Probabl Skills plugin
Probabl skills for machine learning experiments: a scikit-learn and skrub pipeline, evaluation, audit and a link to Skore Hub
- 145 ★ on the repository
- Last commit
- BSD-3-Clause
Minimal access
Why this level
- The agent runs experiment code and writes files into the workspace, so changes need review
- Linking Skore Hub sends experiment results to an external Probabl service
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/probabl-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The agent runs experiment code and writes files into the workspace, so changes need review
- Linking Skore Hub sends experiment results to an external Probabl service
This is third-party code. Review the repository files before installing.
What it does
Probabl Skills is a set of skills from the company Probabl that steer an agent through machine learning experiments. The agent builds a pipeline on core libraries (`scikit-learn`, `skrub`, `skore`, `pandas`, `polars`) and follows a methodology, then evaluates and stores results so they are easy to audit. The skills connect the agent to Skore Hub for an overview of experiments, help plan the next experiment from a Skore audit and user feedback, and organize the workspace by a data science template (cookiecutter). The `eval` folder holds test prompts for skills such as `build-ml-pipeline`, `evaluate-ml-pipeline`, `audit-ml-pipeline`, `explore-ml-data` and `export-ml-notebook`. The installer ships in the `skore` package: the `skore skills install` command opens a picker for source, skills, agent and directory. A pack can be installed by id: `setup`, `data_analysis`, `model`, `loop` or `export`. Each skill declares a model tier (`small`, `medium`, `big`), and the authors tested work in Claude Code, OpenCode, Cursor and GitHub Copilot.
Who it is for. Data scientists and ML engineers who run experiments with an agent and want a single methodology, auditable results and an orderly workspace.
Good fit when
- You want to hand an agent a baseline scikit-learn pipeline from the data in `data/`
- Experiment results have to be stored and audited through Skore
- Next experiments are built from an audit and user feedback
- An ML project workspace should follow a data science template
Not a fit when
- The task is outside tabular machine learning: the skills target tabular data
- You need your own methodology that differs from the chosen libraries
- The `skore` installer and network access for installation are unavailable
Example request
Using `data/README.md` and the data in `data/`, build a first baseline machine learning pipeline and store the evaluation resultLimitations
Installation goes through the `skore` package from PyPI or conda-forge (pip, uv, pixi or conda), and the README has no manual route. The skills target tabular data and the scikit-learn, skrub, skore, pandas and polars stack. The Skore Hub link depends on an external Probabl service. The agent runs experiment code on your machine and writes files into the workspace, so check results by hand. The last commit is dated November 9, 2026, and model tiers and pack contents may change.
How to disable. Delete the installed skill folders from the agent skills directory (project or user) where `skore skills install` put them, and disconnect Skore Hub if you linked it.
FAQ
How do I install the Probabl skills?
Install the `skore` package (pip, uv, pixi or conda) and run `skore skills install`: the command opens a picker for skills and agent.
Which skill packs exist?
`setup` (environment, git, export), `data_analysis`, `model`, `loop` and `export` for notebooks and a documentation site.
Which agents did the authors test?
Claude Code, OpenCode, Cursor and GitHub Copilot, with no significant difference in skill invocation.
What is a skill model tier?
The `metadata.modelTier` field with `small`, `medium` or `big`. A helper never loads a skill of a higher tier than its caller.
Is Skore Hub required?
No, it is only for an overview of experiments. The skills work without it.
Discussion
No questions or reviews yet.
Related
PluginsA skill and Python package for hyperparameter tuning: the agent reads the code, proposes the next set of values, and your own run measures the result
A template for data scientists: AGENTS.md, CLAUDE.md and skills for Python, SQL, Polars and model diagnostics.
High-Stakes Analytics Lab
High-Stakes Analytics & Decision Lab
An evidence-gated analytics skill for high-stakes decisions: profiles data, picks a method the evidence supports, can answer with a data request
24 skills for numerical methods and materials science: the agent checks stability, meshes, convergence and cluster runs.
A skill that runs a Kaggle competition from rules and folds to a scored submission, with hidden-rerun debugging and ensembles
Around 200 skills for Claude Code on bioinformatics, chemistry and biostatistics, from RNA-seq analysis to molecular docking
README in short
The README explains that the skills steer an agent through machine learning experiments and lists five tasks. It shows a sample prompt, then model tiers and testing across several agents. Installation through `skore` and a description of packs follow.