unified-deliverables-flow

name	unified-deliverables-flow
description	Generate spreadsheets, diagrams, and PDF reports in a phased workflow with explicit iteration budgets and error recovery

Unified Multi-Deliverable Generation Workflow

This skill provides a structured, phased approach for creating multiple deliverable types (spreadsheets, diagrams, PDF reports) in a single cohesive workflow with explicit iteration budgeting to prevent premature exhaustion.

Overview

Use this workflow when you need to:

• Generate multiple deliverable types (Excel, images, PDFs) in one task
• Ensure balanced iteration allocation across all deliverables
• Handle tool failures gracefully with retries and fallbacks
• Produce downloadable artifacts with verified paths

Critical: Iteration Budget Allocation

Allocate iterations BEFORE starting work:

Phase	Deliverable	Budget	Checkpoint
Phase 1	Spreadsheet	8 iterations	File exists + readable
Phase 2	Diagram	8 iterations	File exists + >1KB
Phase 3	PDF Report	8 iterations	File exists + ARTIFACT_PATH output
Buffer	Error recovery	6 iterations	Remaining for retries
Total	All deliverables	30 iterations	All verified

Rules:

• Complete each phase before moving to the next
• If a phase exceeds budget, use fallback strategy (see below)
• Never spend >10 iterations on a single deliverable without checkpoint
• Verify each deliverable before proceeding

Step-by-Step Instructions

Phase 0: Pre-Work Setup (1 iteration max)

1. Define all deliverables explicitly:

   Deliverable 1: hardware_selection_table.xlsx (Excel with comparison data)
   Deliverable 2: cnc_workcell_layout.png (PNG diagram of layout)
   Deliverable 3: final_report.pdf (PDF summary report)
   

2. Verify workspace is accessible:

   run_shell
     command: ls -la /workspace/ && mkdir -p /workspace/artifacts
   

3. Set path variables for consistency:

   • All files go to /workspace/ or /workspace/artifacts/
   • Use absolute paths in all code
   • Never use relative paths like ./output.xlsx

Phase 1: Spreadsheet Generation (Budget: 8 iterations)

Step 1.1: Choose approach

Method	Tool	Best For
Direct Python	execute_code_sandbox	Quick generation, simple tables
Shell agent	shell_agent	Complex logic, error recovery

Step 1.2: Create spreadsheet code

from openpyxl import Workbook
from openpyxl.styles import Font, Alignment, Border, Side

def create_hardware_table():
    wb = Workbook()
    ws = wb.active
    ws.title = "Hardware Selection"
    
    # Headers with styling
    headers = ["Component", "Option A", "Option B", "Option C", "Recommendation"]
    for col, header in enumerate(headers, 1):
        cell = ws.cell(row=1, column=col, value=header)
        cell.font = Font(bold=True)
        cell.alignment = Alignment(horizontal='center')
    
    # Data rows
    data = [
        ["Controller", "PLC-X100 ($500)", "PLC-Y200 ($650)", "PLC-Z300 ($800)", "PLC-Y200"],
        ["Motor", "Servo-500W ($300)", "Servo-750W ($400)", "Stepper-1kW ($250)", "Servo-750W"],
        ["Sensor", "Proximity-S1 ($50)", "Vision-V2 ($200)", "Laser-L3 ($150)", "Vision-V2"]
    ]
    
    for row_idx, row_data in enumerate(data, 2):
        for col_idx, value in enumerate(row_data, 1):
            ws.cell(row=row_idx, column=col_idx, value=value)
    
    # Auto-adjust column widths
     col  ws.columns:
        max_length = (((cell.value))  cell  col  cell.value)
        ws.column_dimensions[col[].column_letter].width = (max_length + , )
    
    output_path = 
    wb.save(output_path)
    ()
    ()
     output_path

create_hardware_table()

Step 1.3: Execute with verification

execute_code_sandbox
  code: <spreadsheet code from Step 1.2>
  language: python

Step 1.4: Verify (REQUIRED before proceeding)

run_shell
  command: ls -lh /workspace/*.xlsx && python3 -c "import openpyxl; openpyxl.load_workbook('/workspace/hardware_selection_table.xlsx')"

Checkpoint criteria:

• ✓ File exists
• ✓ Size > 1KB
• ✓ No errors opening file
• ✗ If failed, retry once with shell_agent, then proceed to Phase 2 with note

Phase 2: Diagram Generation (Budget: 8 iterations)

Step 2.1: Choose diagram type and tool

Diagram Type	Tool	Library
Flowchart/Blocks	execute_code_sandbox	matplotlib, graphviz
Layout/Floorplan	shell_agent	matplotlib, PIL
Architecture	execute_code_sandbox	matplotlib, diagram

Step 2.2: Create diagram code (matplotlib example)

import matplotlib.pyplot as plt
import matplotlib.patches as patches

def create_layout_diagram():
    fig, ax = plt.subplots(figsize=(12, 8))
    ax.set_xlim(0, 100)
    ax.set_ylim(0, 80)
    ax.set_aspect('equal')
    ax.set_title('CNC Workcell Layout', fontsize=16, pad=20)
    
    # Define workcell components
    components = [
        {'label': 'CNC Machine', 'xy': (30, 40), 'w': 25, 'h': 20, 'color': '#3498db'},
        {'label': 'Robot Arm', 'xy': (60, 40), 'w': 15, 'h': 15, 'color': '#e74c3c'},
        {'label': 'Material Rack', 'xy': (10, 20), 'w': 15, 'h': 40, 'color': '#2ecc71'},
        {'label': 'Control Panel', 'xy': (, ), : , : , : },
        {: , : (, ), : , : , : , : , : }
    ]
    
     comp  components:
         comp.get():
            rect = patches.Rectangle(
                (comp[][], comp[][]), 
                comp[], comp[],
                linewidth=, 
                edgecolor=comp[],
                facecolor=comp[],
                linestyle=  comp.get()  
            )
        :
            rect = patches.Rectangle(
                (comp[][], comp[][]), 
                comp[], comp[],
                linewidth=, 
                edgecolor=,
                facecolor=comp[]
            )
        ax.add_patch(rect)
        ax.text(
            comp[][] + comp[]/, 
            comp[][] + comp[]/, 
            comp[],
            ha=, va=, fontsize=, fontweight=
        )
    
    plt.grid(, alpha=)
    plt.xlabel()
    plt.ylabel()
    
    output_path = 
    plt.savefig(output_path, dpi=, bbox_inches=)
    plt.close()
    ()
    ()
     output_path

create_layout_diagram()

Step 2.3: Execute and verify

execute_code_sandbox
  code: <diagram code from Step 2.2>
  language: python

Verification:

run_shell
  command: ls -lh /workspace/*.png && file /workspace/*.png

Checkpoint criteria:

• ✓ File exists
• ✓ Size > 1KB
• ✓ File type confirmed as PNG/image
• ✗ If failed after 2 retries, use shell_agent fallback, then proceed

Phase 3: PDF Report Generation (Budget: 8 iterations)

Step 3.1: Prepare content aggregation

Gather data from previous phases:

• Spreadsheet: key findings, recommendations
• Diagram: visual summary
• Additional: scoring, conclusions

Step 3.2: Create PDF generation code

from reportlab.lib import colors
from reportlab.lib.pagesizes import letter
from reportlab.platypus import SimpleDocTemplate, Paragraph, Spacer, Table, TableStyle, Image
from reportlab.lib.styles import getSampleStyleSheet, ParagraphStyle
from reportlab.lib.units import inch
from reportlab.lib.enums import TA_CENTER, TA_LEFT

def create_final_report(output_path):
    doc = SimpleDocTemplate(output_path, pagesize=letter,
                           rightMargin=0.75*inch, leftMargin=0.75*inch,
                           topMargin=0.75*inch, bottomMargin=0.75*inch)
    styles = getSampleStyleSheet()
    story = []
    
    # Custom styles
    title_style = ParagraphStyle('CustomTitle', parent=styles['Heading1'],
                                  fontSize=20, spaceAfter=30, alignment=TA_CENTER,
                                  fontName='Helvetica-Bold')
    heading_style = ParagraphStyle('CustomHeading', parent=styles['Heading2'],
                                    fontSize=14, spaceAfter=12, spaceBefore=20,
                                    fontName='Helvetica-Bold')
    
    # Title
    story.append(Paragraph("CNC Workcell Implementation Report", title_style))
    story.append(Spacer(1, 0.5*inch))
    
    # Executive Summary
    story.append(Paragraph("Executive Summary", heading_style))
    summary_text = """
    This report presents the hardware selection analysis and workcell layout 
    for the proposed CNC implementation. After evaluating multiple options across 
    controllers, motors, and sensors, recommended configurations balance cost, 
    performance, and reliability.
    """
    story.append(Paragraph(summary_text, styles[]))
    story.append(Spacer(, *inch))
    
    
    story.append(Paragraph(, heading_style))
    data = [
        [, , , ],
        [, , , ],
        [, , , ],
        [, , , ]
    ]
    
    table = Table(data, colWidths=[*inch, *inch, *inch, *inch])
    table.setStyle(TableStyle([
        (, (, ), (-, ), colors ),
        (, (, ), (-, ), colors.whitesmoke),
        (, (, ), (-, -), ),
        (, (, ), (-, ), ),
        (, (, ), (-, ), ),
        (, (, ), (-, ), ),
        (, (, ), (-, -), colors.beige),
        (, (, ), (-, -), , colors.black),
        (, (, ), (-, -), )
    ]))
    story.append(table)
    story.append(Spacer(, *inch))
    
    
    story.append(Paragraph(, heading_style))
    story.append(Paragraph(, styles[]))
    story.append(Spacer(, *inch))
    
    
    story.append(Paragraph(, heading_style))
    rec_text = 
    story.append(Paragraph(rec_text, styles[]))
    story.append(Spacer(, *inch))
    
    
    story.append(Spacer(, *inch))
    footer_style = ParagraphStyle(, parent=styles[],
                                   fontSize=, alignment=TA_CENTER, textColor=colors.grey)
    story.append(Paragraph(, footer_style))
    
    doc.build(story)
    ()
    ()
     output_path

create_final_report()

Step 3.3: Execute with explicit artifact path

execute_code_sandbox
  code: <PDF code from Step 3.2>
  language: python

Step 3.4: Final verification

run_shell
  command: ls -lh /workspace/*.pdf && echo "---" && ls -lh /workspace/*.xlsx /workspace/*.png

Checkpoint criteria:

• ✓ PDF file exists
• ✓ Size > 10KB (reports should be substantial)
• ✓ ARTIFACT_PATH was output
• ✓ All three deliverables verified

Error Recovery & Fallback Strategies

When execute_code_sandbox fails repeatedly:

Strategy 1: Use shell_agent (more resilient)

shell_agent
  task: Create an Excel file at /workspace/hardware_selection_table.xlsx with hardware comparison data including controllers, motors, and sensors with costs and recommendations

Strategy 2: Simplify the code

• Remove complex styling
• Use basic libraries only (csv instead of openpyxl if needed)
• Reduce dependencies

Strategy 3: Change output format

• Excel → CSV if openpyxl fails
• PNG → SVG if PIL/matplotlib fails
• PDF → Markdown + convert later

When iteration budget is running low:

Iterations Remaining	Action
< 10	Skip non-critical formatting, use minimal viable output
< 5	Use simplest possible implementation, skip verification
< 3	Output text summary with file paths, request manual generation

Best Practices

1. Phase sequentially - Complete and verify each phase before moving on
2. Use absolute paths - Always /workspace/filename.ext, never relative
3. Output ARTIFACT_PATH - Every successful generation must print this
4. Verify before proceeding - Run shell check after each deliverable
5. Track iteration count - Count tool calls, stop at 25 to leave buffer
6. Simplify on failure - If complex code fails, strip to minimum viable
7. Document decisions - Note why fallback was used for future reference

Quick Reference: Tool Selection

Task	Primary Tool	Fallback Tool
Spreadsheet	execute_code_sandbox (openpyxl)	shell_agent (pandas)
Diagram	execute_code_sandbox (matplotlib)	shell_agent (graphviz)
PDF Report	execute_code_sandbox (reportlab)	shell_agent (fpdf)
Verification	run_shell (ls, file)	read_file (for content check)

Complete Workflow Checklist

• Phase 0: Defined all 3 deliverables with paths
• Phase 0: Verified workspace accessibility
• Phase 1: Spreadsheet created and verified (< 8 iterations)
• Phase 2: Diagram created and verified (< 8 iterations)
• Phase 3: PDF created and verified (< 8 iterations)
• All ARTIFACT_PATH outputs captured
• Final verification: all files exist with appropriate sizes
• Total iterations used ≤ 30

Troubleshooting

Problem	Immediate Action
execute_code_sandbox returns "[ERROR] unknown error"	Retry once, then switch to shell_agent
File not found after execution	Check actual path with run_shell: ls /workspace/
ARTIFACT_PATH not output	Re-run with explicit print statement
Iteration budget nearly exhausted	Skip remaining deliverables, document what was created
Library import fails	Add !pip install <library> at code start

Related Tools

• execute_code_sandbox - Primary tool for code execution (Python)
• shell_agent - Fallback for complex tasks or when sandbox fails
• run_shell - Verification and workspace inspection
• read_file - Content verification (type: xlsx, png, pdf)
• create_file - Alternative for text-based outputs