Borrowing it
Nothing to install: this file belongs to DuqueOM/ML-MLOps-Portfolio. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/DuqueOM/ML-MLOps-Portfolio/main/.devin/skills/drift-detection/SKILL.mdgit clone --depth 1 https://github.com/DuqueOM/ML-MLOps-PortfolioWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/duqueom/ml-mlops-portfolio/drift-detection)<a href="https://agentmods.dev/skills/duqueom/ml-mlops-portfolio/drift-detection"><img src="https://agentmods.dev/badge/skills/duqueom/ml-mlops-portfolio/drift-detection.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00027 | $0.02240 |
| Opus 5 | $0.00014 | $0.01120 |
| Sonnet 5 | $0.00005 | $0.00448 |
| Haiku 4.5 | $0.00003 | $0.00224 |
Grade A, and why
drift-detection scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- Bash(curl:*) How it starts
The opening of the file, as written. The whole thing — 232 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Drift Detection
Two complementary layers (ADR-006):
- Data drift (PSI on feature distributions) — early signal, no labels needed
- Concept drift (sliced AUC/F1 vs baseline, using delayed labels) — ground truth
Always investigate data drift FIRST (cheaper, faster). Escalate to concept drift analysis when (a) PSI alert fires and you need to confirm impact, or (b) a performance alert fires directly (AUC below threshold).
Step 1: Understand the Drift Metric
PSI Interpretation Guide
| PSI Value | Status | Action | Exit Code |
|---|---|---|---|
| < 0.10 | Stable | No action | 0 |
| 0.10 – 0.20 | Warning | Monitor, increase check frequency | 1 |
| > 0.20 | Alert | Trigger retraining | 2 |
ALWAYS use quantile-based bins (not uniform):
breakpoints = np.percentile(reference, np.linspace(0, 100, bins + 1))
Uniform bins can produce empty bins at extremes → PSI dominated by epsilon noise.
Special Cases — When PSI Doesn't Apply
| Feature Type | Problem with PSI | Alternative |
|---|---|---|
| Time series (seasonal) | PSI flags every seasonal change as "drift" | Year-over-Year comparison (same period last year) |
| Text/NLP features | PSI not meaningful for text | OOV (Out-of-Vocabulary) rate: warning > 20%, alert > 35% |
| Low-cardinality categorical | Quantile bins don't work with 3-5 categories | Categorical PSI variant: bins = unique categories |
| Boolean features | Only 2 bins → unstable PSI | Simple proportion test (chi-squared) |
Exit Codes for CronJob Integration
exit 0→ all features stableexit 1→ warning-level drift (monitor)exit 2→ alert-level drift (retraining needed, GitHub Issue created)
Step 2: Run Drift Detection Manually
python src/{service}/monitoring/drift_detection.py \
--reference data/reference/{service}_reference.csv \
--current data/production/{service}_latest.csv \
--output drift_report.json
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 232 lines · 27 tokens per session scan A fa5971133ba1
drift-detection is a skill published in the GitHub repository DuqueOM/ML-MLOps-Portfolio (5 stars, last pushed 6d ago), licensed MIT. It adds 27 tokens to every session and 2,240 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
mlflow
Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform.
setup
Configure MLflow tracing for Claude Code.
Cloud Security & Container Hardening
AWS/Azure/GCP security auditing, container and Kubernetes hardening, Infrastructure as Code scanning, and cloud compliance assessment.
ml-expert
Expert-level machine learning, deep learning, model training, and MLOps. Use when the user mentions machine learning, deep learning, neural networks, MLOps, or data science, or when the task involves Machine Learning Fundamentals, Data Preparation, or Model Training.
mlflow
Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform.
ecs-genai
Use whenever someone runs a GPU / ML / GenAI / LLM workload on Amazon ECS: GPU on ECS, ECS GPU-optimized AMI, g4dn/g5/g6/p4/p5 on ECS, which ECS launch type for GPU, Inferentia/Trainium/Neuron on ECS, distributed training, model inference or vLLM/Triton/TGI/Ray on ECS, Capacity Blocks, GPU sharing, ASG per GPU type.…