Guide · Agent skill
How to use the Uni-Mol Molecular Prediction CLI agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
This guide explains how to install and use the cli-anything-unimol-tools agent skill, which lets an AI agent drive a command-line tool for training and running Uni-Mol molecular property prediction models. It is for researchers, data scientists and developers who have Python and the Uni-Mol Tools backend available. By the end you will know what the skill does, how to install it, how to phrase requests so the agent triggers it, and what to check before trusting the results.
Uni-Mol Molecular Prediction CLIHKUDS/CLI-Anything · Apache-2.0 Download the skill Quick start
- 1Install prerequisites. Confirm Python 3.8+, PyTorch 1.12+ and the Uni-Mol Tools backend are installed, with at least 4GB of RAM.
- 2Download and unzip the skill. Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 3Install the CLI package. From unimol_tools/agent-harness, run pip install -e . to install the package.
- 4Prepare a CSV. Create a CSV with a SMILES column and a target column with the values you want to predict.
- 5Ask the agent to train. Describe the task, the file and the target column in plain language, and specify the task type.
- 6Rank and clean up models. Ask the agent to rank models, then review the results before running cleanup.
- 7Predict on new molecules. Ask the agent to predict on a new CSV using a model ID from models list.
What an agent skill is and what this one does
An agent skill is a folder containing a SKILL.md file that tells an AI agent when and how to use a tool. This skill wraps a command-line interface for Uni-Mol Tools, a backend for molecular property prediction.
You give the agent a CSV file of molecular structures written as SMILES strings, plus the name of a target column. The agent can then train classification or regression models, list and rank the trained models, clean up weaker ones, and make predictions on new molecules.
- Organise experiments into named projects
- Train five task types: binary classification, regression, multiclass, multilabel classification and multilabel regression
- Track model performance history and rank models
- Analyse storage usage and clean up underperforming models
- Output machine-readable JSON with the --json flag
Install the skill
Download the skill ZIP from this page and unzip it. For a personal setup, place the skill folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.
Alternatively, upload the ZIP in the Claude apps under Settings → Capabilities → Skills.
The skill's SKILL.md describes installing the command-line package from the agent-harness folder of the Uni-Mol Tools project using pip install -e . . Run that from inside unimol_tools/agent-harness.
cd unimol_tools/agent-harness
pip install -e .Check the requirements before you start
The skill cannot run without its dependencies. The source lists Python 3.8 or newer, PyTorch 1.12 or newer, the Uni-Mol Tools backend, and at least 4GB of RAM. The source recommends 8GB or more for training.
The skill runs scripts and uses the network, so review the files before installing it on a shared machine.
- Python 3.8+
- PyTorch 1.12+
- Uni-Mol Tools backend installed
- 4GB+ RAM (8GB+ recommended for training)
Prepare your data as a CSV
Your CSV must contain a column named SMILES holding the molecular structures, and one or more target columns with the values to predict. You pass the target column name when training.
The source gives this example, where target is the column to predict:
- Required column: SMILES
- Target column: any name you pass with --target-col
- The source provides example datasets for all five task types in the unimol_tools/examples folder of the CLI-Anything-unimol-tools repository
SMILES,target
CCO,1
CCCO,0
CC(C)O,1Trigger the skill with natural requests
You do not need to type commands yourself. Describe the goal in plain language and mention your file and target column so the agent can pick the right task type. The skill's description is short, so naming the task (classification, regression, ranking models) helps the agent match it.
Examples of requests that fit the skill:
- Create a project called drug_discovery and train a classification model on train.csv with target column active for 10 epochs.
- Train a regression model on affinity values in train.csv.
- List my trained models and rank them by performance.
- Clean up models that score below 0.7, keeping at most 10.
- Predict activity for test.csv using model 3.
Commands the agent runs
Behind the scenes the agent uses commands like these. You can run them yourself to check its work. Replace placeholders such as <id> with real values from your project.
- project create / project list / project switch
- train with --data-path, --target-col, --task-type and --epochs
- models list, models show --model-id <id>, models rank
- storage analyze, cleanup auto, cleanup manual --max-models 10 --min-score 0.7
- predict --model-id <id> --data-path test.csv
python3 -m cli_anything.unimol_tools --json models listSafety notes and limits
The skill runs scripts on your machine and uses the network, so install it only from a source you trust and read SKILL.md first. The cleanup commands can remove trained models, so run models rank and storage analyze before cleanup.
The SKILL.md does not describe any validation of model accuracy. Predictions are only as good as your training data, so do not use them alone for safety-critical or regulatory decisions.
- Review scripts and network use before installing
- Check models rank before running cleanup
- Validate predictions independently before acting on them
What people use it for
- Training a binary classifier that labels candidate molecules as active or inactive
- Comparing several training runs and ranking them by performance
- Training a regression model on a continuous property such as affinity
- Running batch predictions on a test set with a saved model
- Feeding JSON output from models list into an automated pipeline
Frequently asked questions
How do I install the cli-anything-unimol-tools skill?
Download the skill ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. You can also upload it in the Claude apps under Settings → Capabilities → Skills. The CLI package itself is installed from unimol_tools/agent-harness with pip install -e .
What are the requirements for this skill?
The source lists Python 3.8+, PyTorch 1.12+, the Uni-Mol Tools backend and 4GB+ RAM, with 8GB+ recommended for training. The skill cannot run without Python, PyTorch and Uni-Mol Tools installed.
Is the cli-anything-unimol-tools skill free?
The skill is licensed under Apache-2.0, as listed on this page. The source does not mention pricing, so check the repository's licence file for the terms that apply to you.
What data format does it need?
Your CSV must have a SMILES column with molecular structures and a target column whose name you pass with --target-col. The source provides example datasets for all five task types.
Are there alternatives to this skill?
The source does not compare this skill with other tools. You can use the Uni-Mol Tools project directly from its GitHub repository, which is linked from the skill's source, if you prefer to write code yourself.
Why does the agent fail with a CUDA error or low accuracy?
The source suggests reducing the batch size or using CPU mode for CUDA errors, and trying more epochs or adjusting the learning rate for low accuracy. If the CSV is not recognised, check that a SMILES column exists.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
This skill suits chemistry and drug-discovery researchers, data scientists comparing Uni-Mol runs, and developers who need JSON output for pipelines, provided Python, PyTorch and Uni-Mol Tools are already set up. It is not a general chemistry assistant and should not be the sole basis for safety-critical or regulatory decisions.
Download the skill