v0.4.1

What's Changed

• Removal of some CLI commands + updated README.md by @MarcusTantakoun in #31

Full Changelog: v0.4.0...v0.4.1

v0.4.0

What's Changed

New Features

CLI Feature

New addition to L2P. Allows users and agents to interact with the library via CLI easily.

• CLI for user use: l2p init, l2p config, l2p models, l2p chat, l2p new l2p generate - all interactive use
• CLI for agentic use: l2p set, l2p build, l2p validate, l2p plan, l2p schema - all stateless, pipeable JSON commands designed for LLM agent consumption

PDDL Types: Pydantic BaseModel

• Migrated from data classes to Pydantic library BaseModel classes (JSON formatting). This removes the labour of regex parsing and instead the LLM declares strict JSON objects
• Incorporated PDDL types: 1.2, 2.1, 2.2, 3.0 and PDDL+ support for both domain and problem files
• Incorporated built-in class examples for LLM to follow with ease

Builder Classes

• Domain and Problem builders: Migrated from 20+ formalization functions into a single formalize_component() function that passes in the specific Pydantic type to generate. Users can now generate multiple types in a single query with the LLM.
• New prompt templates: migrated from .txt files to markdown .md files. Changed code block extraction from markdown ticks to xml tags. This is a critical improvement to LLM formatting mistakes.
• Prompt builder: structured Markdown prompt generation with _example() classmethods on every Pydantic model
• 20+ prompt templates for each class extraction - following standardized template (that is also applied in PromptBuilder)
• Custom templates: support for multi-component prompt templates and user-defined template directories
• Feedback builder: completely refactored class that incorporates different LLM feedback techniques (i.e, diagnose, evaluate, reflect, revise, select, with plan evaluation and diagnosis

Syntax Validator

Migrated from built-in validators for formalization extraction to decoupled classes / functions (apply post-generation)

• Syntax validator engine: DomainValidator / ProblemValidator with 12+ decorator-based rules: naming, type hierarchy, cycles, parameter types, arity, variable scoping, symbol references
• Parsing and formatting has changed to accommodate new Pydantic model types

Planner Backend: planner_builder.py

Refactored Planner class - now is an abstract class to incorporate other planners.

• Unified Planning backend support - UnifiedPlanning planner class alongside FastDownward
• New class PlanningResult that includes all components of the generated plan

Model Config

Refactored format of .yaml files for LLM configurations

• Added in current latest LLMs available
• Fixed bug between UnifiedLLM and OPENAI classes

Removed

• vllm.py VLLM class - deprecated since this was experimental
• paper_reconstructions - migrating folder to another repo for ease of updates

Fixed

• Python=3.10.0 compatibility error for f-string parsing fixed

Full Changelog: v0.3.3...v0.4.0

0.3.3

What's Changed

• Shifted LLM interface to llm which can be found at l2p/llm/unified.py. Configuration file found in l2p/llm/utils/llm.yaml
• Maintained OpenAI SDK class in l2p/llm/openai.py. Configuration file found in l2p/llm/utils/openaiSDK.yaml
• Updated newest Gemini 3 and GPT-5 models to config files

Full Changelog: v0.3.2...v0.3.3

0.3.2

What's Changed

• Minor Bug Fixes + parse cleanup by @MarcusTantakoun in #17

• Fixed setup.py
• Fixed llm/utils missing folder dependencies

Full Changelog: v0.3.1...v0.3.2

0.3.0

What's Changed

• Major version update by @MarcusTantakoun in #14

This PR introduces major refactors and feature additions across the L2P library. Renamed (and added new) extract_* to formalize_* functions in DomainBuilder and TaskBuilder for generating PDDL components with relaxed input requirements and enhanced syntax validation. FeedbackBuilder now provides string-based feedback without auto-regeneration and enforces a JUDGMENT header for LLM outputs. Utility functions are modularized, and the syntax validator is fully refactored to use a function-name-based interface. The LLM module adopts YAML-based configuration and removes llm_builder.py. Test cases are completed, the paper recreation code is updated, and documentation is moved to a separate repo for GitHub Pages.

Full Changelog: v0.2.2...v0.3.0

v0.2.2

Fixed version bump.

v0.2.1

What's Changed

• fixed huggingface initialization model bug (line 265 of llm_builder.py file) by @MarcusTantakoun in #10
• Fixed type absence case by @MarcusTantakoun in #12

Full Changelog: v0.2.0...v0.2.1

0.2.0

What's Changed

• Paper Reconstruction + updated papers by @MarcusTantakoun in #9

Full Changelog: v0.1.8...v0.2.0

0.1.8

Fixing the docs for l2p

0.1.7

Shifting lib name to l2p