Skip to content

AGENTS.md

Latest commit

 

History

History
29 lines (19 loc) · 1.23 KB

AGENTS.md

File metadata and controls

29 lines (19 loc) · 1.23 KB

AGENTS.md

Project Structure

JAX-based diffusion policy RL algorithms.

  • Agents: flowrl/agent/online/ (online) and flowrl/agent/offline/ (offline)
  • Configs: flowrl/config/ (dataclasses), examples/**/config/ (YAML hyperparams)
  • Entry points: examples/ — organized by scenario (online/offline) and benchmark (DMControl, HumanoidBench, MuJoCo)

General Instructions

  • Keep the code clean, readable and consistent with existing ones.
  • You can sacrifice grammar in your response for conciseness.

Adding a New Algorithm

  1. Read the official implementation; identify all hyperparameters.
  2. Implement in flowrl/agent/. Match existing code style — reference SDAC (online) or DAC (offline).
  3. Add config dataclasses and YAML files consistent with the official implementation.

Proofreading Algorithms

  • Flag anything abnormal vs. standard RL practices.
  • If an official implementation is provided, diff against it for logic and hyperparameter mismatches.

Pre-Release Checklist

  • README: Identify undocumented algorithms. Prompt me to add experiment links — only include algorithms with results. Do not overwrite existing README content.
  • Dependencies: If changes are needed, ask me before updating.