skills.rest
OSU-NLP-Group
OSU-NLP-GroupOFFICIAL
Explore all skills by this author →
𝕏@osunlpGitHub

early-experience-data

Generate SFT-ready expert, IWM, and reflection training data for agent environments.

137|1|Updated Jun 8, 2026
Search Tags:#training data#agent learning#early-experience#IWM#SR
One-click install
npx skills add https://github.com/OSU-NLP-Group/EarlyExperience --skill early-experience-data
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: early-experience-data
Source: https://github.com/OSU-NLP-Group/EarlyExperience/tree/main/skill
Command: npx skills add https://github.com/OSU-NLP-Group/EarlyExperience --skill early-experience-data

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill addresses the challenge of generating SFT-ready training data for early-experience paradigms in agent environments, specifically for Implicit World Modeling (IWM) and Self-Reflection (SR) methods.

Core Features & Use Cases

  • Data Generation: Automates the creation of expert, IWM, and reflection data for agent environments.
  • Method Mapping: Guides the mapping of methods like IWM and SR to specific environments.
  • Alternative Action Sampling: Provides strategies for sampling alternative actions in environments with enumerable or open action spaces.
  • Reflection Generation: Facilitates the generation of self-reflection content for training policies.
  • Use Case: Utilize this Skill to generate training data for an agent navigating a virtual world, such as predicting the next state or generating reasoning for expert actions.

Quick Start

Use the early-experience-data skill to generate expert and reflection data for a virtual environment by pointing the agent at skill/SKILL.md.

Frequently Asked Questions about early-experience-data

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

FAQPage Schema
How do I generate SFT-ready training data for agent learning environments?▼

Generate SFT-ready training data for agent learning by automating expert, IWM, and reflection data collection, alternative action sampling, and reasoning generation through Python scripts and LLM capabilities.

What is early-experience data in reinforcement learning and imitation learning?▼

Early-experience data in reinforcement learning and imitation learning is SFT-ready training content that captures expert actions, implicit world modeling, and self-reflection to bootstrap agent policies in virtual environments.

How do I sample alternative actions in open action spaces for IWM training?▼

Sample alternative actions in open action spaces for IWM training by applying strategies that gather enumerable or open action variations, facilitating the generation of implicit world modeling datasets.

Does this agent data generation approach require specific Python dependencies?▼

This agent data generation approach requires Python scripts and LLM capabilities to process environments, map IWM and SR methods, and output SFT-ready training datasets without external dependencies.

When do I need self-reflection data for training agent policies?▼

You need self-reflection data for training agent policies when utilizing Self-Reflection (SR) methods, which require generated reasoning content to improve decision-making in virtual environments.

Related Skills

Explore More Software Engineering Skills→
raphaelchristiraphaelchristi
43

harness:setup

Configure Harness Evolver, LangSmith, and baseline evaluations for Python projects.

Community
Advanced
MarcoDavaMarcoDava
1

AgentDB Learning Plugins

Create and train AI learning plugins with AgentDB reinforcement learning algorithms.

Community
Advanced
yiwei79yiwei79

agentic-eval

Evaluate and refine agent outputs through structured reflection loops.

Official
Advanced
M4n5terM4n5ter
3

training-data-curation

Standardize LLM post-training data collection and evaluation across SFT, DPO, and RLHF workflows.

Community
Intermediate