ML-MLOps-Production-Template

Recurring 23-domain enterprise/ISO repository audit — governance, traceability, security, supply chain, reproducibility, observability, architecture, technical debt, DX — produces a findings report with file:line evidence and risk ratings, and preserves the audit bar via anti-patterns Q-01…Q-08

Recurring 23-domain enterprise/ISO repository audit — governance, traceability, security, supply chain, reproducibility, observability, architecture, technical debt, DX — produces a findings report with file:line evidence and risk ratings, and preserves the audit bar via anti-patterns Q-01…Q-08 (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Review cloud costs against budget and identify optimization opportunities

Debug ML inference issues — latency spikes, wrong predictions, event loop blocking

Scan a service for edge-protection coverage (WAF, DDoS mitigation, rate limiting) against ADR-042/D-38 — produces a PASS/FAIL report with file:line evidence

Scan a service for edge-protection coverage (WAF, DDoS mitigation, rate limiting) against ADR-042/D-38 — produces a PASS/FAIL report with file:line evidence (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Systematic root-cause diagnosis for any non-ML-serving bug (CI, infra, agentic tooling, drift scripts) — reproduce, minimize, hypothesize, instrument, fix, regression-test

Structured, blameless post-incident review after /incident, /rollback, or /secret-breach closes — timeline reconstruction from primary sources, 5-whys root cause, action items with owner and date, written to docs/incidents/

Capture the ML problem specification — label definition, fairness-sensitive attributes, false-positive/negative cost, data source — BEFORE scaffolding, so quality_gates.yaml thresholds have a documented origin instead of unexamined defaults

Dual-axis change review — Standards (repo conventions, anti-patterns) and Spec (does the diff implement what the linked ADR/issue asked for) — evaluated in isolation so neither pass contaminates the other's verdict

Systematic root-cause diagnosis for any non-ML-serving bug (CI, infra, agentic tooling, drift scripts) — reproduce, minimize, hypothesize, instrument, fix, regression-test (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Structured, blameless post-incident review after /incident, /rollback, or /secret-breach closes — timeline reconstruction from primary sources, 5-whys root cause, action items with owner and date, written to docs/incidents/ (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Capture the ML problem specification — label definition, fairness-sensitive attributes, false-positive/negative cost, data source — BEFORE scaffolding, so quality_gates.yaml thresholds have a documented origin instead of unexamined defaults (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Dual-axis change review — Standards (repo conventions, anti-patterns) and Spec (does the diff implement what the linked ADR/issue asked for) — evaluated in isolation so neither pass contaminates the other's verdict (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Keep enterprise documentation coherent across the repo — version, CHANGELOG, ADRs, READMEs, schemas, llms.txt, and agentic surface counts — driven by check_doc_coherence.py (rule 16, ADR-031)

Verify GitHub Actions CI status before a promote/release/deploy action proceeds — read-only check, never an override

Deploy ML service to EKS with Kustomize overlays and IRSA

Deploy ML service to GKE with Kustomize overlays and Workload Identity

Verify GitHub Actions CI status before a promote/release/deploy action proceeds — read-only check, never an override (Mode: STOP — AGENTS.md Agent Behavior Protocol applies.)

Full release checklist for multi-cloud deployment (GCP + AWS)

Scaffold and run batch scoring jobs (CronJob + Parquet output) that reuse the service's model + feature-engineering code without opening the live API

Automated scan of a service/repo for compliance with AGENTS.md invariants D-01 through D-34 — produces a PASS/FAIL report with file:line evidence

Switch a scaffolded service between stack profiles (local, staging, prod)

Interview the adopter and emit a context file for their organization

Keep enterprise documentation coherent across the repo — version, CHANGELOG, ADRs, READMEs, schemas, llms.txt, and agentic surface counts — driven by check_doc_coherence.py (rule 16, ADR-031) (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Switch a scaffolded service between stack profiles (local, staging, prod) (Mode: CONSULT — AGENTS.md Agent Behavior Protocol applies.)

Interview the adopter and emit a context file for their organization (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Create a complete new ML service from template — end-to-end scaffolding

End-to-end RCA for a performance-degradation incident — correlates sliced metrics, drift, deploy history, upstream data changes, and prediction logs into one evidence-backed root cause.

Update an existing scaffolded service with the latest template changes via copier update

Automated scan of a service/repo for compliance with AGENTS.md invariants D-01 through D-34 — produces a PASS/FAIL report with file:line evidence (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Update an existing scaffolded service with the latest template changes via copier update (Mode: CONSULT — AGENTS.md Agent Behavior Protocol applies.)

Scaffold and run batch scoring jobs (CronJob + Parquet output) that reuse the service's model + feature-engineering code without opening the live API (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Root-cause a performance alert using sliced metrics + ground-truth (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Review cloud costs against budget and identify optimization opportunities (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Debug ML inference issues — latency spikes, wrong predictions, event loop blocking (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Deploy ML service to EKS with Kustomize overlays and IRSA (Mode: CONSULT — AGENTS.md Agent Behavior Protocol applies.)

Deploy ML service to GKE with Kustomize overlays and Workload Identity (Mode: CONSULT — AGENTS.md Agent Behavior Protocol applies.)

Run and interpret DATA drift (PSI) AND CONCEPT drift (sliced performance) for an ML service (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)

Run 6-phase exploratory data analysis on a new dataset — ingest, profile, univariate, correlations, leakage gate, feature proposals (Mode: AUTO — AGENTS.md Agent Behavior Protocol applies.)