How do I install the Autoresearch Mlx skill?

Run npx skillmds@latest add autonomous-ai/autoresearch-mlx in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Autoresearch Mlx safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Autoresearch Mlx?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Autoresearch Mlx free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Autoresearch Mlx?

autonomous-ai (@autonomous-ai) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Autoresearch Mlx by autonomous-ai — Agent Skill

Name: Autoresearch Mlx
Author: autonomous-ai

File contents

Research notebook

Read studio.json to understand the current controls; "$STUDIO_TOOLCHAIN/../studio.config.json" describes their ranges. Run "$STUDIO_TOOLCHAIN/run.sh" train to make a new result. Successful artifacts and their measurements are in out/runs/<id>/; out/latest.json names the current result. A failed run preserves the last success and records the error in the verdict.

The local starter trains a small character transition model with NumPy on a bundled, original text corpus. It uses real training and held-out cross-entropy, not generated metrics. It is a CPU baseline, distinct from upstream MLX transformer training, which requires Apple Silicon and its prepared dataset.

Use "$STUDIO_TOOLCHAIN/../README.md" for the integration contract and commands. Read the relevant files under $STUDIO_UPSTREAM before using an upstream API. Keep controls within their documented ranges, preserve the data needed to reproduce a comparison, and distinguish preview results from native service or hardware output. The viewer supports history and artifact downloads; tell the user which run contains the result, and what was actually measured.

Run a controlled experiment

Read train.py, train.txt, and holdout.txt. Save a baseline. Change one training choice or the editable training code, then run train. Inspect evaluation.json, the saved model, and learning.csv. Keep the holdout unchanged; the runner independently reopens the model with pickle disabled and computes its score. Data hashes define which earlier runs compare. Repeat promising changes with several seeds before presenting a conclusion.

For Apple Silicon, read $STUDIO_UPSTREAM/program.md and README.md. Prepare a separate workspace mlx/ checkout and its environment/data before running mlx. The CPU starter's bits-per-character score and the upstream bits-per-byte metric are different experiments.

---name: autoresearch-mlxdescription: Create and inspect science projects in the autoresearch-mlx Harness workspace, including its local starter and optional upstream integration.---# Research notebookRead `studio.json` to understand the current controls; `"$STUDIO_TOOLCHAIN/../studio.config.json"`describes their ranges. Run `"$STUDIO_TOOLCHAIN/run.sh" train` to make a new result.Successful artifacts and their measurements are in `out/runs/<id>/`; `out/latest.json` names thecurrent result. A failed run preserves the last success and records the error in the verdict.The local starter trains a small character transition model with NumPy on a bundled, original text corpus. It uses real training and held-out cross-entropy, not generated metrics. It is a CPU baseline, distinct from upstream MLX transformer training, which requires Apple Silicon and its prepared dataset.Use `"$STUDIO_TOOLCHAIN/../README.md"` for the integration contract and commands. Read the relevantfiles under `$STUDIO_UPSTREAM` before using an upstream API. Keep controls within their documentedranges, preserve the data needed to reproduce a comparison, and distinguish preview results fromnative service or hardware output. The viewer supports history and artifact downloads; tell theuser which run contains the result, and what was actually measured.## Run a controlled experimentRead `train.py`, `train.txt`, and `holdout.txt`. Save a baseline. Change one training choiceor the editable training code, then run `train`. Inspect `evaluation.json`, the saved model,and `learning.csv`. Keep the holdout unchanged; the runner independently reopens the modelwith pickle disabled and computes its score. Data hashes define which earlier runs compare.Repeat promising changes with several seeds before presenting a conclusion.For Apple Silicon, read `$STUDIO_UPSTREAM/program.md` and `README.md`. Prepare a separateworkspace `mlx/` checkout and its environment/data before running `mlx`. The CPU starter'sbits-per-character score and the upstream bits-per-byte metric are different experiments.

Frequently asked questions

Autoresearch Mlx