v0.1.3

Full Changelog: 0.1.2...0.1.3

Changelog

All notable changes to this project will be documented in this file.

v0.1.3 (2025-12-22)

Added

• SYSTEM_GUARDRAIL to enforce strict rules for handling untrusted data in LLM prompts.
• Support for flexible input types (dict, DataFrame, str) in generate_wrapped.
• JSON serialization of input data to prevent prompt injection.
• Explicit data delimiters (<DATA></DATA>) in prompts for safety.
• Optional user-provided model_api_url, model_api_key, and model_name.

Changed

• generate_wrapped function now normalizes inputs and applies system guardrails automatically.
• Model.call_model updated to pass SYSTEM_GUARDRAIL for both OpenAI (system role) and Anthropic (system top-level field).
• Anthropic response parsing updated for messages array format.
• OpenAI and Anthropic defaults unified and made explicit for safety.

Fixed

• Prevented prompt injection vulnerabilities by:
  • Serializing user data
  • Enforcing system-level guardrails
  • Ignoring instructions inside input data

Docs

• Updated How it Works section to clarify system token handling.
• Added clear guidelines for structured data input and expected output style.