is present, what is missing)\\\\n" # Escaped newline

1784 - f"- Summary of results (from all completed tasks)\\\\n" # Escaped newline

1785 - f"- Recommendations for improvement or next steps\\\\n" # Escaped newline

1815 + f"Integration Step 1: Review Planning/Results Document\\n" # Escaped newline

1816 + f"Read the current planning/results document at {doc_path} using `cat {doc_path}`. Summarize its contents, especially the Project Overview and any previous Integration Review sections.\\n" # Escaped newline

1817 + f"Then review the current project directory and files using `ls -R` or similar terminal commands. Validate that all expected deliverables are present and organized. Summarize results, check for gaps, and provide recommendations.\\n" # Escaped newline

1818 + f"Your response should include:\\n" # Escaped newline

1819 + f"- Directory/file review (list and describe key files and structure using terminal commands)\\n" # Escaped newline

1820 + f"- Checklist of deliverables (what was expected, what is present, what is missing)\\n" # Escaped newline

1821 + f"- Summary of results (from all completed tasks)\\n" # Escaped newline

1822 + f"- Recommendations for improvement or next steps\\n" # Escaped newline

1823 f"Always state which directories and files you are reviewing in your output."

1824 )

1825 team_leader.hist_add_user_message(UserMessage(message=review_doc_prompt, attachments=[]))

Edit: Team Agent System Prompts