Fallback Document Generation with Delegation

TOOL_FAILURE_INDICATORS = [
    "read_file returns binary or image data",
    "search_web returns unknown error or empty results",
    "execute_code_sandbox fails repeatedly",
    "Multiple consecutive tool failures on data retrieval"
]

Is the output a PDF or complex formatted document?
├── YES → Use shell_agent (delegates to skilled agent)
└── NO → Is it markdown or plain text?
    ├── YES → Use write_file (direct content generation)
    └── NO → Use shell_agent (handles uncertainty)

# Delegate to shell_agent with clear task description
shell_agent(
    task="Generate a professional [document type] about [topic]. "
         "Use appropriate libraries (e.g., reportlab for PDFs). "
         "Include: [list key sections/requirements]. "
         "Save to working directory as [filename]."
)

shell_agent(
    task="Create a 1-page SBAR (Situation-Background-Assessment-Recommendation) "
         "template PDF with 4 sections. Each section should have guiding points "
         "and blank lined spaces for user input. Use reportlab library. "
         "Save as SBAR_template.pdf in current directory."
)

# Generate content with embedded domain knowledge
report_content = f"""# [Document Title]

## Executive Summary
[Brief overview of key findings/content]

## Background
[Context and scope - use embedded knowledge]

## Main Content
[Organized sections with headers, lists, tables as appropriate]

## Limitations & Notes
[Transparent about data source limitations if relevant]

## Recommendations/Next Steps
[Actionable guidance based on available information]
"""

write_file(path="output/report.md", content=report_content)

# [Document Title]

## Executive Summary
[Brief overview of key findings/content]

## Background
[Context and scope - use embedded knowledge]

## Main Content
[Organized sections with headers, lists, tables as appropriate]

## Limitations & Notes
[Transparent about data source limitations if relevant]

## Recommendations/Next Steps
[Actionable guidance based on available information]

> **Note**: Specific [metric/data point] would typically be sourced from 
> [expected source]. The guidance below reflects established best practices 
> in this domain.

# For shell_agent: verify the output exists and is valid
list_dir(path=".")  # Confirm file was created

# For PDFs, optionally verify with pdfinfo via run_shell
# For markdown, optionally read to confirm content

# For write_file: confirm file exists
list_dir(path=".")  # Confirm file exists

# Complete detection and pivot pattern
def generate_fallback_document(task_goal, output_format="markdown"):
    # After detecting tool failures:
    
    # Step 1: Choose method based on format
    if output_format == "pdf" or "complex" in task_goal:
        # Use shell_agent for complex formats
        shell_agent(
            task=f"Generate a professional {output_format} document about "
                 f"{task_goal}. Use appropriate libraries. Include executive "
                 f"summary, main content, limitations, and recommendations. "
                 f"Save to working directory."
        )
        return "Document generated via shell_agent delegation"
    
    else:
        # Use write_file for simple formats
        report_content = f"""# {task_goal} Report

## Executive Summary
This report was generated using established domain knowledge due to 
temporary unavailability of primary data sources.

## Key Frameworks and Guidance
[Structured content with headers, bullets, tables]

## Limitations
- Specific data points from [expected sources] were unavailable
- Recommendations based on general best practices

## Action Items
1. [Concrete step 1]
2. [Concrete step 2]
"""
        write_file(path="generated_report.md", content=report_content)
        return "Report generated with write_file"