메뉴
Apache Airflow DAG development, testing, performance tuning, security, and operations. Use when writing or reviewing DAGs, configuring Airflow, optimizing scheduler performance, setting up CI/CD for DAGs, managing secrets backends, or migrating to Airflow 3.x.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Shape answers so they are immediately actionable instead of long. Use on any response that would otherwise run long — multi-step tasks, explanations, plans, reviews with many findings, or answers that hand back commands, paths, or snippets to run. Leads with the next action, numbers steps, restates progress, caps lists, and gives concrete time estimates. Complements caveman (which handles tone); this handles shape.
Senior software architect for system design, ADRs, microservices evaluation, technology trade-off analysis. Use when designing system architecture, evaluating patterns, creating ADRs, analyzing technology choices, or planning migrations.
Go CLI development with Cobra command hierarchy, Bubbletea interactive TUI, Lipgloss terminal styling, Charmbracelet ecosystem (Bubbles, Huh forms, Glamour markdown rendering), and shell completions. Use when building CLI tools, adding subcommands, creating interactive TUI, styling terminal output, or implementing shell completions.
Cloud architecture design, migration, optimization, and disaster recovery across AWS and Azure. Use when designing VPCs/VNets, IAM/RBAC, compute/storage architectures, cost optimization, multi-cloud patterns, or planning migrations. Both Terraform and CLI approaches.
Datadog monitoring strategy, alert design, tagging governance, dashboard patterns, SLO strategy, log management, cost optimization, and cross-pillar correlation workflows. Use when making architectural decisions about what to monitor, how to alert, how to tag, or how to structure Datadog for a service or platform. Delegates execution to pup CLI and infrastructure-as-code to sre-engineer.
Kubernetes workload management, Helm chart development, cluster troubleshooting, custom operators, configuration management, and multi-cluster administration. Use when deploying workloads, writing Helm charts, debugging pods, creating operators, managing ConfigMaps/Secrets/RBAC, or working with multi-cluster setups.
SOC 직업 분류 기준
| name | airflow-engineer |
| description | Apache Airflow DAG development, testing, performance tuning, security, and operations. Use when writing or reviewing DAGs, configuring Airflow, optimizing scheduler performance, setting up CI/CD for DAGs, managing secrets backends, or migrating to Airflow 3.x. |
| metadata | {"version":"1.0.0","domain":"data-engineering","triggers":"airflow, DAG, dag_id, data pipeline, task idempotency, airflow scheduler, airflow secrets, airflow testing, dag.test, airflow 3, airflow migration, airflow security, airflow performance, parse time, min_file_process_interval, TaskGroup, SubDAG, data aware scheduling, assets, deadline alerts","role":"specialist","scope":"implementation","output-format":"code","related-skills":"sre-engineer, datadog-advisor, devsecops-expert, secrets-management, ci-cd"} |
Tasks are transactions. Each task completes entirely or not at
all. Use UPSERT, partition by data_interval_start, never
datetime.now(). Re-runs must produce identical results.
Parse time is the bottleneck. Top-level DAG code runs every
min_file_process_interval seconds. No DB calls, no network, no
Variable.get() at module level. Defer to Jinja or execution time.
DAG authors are trusted users. They execute arbitrary code on workers, DAG File Processor, and Triggerer. Treat DAG deployment like code deployment: PR reviews, static analysis, gated CI/CD.
| Topic | Reference |
|---|---|
| Task design, idempotency, atomicity, granularity | dag-design.md |
Testing strategies, dag.test(), CI/CD pipelines | testing-cicd.md |
| Scheduler tuning, parse-time hygiene, worker scaling | performance-tuning.md |
| Secrets backends, trust model, DAG author security | security-config.md |
| Airflow 3.x breaking changes, migration checklist | airflow3-migration.md |
| Anti-Pattern | Fix |
|---|---|
Variable.get() at module level | Use {{ var.value.get('foo') }} in templates |
INSERT in tasks | Replace with UPSERT for idempotency |
datetime.now() in task logic | Use data_interval_start / data_interval_end |
| Monolithic extract-transform-load task | Split into one task per logical step |
| SubDAGs (removed in Airflow 3) | Use TaskGroup or Assets |
| Direct DB import in task code | Use Airflow Python Client or DB hooks |
| Many DAGs per file | One DAG per file for scalability |
| Deeply nested DAG structures | Prefer linear A -> B -> C chains |
| Removed | Replacement |
|---|---|
| SubDAGs | TaskGroups, Assets, Data Aware Scheduling |
| SLAs | Deadline Alerts (3.1, AIP-86) |
| Direct metadata DB access in tasks | Execution API (workers have no DB creds) |
from airflow.models import ... in tasks | Airflow Python Client or DB hooks |
