--- name: moai-formats-data description: > Data format specialist covering TOON encoding, JSON/YAML optimization, serialization patterns, and data validation. Use when optimizing data for LLM transmission or implementing serialization. license: Apache-2.0 compatibility: Designed for Claude Code allowed-tools: Read, Write, Edit, Grep, Glob, mcp__context7__resolve-library-id, mcp__context7__get-library-docs user-invocable: false metadata: version: "2.0.0" category: "library" status: "active" updated: "2026-01-11" modularized: "true" tags: "formats, data, toon, serialization, validation, optimization" author: "MoAI-ADK Team" # MoAI Extension: Triggers triggers: keywords: ["serialization", "data format", "json", "yaml", "toon", "validation", "schema", "optimization"] --- # Data Format Specialist ## Quick Reference Advanced Data Format Management - Comprehensive data handling covering TOON encoding, JSON/YAML optimization, serialization patterns, and data validation for performance-critical applications. Core Capabilities: - TOON Encoding: 40-60% token reduction vs JSON for LLM communication - JSON/YAML Optimization: Efficient serialization and parsing patterns - Data Validation: Schema validation, type checking, error handling - Format Conversion: Seamless transformation between data formats - Performance: Optimized data structures and caching strategies - Schema Management: Dynamic schema generation and evolution When to Use: - Optimizing data transmission to LLMs within token budgets - High-performance serialization/deserialization - Schema validation and data integrity - Format conversion and data transformation - Large dataset processing and optimization Quick Start: Create a TOONEncoder instance and call encode with a dictionary containing user and age fields to compress the data. The encoded result achieves 40-60% token reduction. Call decode to restore the original data structure. Create a JSONOptimizer instance and call serialize_fast with a large dataset to achieve ultra-fast JSON processing. Create a DataValidator instance and call create_schema with a dictionary defining name as a required string type. Call validate with the data and schema to check validity. --- ## Implementation Guide ### Core Concepts TOON (Token-Optimized Object Notation): - Custom binary-compatible format optimized for LLM token usage - Type markers: # for numbers, ! for booleans, @ for timestamps, ~ for null - 40-60% size reduction vs JSON for typical data structures - Lossless round-trip encoding/decoding Performance Optimization: - Ultra-fast JSON processing with orjson achieving 2-5x faster than standard json - Streaming processing for large datasets using ijson - Intelligent caching with LRU eviction and memory management - Schema compression and validation optimization Data Validation: - Type-safe validation with custom rules and patterns - Schema evolution and migration support - Cross-field validation and dependency checking - Performance-optimized batch validation ### Basic Implementation TOON Encoding for LLM Optimization: Create a TOONEncoder instance. Define data with user object containing id, name, active boolean, and created datetime, plus permissions array. Call encode to compress and decode to restore. Compare sizes to verify reduction. Fast JSON Processing: Create a JSONOptimizer instance. Call serialize_fast to get bytes and deserialize_fast to parse. Use compress_schema with a type object and properties definition to optimize repeated validation. Data Validation: Create a DataValidator instance. Define user_schema with username requiring string type, minimum length 3, email requiring email type, and age as optional integer with minimum value 13. Call validate with user_data and schema, then check result for valid status, sanitized_data, or errors list. ### Common Use Cases API Response Optimization: Create a function to optimize API responses for LLM consumption by encoding data with TOONEncoder. Create a corresponding function to parse optimized responses by decoding TOON data back to dictionary. Configuration Management: Create a YAMLOptimizer instance and call load_fast with a config file path. Call merge_configs with base_config, env_config, and user_config for multi-file merging. Large Dataset Processing: Create a StreamProcessor with chunk_size of 8192. Define a process_item function that handles each item. Call process_json_stream with the file path and callback to process large JSON files without loading into memory. --- ## Advanced Features Overview ### Advanced TOON Features See modules/toon-encoding.md for custom type handlers (UUID, Decimal), streaming TOON processing, batch TOON encoding, and performance characteristics with benchmarks. ### Advanced Validation Patterns See modules/data-validation.md for cross-field validation, schema evolution and migration, custom validation rules, and batch validation optimization. ### Performance Optimization See modules/caching-performance.md for intelligent caching strategies, cache warming and invalidation, memory management, and performance monitoring. ### JSON/YAML Advanced Features See modules/json-optimization.md for streaming JSON processing, memory-efficient parsing, schema compression, and format conversion utilities. --- ## Works Well With - moai-domain-backend - Backend data serialization and API responses - moai-domain-database - Database data format optimization - moai-foundation-core - MCP data serialization and transmission patterns - moai-workflow-docs - Documentation data formatting - moai-foundation-context - Context optimization for token budgets --- ## Module References Core Implementation Modules: - modules/toon-encoding.md - TOON encoding implementation - modules/json-optimization.md - High-performance JSON/YAML - modules/data-validation.md - Advanced validation and schemas - modules/caching-performance.md - Caching strategies Supporting Files: - modules/INDEX.md - Module overview and integration patterns - reference.md - Extended reference documentation - examples.md - Complete working examples --- ## Technology Stack Core Libraries: - orjson: Ultra-fast JSON parsing and serialization - PyYAML: YAML processing with C-based loaders - ijson: Streaming JSON parser for large files - python-dateutil: Advanced datetime parsing - regex: Advanced regular expression support Performance Tools: - lru_cache: Built-in memoization - pickle: Object serialization - hashlib: Hash generation for caching - functools: Function decorators and utilities Validation Libraries: - jsonschema: JSON Schema validation - cerberus: Lightweight data validation - marshmallow: Object serialization/deserialization - pydantic: Data validation using Python type hints --- ## Resources For working code examples, see [examples.md](examples.md). Status: Production Ready Last Updated: 2026-01-11 Maintained by: MoAI-ADK Data Team ## Common Rationalizations | Rationalization | Reality | |---|---| | "JSON is universal, I do not need to optimize the format" | JSON verbosity wastes tokens in LLM contexts. TOON encoding reduces token usage by 30-50% for structured data. | | "YAML is more readable than JSON for configuration" | YAML readability comes with parsing ambiguity (Norway problem, indentation errors). Use JSON for machine-read config, YAML for human-edited. | | "Schema validation is unnecessary for internal data" | Internal data formats drift silently. Schema validation catches format changes before they cause runtime errors. | | "I will handle serialization edge cases as they appear" | Date formats, null handling, and unicode escaping are known edge cases. Handle them in the serializer, not in each consumer. | | "The data payload is small, optimization is unnecessary" | Small payloads multiply by request count. Optimizing a 1KB payload saves 1MB per 1000 requests. | ## Red Flags - JSON payload contains redundant keys or unnecessary nesting for LLM consumption - YAML configuration files parsed without schema validation - Date/time values serialized in inconsistent formats across endpoints - Serialization/deserialization errors silently swallowed instead of reported - No compression applied to large data transfers ## Verification - [ ] Data format chosen with documented rationale (JSON vs YAML vs TOON) - [ ] Schema validation configured for structured data inputs (show schema file) - [ ] Date/time format consistent across all serialization points (ISO-8601) - [ ] Payload size measured and within acceptable limits for the use case - [ ] Serialization round-trip test passes (serialize -> deserialize -> compare)