Airflow and Workflow Orchestration

Wymusza wybór modelu, idempotentne ponowienia i bramę publikacji przed kodem DAG

Autor: vaquarkhan · vaquarkhan/data-engineering-agent-skills

Testowano · Działa ★ 8.4/10

Airflow and Workflow Orchestration — Wymusza wybór modelu, idempotentne ponowienia i bramę publikacji przed kodem DAG

Co robi ten skill

Lista kontrolna projektowania orkiestracji potoków danych w Airflow, MWAA, Cloud Composer, Step Functions, Azure Data Factory i Databricks Workflows. Zmusza do ustalenia workflow contract, orchestration model, retry idempotency, backfill path i publish gate before writing DAG code, and triggers on questions about dependencies, sensors, retries, SLAs, backfills, or which orchestrator to use.

Raport z testu

Zaprojektowano ten sam 40-tenant S3-to-Redshift pipeline twice. The cold answer reached for a stock MWAA DAG: forty S3 sensors, a Python loop building forty TaskGroups, dbt in a BashOperator, retries=3 on every task, and catchup=True for the 90-day backfill. The skill's six steps overturned five of those - it routed the workload to metadata-driven fan-out off a control table instead of copy-pasted TaskGroups, replaced the sensor fleet with an event trigger, caught that retrying a partial Redshift COPY duplicates rows, split the backfill onto its own throttled entry point with reconciliation, and reordered the run so the load lands in staging and only promotes after the quality gate. Its 'Common Rationalizations' table names the two traps the cold answer walked into verbatim. One snag: SKILL.md points at references/pipeline-orchestration-patterns.md, which lives at the repo root, not beside the skill, so a plain copy of the skill folder silently loses it.

Testowano: 2026-07-30 · Claude Code 2.x (agent harness)

Instalacja

git clone --depth 1 https://github.com/vaquarkhan/data-engineering-agent-skills.git
mkdir -p ~/.claude/skills/airflow-and-workflow-orchestration/references
cd data-engineering-agent-skills && cp skills/airflow-and-workflow-orchestration/SKILL.md ~/.claude/skills/airflow-and-workflow-orchestration/ && cp references/pipeline-orchestration-patterns.md ~/.claude/skills/airflow-and-workflow-orchestration/references/

Komendy i przykładowe prompty

/airflow-and-workflow-orchestrationWymusza wybór modelu, idempotentne ponowienia i bramę publikacji przed kodem DAG

Skille uruchamiają się na zwykłe polecenia — bez komend do zapamiętania. Po instalacji aktywują go prompty takie jak te (po angielsku):

Should I use Step Functions or MWAA for this 40-tenant ingestion pipeline?
Add retries and a backfill path to my Airflow DAG without duplicating rows
Design the dependency graph for our ingestion-to-publish data pipeline

Pełny raport →