Auto Research Agent

A reference framework for understanding autonomous AI research pipelines. Learn how AI can optimize ML training with fixed time budgets and metric-driven ite...

Security Scan

This is an instruction-only, educational reference about autonomous ML research; nothing in the skill requests credentials or installs code, though there are minor inconsistencies between the documentation and the package contents.

Purpose & Capability

The skill's name/description match the SKILL.md content (an educational reference about autoresearch). However, the manifest claims three files (prepare.py, train.py, program.md) that are not included, and the skill declares python3 as a required binary despite stating it does not run code. These are small inconsistencies but not evidence of malicious intent.

Instruction Scope

SKILL.md is documentation only and does not instruct the agent to read system files, access credentials, or transmit data externally. The instructions stay within an educational scope.

Install Mechanism

No install spec and no code files are present, so nothing would be written to disk or executed during installation.

Credentials

No environment variables, credentials, or config paths are requested; requested privileges are minimal and proportionate to a read-only documentation skill.

Persistence & Privilege

The skill is not always-enabled and doesn't request persistent system presence or modifications to other skills or system-wide settings.

Assessment

This appears to be a harmless, read-only educational guide. Before installing, consider: (1) If you expected runnable example code, note that the referenced files are not included—the skill is documentation-only. (2) The declared dependency on python3 is unnecessary for pure documentation but not harmful. (3) Because the skill can be invoked by the agent, review the SKILL.md to confirm it contains only the documentation you expect; if you need runnable experiments, obtain the code from the original karpathy/autoresearch repository and verify it separately.

AutoResearch Framework

A reference guide for understanding how autonomous AI research works. This skill documents the methodology from karpathy/autoresearch for educational purposes.

What This Is

This skill does NOT run any code. It serves as a reference for understanding:

• Fixed time budget experiments (5 minutes)
• Metric-driven iteration (val_bpb)
• Single-file training scope
• Self-contained ML training setup

Key Concepts

Concept	Description
val_bpb	Validation bits per byte — lower is better
Fixed Budget	Experiments run for exactly 5 minutes
Single Scope	One file to modify per experiment

Architecture Overview

The framework consists of three files:

File	Purpose
prepare.py	Data preparation (do not modify)
train.py	Model training loop reference
program.md	Research strategy template

Design Patterns

• Fixed time budget: Makes experiments directly comparable
• Single file scope: Keeps changes manageable
• Metric-driven: Uses val_bpb to compare results

For Educational Use

This skill is a reference implementation based on karpathy/autoresearch by Andrej Karpathy. It demonstrates autonomous research methodologies used in modern AI development.

Inspiration

Based on karpathy/autoresearch by Andrej Karpathy.