skills.rest
tachyon-beep
tachyon-beepCOMMUNITY
Explore all skills by this author →
GitHub

rl-debugging

Diagnose reinforcement learning training failures with structured diagnosis trees.

1|Updated Oct 29, 2025
Search Tags:#reinforcement-learning#debugging-framework#rl-debugging#diagnosis-trees#training-instability#policy-optimization
One-click install
npx skills add https://github.com/tachyon-beep/hamlet --skill rl-debugging
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rl-debugging
Source: https://github.com/tachyon-beep/hamlet/tree/main/.claude/skills/yzmir-deep-rl/skills/rl-debugging
Command: npx skills add https://github.com/tachyon-beep/hamlet --skill rl-debugging

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill provides a systematic framework to diagnose and fix RL training failures, guiding engineers through a prioritized set of checks and diagnostic trees to identify root causes.

Core Features & Use Cases

  • Systematic debugging framework for RL: diagnosis trees, step-by-step checks, and practical guidance.
  • Comprehensive symptom-based paths: "Agent Won't Learn", "Training Unstable", "Suboptimal Policy".
  • Practical logging and monitoring guidance to surface actionable signals.
  • Useful for rapid triage of ML training issues and improving reproducibility.

Quick Start

Invoke this skill to guide systematic diagnosis when an RL agent exhibits learning failures, instability, or suboptimal performance. Start by checking reward scale and environment sanity, then follow the diagnosis trees to identify root causes, compare against a common bugs catalog, and apply recommended fixes.

Frequently Asked Questions about rl-debugging

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I debug reinforcement learning training instability?▼

Debug reinforcement learning training instability by applying a structured diagnosis tree that checks reward scales and environment sanity to identify root causes. The framework provides step-by-step checks and a bug catalog to systematically resolve unstable training.

Why does my reinforcement learning agent won't learn or shows suboptimal policies?▼

A reinforcement learning agent may exhibit non-learning or suboptimal policies due to environment bugs or incorrect policy optimization. Follow symptom-based diagnosis paths to run systematic checks, compare against a bug catalog, and apply targeted fixes for rapid triage.

What is the best way to diagnose non-learning in RL agents?▼

The best way to diagnose non-learning in RL agents is using a formal debugging framework with diagnosis trees. Start by verifying environment sanity and reward scales, then navigate specific symptom paths to isolate root causes and improve training reproducibility.

What logging and monitoring practices should I use for RL debugging?▼

For RL debugging, implement recommended logging and monitoring practices to surface actionable signals during training. These practices help track policy optimization and environment sanity, enabling rapid triage of learning failures and improving overall training reproducibility.

Related Skills

Explore More Software Engineering Skills→
dawsonblockdawsonblock

slime-rl-training

Implement RL post-training for GLM models with slime, Megatron-LM, and SGLang.

Community
Advanced
loopyardloopyard
21

fix-bug

Reproduce bugs in Elixir/Mix projects using a test-first workflow.

Official
Intermediate
AnushkaTandon26AnushkaTandon26

hugging-face-trackio

Log and visualize ML training metrics with Trackio Python API and CLI.

Community
Advanced
tachyon-beeptachyon-beep
1

multi-agent-rl

Implement QMIX and MADDPG algorithms for multi-agent reinforcement learning with PyTorch.

Community
Intermediate