synth-env 0.0.1.dev2

No project description provided

Environments

Synthetic Environments / Long Horizon Tasks / Digital Control Agents

Motivation

• We're building Environments to have key desiderata for long-horizon language agents
• Snapshotting and reproducibility
• Statefulness as a first-class citizen
• Consistent abstractions for engine interaction and state reads across settings
• Environment observability and tracing
• HTTP access for simplified training and evaluation
• Strong data abstractions to enable easily-configurable filtering and curriculum learning
• Simple abstractions for agent-originated tool edits, etc.

User Guide

Note - this repo is under extremely active development. Hic sunt dracones, if not contributing it may be more useful as a reference or development resource than as core code for your production systems.

Supported Environments

[] Sokoban (maturing, not active dev) - This environment is not of our own making! Please credit the great researchers at Deepmind and Ragen, among others. - Checkout test_synth_react.py for a hello world example! [] Hendryks Math [] (maturing) [] Crafter-Classic (maturing) [] EnronBench (active dev) [] SWE-Bench (active dev - do not attempt) [] NMMO (active dev - do not attempt) [] Red (active dev) [] Verilog (maturing)

...

Development

To use the Astral suite of devtools:

• Package management: uv sync, uv add, uv remove
• Linting: ruff format .
• Type checking: uvx ty check

Testing and Coverage

Run tests with coverage:

# Run all tests with coverage report
pytest tests/ --cov=src --cov-report=html --cov-report=term

# Run specific test files
pytest tests/unit/test_registry.py

# Run with verbose output
pytest tests/ -v

Current test coverage: 42.3%

The test suite includes:

• Unit tests for core components (registry, environments, engines)
• Integration tests for service API endpoints
• Q solver tests* for Sokoban environment
• No AI agent demos in tests (algorithmic solving only)