analyze-rl-behavior

Synthesize evaluation traces, training logs, and GPT-5 assessments into a structured behavioral report.

276|37|Updated Dec 5, 2025

Search Tags:#automation #gpt-5 #behavioral-analysis #rl #training-logs #evaluation-metrics #model-traces

Ready to run

npx skills add https://github.com/open-thoughts/OpenThoughts-Agent --skill analyze-rl-behavior

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: analyze-rl-behavior
Source: https://github.com/open-thoughts/OpenThoughts-Agent/tree/main/.claude/skills/analyze-rl-behavior
Command: npx skills add https://github.com/open-thoughts/OpenThoughts-Agent --skill analyze-rl-behavior

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Analyze RL behavior changes in trained models by synthesizing evaluation traces, training logs, and GPT-5 assessments into a structured behavioral report.

Core Features & Use Cases

Orchestrates the local RL-behavior analysis pipeline to compare pre-RL baselines with RL checkpoints, attribute changes, and evaluate persistence.
Generates Q1–Q4 outputs (behavioral_delta, llm_judge_diff, temporal_trace_analysis, parse_skyrl_metrics, etc.) and writes results to dedicated output directories for reproducibility.
Supports optional GPT-5-based judgement and failure-mode annotation when evaluation repos are writable; designed for local Mac environments without GPUs.

Quick Start

Run from the repo root with the analyze_rl_behavior script to analyze a trained RL model's artifacts and generate the Q1–Q4 reports.

Frequently Asked Questions about analyze-rl-behavior

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

FAQPage Schema

How do I analyze RL behavior changes between pre-RL baselines and RL checkpoints?▼

You can analyze RL behavior changes by running a local pipeline that synthesizes evaluation traces, training logs, and GPT-5 assessments to compare pre-RL baselines with RL checkpoints. This generates a structured behavioral report detailing changes and persistence.

Can I run reinforcement learning evaluation trace analysis on a Mac without a GPU?▼

Yes, you can run reinforcement learning evaluation trace analysis locally on a Mac without a GPU. The pipeline is specifically designed for Mac-based environments to orchestrate behavioral analysis and generate structured Q1–Q4 reports.

What is the best way to automate behavioral analysis reporting for trained models?▼

The best way to automate behavioral analysis reporting is to use an orchestration pipeline that synthesizes model traces and evaluation metrics into structured Q1–Q4 outputs. It automatically writes results to dedicated output directories for reproducibility.

Does the behavioral analysis pipeline support GPT-5 assessments for failure-mode annotation?▼

Yes, the behavioral analysis pipeline supports optional GPT-5-based judgement and failure-mode annotation. This feature is available when your evaluation repositories are writable, allowing for deeper model assessment.

What outputs are generated when comparing RL checkpoints to baselines?▼

Comparing RL checkpoints to baselines generates Q1–Q4 outputs including behavioral_delta, llm_judge_diff, temporal_trace_analysis, and parse_skyrl_metrics. These results are written to dedicated output directories for reproducibility.

Do I need to perform preflight artifact checks before running RL behavioral analysis?▼

Yes, preflight artifact checks are performed automatically when you run the RL behavioral analysis script. This ensures all necessary evaluation traces and training logs are present before orchestrating the local analysis pipeline.