Tabular ml lab

Skill howardxie-dev/ml-agent-skills/skills/tabular-ml-lab

Portable Agent Skills for reproducible ML workflows, starting with tabular data profiling, leakage checks, baseline training, evaluation, and reports.

Install

npx -y skills add howardxie-dev/ml-agent-skills --skill tabular-ml-lab

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when the user wants to use an LLM coding agent such as Codex or Claude Code to run a reproducible machine learning workflow on tabular CSV data, including data profiling, leakage checks, binary classification baselines, evaluation metrics, threshold reports, model cards, and final reports.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.7 KB, as published. Nobody here has run it

Tabular ML Lab

Use this skill when Codex, Claude Code, or another LLM coding agent should act as a workflow agent for local, reproducible CSV binary classification baseline experiments.

Scope

CSV input only.
binary_classification only.
Logistic Regression and RandomForestClassifier baselines.
Seeded holdout split.
No Web UI, Jupyter, AutoML backend, deployment, monitoring, or production-readiness guarantee.

Before Running

Work from the ml-agent-skills repository root.

Read the task file and confirm:

data.format is csv.
task.type is binary_classification.
data.path points to an existing CSV.
data.target exists in that CSV.
run.random_seed is present or defaults to 42.
Output will be written to the requested output directory.

Use assets/task.template.yaml when creating a new task file.

Commands

Preferred full workflow:

uv run atm run path/to/task.yaml --output path/to/output_dir

Use staged scripts only for debugging phase-by-phase behavior:

uv run python skills/tabular-ml-lab/scripts/inspect_dataset.py --task path/to/task.yaml --output path/to/output_dir
uv run python skills/tabular-ml-lab/scripts/train_baseline.py --task path/to/task.yaml --output path/to/output_dir
uv run python skills/tabular-ml-lab/scripts/evaluate_model.py --task path/to/task.yaml --output path/to/output_dir
uv run python skills/tabular-ml-lab/scripts/render_report.py --task path/to/task.yaml --output path/to/output_dir

If a step fails, stop and report the failing command, error, and likely fix. Do not skip ahead.

References

Load only the reference needed for the task:

references/workflow.md for end-to-end run order and validation.
references/artifact-contract.md for required files and metadata expectations.
references/leakage-checklist.md for leakage policy and review rules.
references/metrics.md for metrics and threshold report interpretation.
references/reporting.md for user-facing summary expectations.

Reporting Expectations

When summarizing results:

Mention the selected baseline model and holdout metrics.
Mention leakage warnings and class imbalance warnings.
Mention threshold_report.csv when threshold trade-offs matter.
State that results are local baseline experiment results, not production validation.
Use actual generated artifacts; do not invent metrics or report contents.