skills.rest
junhyekh
junhyekhCOMMUNITY
Explore all skills by this author →
GitHub

rlxp-curriculum-design

Design curriculum adjustments for reinforcement learning tasks using per-bin evidence.

1|Updated May 14, 2026
Search Tags:#curriculum#reinforcement learning#experiment analysis#scene generation#adaptive sampling#difficulty bins
One-click install
npx skills add https://github.com/junhyekh/rlxp --skill rlxp-curriculum-design
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rlxp-curriculum-design
Source: https://github.com/junhyekh/rlxp/tree/main/plugins/rl-experiment-assistant/skills/rlxp-curriculum-design
Command: npx skills add https://github.com/junhyekh/rlxp --skill rlxp-curriculum-design

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill helps you tune learning distribution instead of reward so reinforcement-learning systems can improve on hard cases without changing the final evaluation goal.

Core Features & Use Cases

  • Curriculum Diagnosis: Compare easy and hard performance using per-bin evidence to identify where learning stalls.
  • Bounded Curriculum Changes: Propose sampling reweighting, difficulty schedules, or environment-generation bridges that stay within contract scope.
  • Evaluation Safety: Preserve final success metrics and held-out evaluation semantics while adapting training exposure.
  • Use Case: When a policy succeeds on simple scenes but fails on difficult ones, use this Skill to decide whether to rebalance samples, ease progression, or add intermediate scene families.

Quick Start

Ask the assistant to review the curriculum bins, compare easy-versus-hard performance, and propose a bounded sampling or difficulty change that preserves the final evaluation metric.

Frequently Asked Questions about rlxp-curriculum-design

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

FAQPage Schema
How do I adjust curriculum sampling when my reinforcement learning policy fails on hard scenes?▼

Curriculum diagnosis compares easy versus hard performance using per-bin evidence to identify where reinforcement learning stalls. It analyzes per-bin success rates and latest training data to pinpoint difficulty levels where the policy plateaus.

What is the best way to rebalance training distribution without changing the final evaluation goal?▼

Adaptive sampling reweighting stays within contract scope by proposing difficulty schedules and environment-generation bridges that preserve final success metrics. These bounded curriculum changes adapt training exposure without altering the held-out evaluation semantics.

How to add intermediate scene families when easy cases are solved but harder cases fail?▼

Intermediate scene families bridge the gap between solved easy cases and failing hard ones by inserting environment-generation steps that ease difficulty progression. The Skill uses per-bin evidence and scene configuration to propose bridging scenes within contract scope.

Can I modify difficulty progression while preserving held-out evaluation semantics?▼

Yes, difficulty progression modifications preserve held-out evaluation semantics by bounding curriculum changes to training exposure only. The Skill ensures final success metrics remain intact while adapting sampling weights and scene difficulty schedules during training.

Why does my reinforcement learning policy succeed on simple scenes but fail on difficult ones?▼

Policies succeed on simple scenes but fail on difficult ones due to unbalanced curriculum sampling that overexposes easy cases. This Skill diagnoses per-bin performance gaps and proposes reweighting or difficulty schedule adjustments to address the imbalance.

Related Skills

Explore More Education & Research Skills→
Ronin15Ronin15
14

hammer-benchmark-regression

Compare SDL3 HammerEngine metrics against baselines to flag performance regressions.

Community
Intermediate
Ocean326Ocean326

research-training-and-ablation-loop

Analyze AI training runs and ablation results to recommend next experiments.

Community
Advanced
qiphonqiphon
2

AgentDB Learning Plugins

Create, train, and deploy reinforcement learning plugins for autonomous agents.

Community
Intermediate
junhyekhjunhyekh
1

rl-reward-curriculum-design

Route legacy RL reward and curriculum requests to RLXP intervention skills.

Community
Intermediate