ml-pipeline

An agent skill by Jeffallan, from Jeffallan/claude-skills. Tags: analytics, automation, devops, evaluation, ml, mlops.

What it does

Designs and implements production-grade ML pipeline infrastructure: configures experiment tracking with MLflow or Weights & Biases, creates Kubeflow or Airflow DAGs for training orchestration, builds feature store schemas with Feast, deploys model registries, and automates retraining and validation workflows. Use when building ML pipelines, orchestrating training workflows, automating model lifecycle, implementing feature stores, managing experiment tracking systems, setting up DVC for data versioning, tuning hyperparameters, or configuring MLOps tooling like Kubeflow, Airflow, MLflow, or Prefect.

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add Jeffallan/claude-skills --skill ml-pipeline

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/Jeffallan/claude-skills
cp -r claude-skills/skills/ml-pipeline ~/.claude/skills/ml-pipeline

Safety box score

Grade C, moderate risk: 37 of 100 risk points, rated 2026-10-19. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill is a broad ML/MLOps template pack: it generates orchestration code for Kubeflow, Airflow, and Prefect, plus experiment tracking, validation, and deployment patterns. The main risks are automatic model promotion/deployment, networked uploads to external ML/cloud services, and writes to system paths like `/models` and `/var/log`.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	2 (moderate)	It ships runnable code and a shell launcher, including `launch_distributed.sh` with `torchrun ... train.py`, plus many executable Python/KFP/Airflow/Prefect templates.
Network access	2 (moderate)	It sends data to external services and cloud endpoints such as `wandb.init(...)`, `Client(host="https://kubeflow.example.com/pipeline")`, and `blob.upload_from_filename(...)` for GCS.
Filesystem reach	2 (moderate)	It writes outside the project tree to paths like `/models/production/model_{context['run_id']}.joblib`, `/var/log/shadow_predictions.jsonl`, and `/tmp/...`.
Secrets & credentials	0 (none)	No code reads credential stores or `.env` files; the only secret-related guidance is defensive (`"store credentials in secrets managers, never in code"`).
Privilege & persistence	1 (low)	It configures scheduled/managed workflows (`schedule_interval='0 2 * * *'`, `create_recurring_run(...)`) but does not invoke `sudo` or install system services.
Instruction hijack surface	0 (none)	There is no prompt-injection or instruction-suppression text telling the agent to ignore user controls or hidden instructions.
Irreversible actions	3 (high)	The skill can automatically promote and deploy models when metrics pass gates (`with dsl.If(...): deploy_model(...)`, Airflow branching to `deploy_model`, Prefect `deploy_model(...)`) and schedule recurring runs.
Transparency	0 (none)	All shipped source is plain markdown/Python/bash; there are no binaries, base64 blobs, or fetched opaque installers.

Before installing

Review `references/pipeline-orchestration.md` for the auto-deploy and recurring-run paths before using it on real models.
Test the tracking integrations (`wandb`, MLflow, Kubeflow, GCS) in a sandbox project with non-production credentials.
Confirm the runtime can write to `/models` and `/var/log`, or patch those paths to a project-local directory first.
Run `launch_distributed.sh` only in an isolated GPU environment; it will start `torchrun` on the host you give it.

Rated 2026-10-19 by an automated review of 6 files from the skill's repository.

Source

Repository: Jeffallan/claude-skills (all skills from this repository)
Path: skills/ml-pipeline/SKILL.md
Branch: main
Updated: 2026-09-19

fine-tuning-expert — Use when fine-tuning LLMs, training custom models, or adapting foundation models for specific tasks.
monitoring-expert — Configures monitoring systems, implements structured logging pipelines, creates Prometheus/Grafana dashboards, defines alerting rules.
advanced-evaluation — This skill should be used for advanced LLM evaluation: LLM-as-judge systems, direct scoring, pairwise comparison, rubric calibration.
dfam-check — Measure mesh files against Design for Additive Manufacturing (DfAM) rules and report printability findings per process (FDM, SLS, SLA/DLP, metal PBF, MJF).
evaluation — This skill should be used when building agent evaluation systems: deterministic checks, regression suites, multi-dimensional rubrics, quality gates.
gcode — Generate, inspect, dry-run, and statically validate plain FDM `.gcode` from 3D mesh files by orchestrating real slicer CLIs. Use when Codex needs to slice.