Add smart input truncation and schema simplification

[michaeldistel]

Problem

When inputs or schemas exceed token budgets, we need graceful handling instead of hard failures.

Dependencies

• Requires: Add basic token counting and budget warnings #39 (basic token counting)

Proposed Solution

Implement intelligent truncation strategies:

1. Input truncation

   • Truncate from middle (keep start/end context)
   • Preserve structure (complete sentences/paragraphs)
   • Add truncation markers: "...[truncated X tokens]..."
   • Log original vs. truncated length

2. Schema simplification

   • Remove field descriptions if over budget
   • Omit optional fields
   • Simplify complex nested structures
   • Warn user about simplification

3. Prioritization

interface TruncationConfig {
  strategy: 'middle' | 'end' | 'start';
  preserveStructure: boolean;
  prioritize: 'input' | 'schema';
}

4. User control
   • CLI flag: --max-input-tokens
   • CLI flag: --truncate-strategy middle|end|start
   • Config file support

Acceptance Criteria

• Input truncation implementation
• Schema simplification logic
• Truncation preserves text structure
• Clear warnings about data loss
• User-configurable strategies
• Tests for different content types
• Documentation with examples

Related

• Parent: Improve LLM prompt engineering and clarity #30 (closed - split into focused issues)
• Prerequisite: Add basic token counting and budget warnings #39 (token counting)
• Follow-up: Advanced chunking (separate issue)