Works with the AI Agents you already use
data science ml specialist
Staff-level ML engineer skill for rigorous tabular data science, leakage prevention, and production-ready pipelines.
Free
data science ml specialist
Example session with this skill installed
Senior mode: Build a model to predict customer churn using the provided CSV. The data contains multiple entries per customer ID and is time-ordered.
- Read your context and instructions
- Compiled the data science ml
I have initiated Senior mode for the churn prediction project. I have detected repeated customer IDs and temporal ordering, so I will use a GroupShuffleSplit with a temporal cutoff to prevent leakage.
Project Initiation
I have scaffolded the project structure and completed the initial data audit.
| Metric | Value |
|---|---|
| Initial Row Count | 50,000 |
| Unique Customers | 12,400 |
| Exact Duplicates Removed | 450 |
| Target Imbalance (Churn=1) | 18.2% |
| Split Strategy | Temporal Grouped (80/20) |
| Test Set SHA256 | e3b0c44298fc1c149afbf4c8996fb924... |
# dstk/pipeline_config.py
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScalar
from lightgbm import LGBMClassifier
import dstk
# Pre-declared selection rule:
# Simplest model within 1 std dev of best CV score.
pipeline = Pipeline([
('preprocess', dstk.preprocess.ChurnTransformer()),
('clf', LGBMClassifier(random_state=42, class_weight='balanced'))
])
Next steps
- Review the
reports/project_brief.mdfor target definition alignment. - Run
python scripts/train.pyto execute the baseline ladder. - Check
reports/data_audit.mdfor the full list of detected feature correlations.
Connects securely to your tools. The creator never sees your data.
About this skill
Stop relying on unverified metrics and leaky models. This skill enforces a rigorous, staff-level data science workflow that prioritizes decision-making over hype. It prevents the most common causes of ML failure, like data leakage, duplicate rows in splits, and silent training-serving skews.
What it does
- Rigorous evaluation splits data correctly for time-series, grouped, or stratified sets and freezes a test set with SHA256 integrity checks.
- Leakage prevention detects target leakage through label-shuffling, adversarial validation, and split-gap analysis.
- Baseline ladders mandates starting with trivial heuristics and simple models before moving to complex tree ensembles.
- Automated auditing generates data quality reports, row-count provenance, and integrity-stamped results that cannot be hand-edited.
- Production readiness builds pipelines that wrap all learned transforms and includes parity tests to ensure training and inference match.
How it works
- Mode selection starts in Lite by default for quick snippets or switches to Senior mode for full end-to-end lifecycle management.
- Problem framing generates a Project Brief focusing on the business decision, unit of analysis, and success metrics.
- Execution runs a tiered experiment log, performing data audits, pipeline construction, and model selection based on pre-declared rules.
- Validation executes
pytestsuites for leakage and regression while linting reports to remove hype and verify number provenance.
Frameworks & tools
Python-centric stack using scikit-learn, LightGBM, XGBoost, CatBoost, Optuna, pytest, Polars, DuckDB, and FastAPI.
Why this beats prompting it yourself
Standard prompts often ignore data leakage or use incorrect split strategies for time-series data. This skill forces a Pipeline architecture and uses a dedicated dstk helper library to ensure every metric is statistically significant and every result is reproducible.
Use cases
- Forecasting & Tabular ML for building regression or classification models with temporal integrity.
- Model Debugging to diagnose why a model overfits or fails to generalize in production.
- Ranking & Recommendations using specialized metrics and grouped evaluation strategies.
- A/B Test Analysis applying bootstrap intervals and paired tests to compare variants.
Known limitations
Does not support deep learning fine-tuning, computer vision, audio processing, or pure SQL/ETL work. Binary model artifacts like pickles should only be loaded from trusted sources.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
No support for deep learning, computer vision, or audio. Requires project-specific hooks for full test suite coverage. No hand-editing results.json.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 7 days ago
- Free with an Agensi account
- Software
Git