Agent Quick Reference

This page is for maintainers and agents who need the shortest route to current repo facts.

Start Here

  • Install and smoke check: Installation

  • Backend choice: Choosing a Backend

  • Task index: Tasks

  • Algorithms index: Algorithms

  • PPO entrypoint: scripts/train_rsl_rl.py

  • MLX PPO entrypoint: scripts/train_mlx_ppo.py

  • APPO entrypoint: scripts/train_appo.py

  • SAC / TD3 / FlashSAC entrypoint: scripts/train_offpolicy.py

  • HIM-PPO entrypoint: scripts/train_him_ppo.py

  • HORA distillation entrypoint: scripts/train_hora_distill.py

Contracts To Keep In Mind

  • Env contract: src/unilab/base/np_env.py

  • Backend contract: src/unilab/base/backend/base.py

  • Training helpers: src/unilab/training/run.py

  • Config schema: src/unilab/structured_configs.py

  • Developer standard: Architecture Overview

  • High-risk areas: see the repo-root AGENTS.md.

Use uv run train, uv run eval, or uv run demo for command examples. Choose algorithm, task, and backend through --algo, --task, and --sim; write only facts that can be traced to code, config, tests, or current docs.