Data Orchestration Skill for Claude

Expert guidance for orchestrating data pipelines. Dagster-first for greenfield projects, Airflow for brownfield. Covers scheduling, dependencies, monitoring, retries, alerting, and dbt/DLT integration.

When to Use This Skill

Activate when:

• Choosing between orchestration tools (Dagster vs Airflow vs Prefect vs embedded)
• Building Dagster assets, resources, sensors, schedules, or partitions
• Writing Airflow DAGs, operators, TaskFlow tasks, or connections
• Integrating orchestrators with dbt or DLT
• Implementing retry logic, alerting, failure handling, or partitioned backfills
• Deciding whether you need an orchestrator at all

Don't use for: dbt model writing (use dbt-skill), DLT source/destination config (use integration-patterns-skill), Kafka/Flink streaming (use streaming-data-skill), IaC provisioning, or CI/CD pipelines.

Scope Constraints

• Generate orchestration code, DAG definitions, asset configs, and scheduling logic only.
• Credential management: reference environment variables and secrets managers. Never hardcode secrets. See Security & Compliance Patterns.
• Limit scope to orchestration concerns. Hand off transformation logic to dbt-skill and ingestion logic to integration-patterns-skill.

Model Routing

reasoning_demand	preferred	acceptable	minimum
medium	Sonnet	Opus, Haiku	Haiku

Core Principles

1. Assets over tasks — Define persistent data artifacts (tables, views, models), not computation steps. Asset lineage is visible, dependencies declarative, backfills targeted, freshness observable.
2. Idempotent runs — Use MERGE/upsert, partition by date/key, track state externally. Every run must be safe to re-execute.
3. Declarative dependencies — Declare dependencies; the framework resolves execution order.
4. Observable pipelines — Log metadata (row counts, schema changes, execution times). Monitor freshness SLAs. Trace lineage from source to dashboard. Alert with actionable context.
5. Graceful failure — Retry with backoff for transient failures. Re-run only failed assets. Capture failed records via dead letter patterns. Group alerts to prevent fatigue.

---

Orchestrator Decision Matrix

Factor	Dagster	Airflow	Prefect	Embedded
Philosophy	Asset-oriented	Task-oriented	Flow-oriented	Tool-native
Best for	Greenfield platforms	Brownfield, large DAGs	Python-native, event-driven	Single-tool workflows
dbt integration	dagster-dbt (first-class)	cosmos (good)	CLI wrapper	Native (dbt Cloud)
DLT integration	dagster-dlt (first-class)	Task wrapper	Task wrapper	N/A
Asset lineage	Built-in, UI-native	Via plugins (limited)	Via artifacts	Tool-specific
Partitioning	First-class	Dynamic task mapping	Map/reduce	Tool-specific
Local dev	dagster dev (full UI)	Local executor (limited)	prefect server start	N/A
Managed offering	Dagster Cloud	MWAA, Composer, Astronomer	Prefect Cloud	Built into platform

See Embedded Orchestration Reference for dbt Cloud, Databricks Workflows, Snowflake Tasks, and Prefect patterns.

---

The Trifecta: Dagster + DLT + dbt

Each tool handles one concern:

Layer	Tool	Responsibility
Orchestration	Dagster	Scheduling, dependency resolution, monitoring, alerting
Ingestion	DLT	Extract from sources, load to warehouse (raw layer)
Transformation	dbt	SQL transformations (staging, intermediate, marts)

┌─────────────────────────────────────────────────────┐
│                  Dagster (Orchestrator)              │
│  ┌───────────┐     ┌────────────┐     ┌──────────┐ │
│  │ DLT Assets│ ──→ │ dbt Assets │ ──→ │ Quality  │ │
│  │ (ingest)  │     │ (transform)│     │ (assert) │ │
│  └───────────┘     └────────────┘     └──────────┘ │
│       ▼                  ▼                  ▼       │
│  ┌──────────────────────────────────────────────┐   │
│  │         Snowflake / BigQuery                 │   │
│  │  raw.* → staging.* → intermediate.* → marts.*│  │
│  └──────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────┘

Key benefits: DLT assets auto-depend on source availability. dbt assets auto-depend on raw tables DLT creates. Dagster shows complete lineage from API to marts. Backfills target specific date ranges across both DLT and dbt. Failures in DLT prevent dbt from running on stale data.

For full trifecta code example, see Dagster Integrations Reference.

---

Dagster Quickstart

pip install dagster dagster-webserver
dagster dev -f my_pipeline.py        # Full UI at http://localhost:3000

For asset examples (basic, resources, schedules, sensors), see Dagster Patterns Reference. For dbt and DLT integrations, see Dagster Integrations Reference.

---

Airflow Quickstart

Airflow stores credentials in Connections — never in DAG code. Configure via Airflow UI, environment variables (AIRFLOW_CONN_*), or secrets backends (Vault, AWS SM, GCP SM).

For TaskFlow API, classic operator pattern, dynamic task mapping, cosmos dbt integration, and managed Airflow (MWAA/Composer/Astronomer), see Airflow Patterns Reference.

---

Common Patterns

Retry Strategy

# Dagster
@asset(retry_policy=RetryPolicy(max_retries=3, delay=30, backoff=Backoff.EXPONENTIAL))
def flaky_api_asset():
    return call_external_api()

# Airflow
default_args = {"retries": 3, "retry_delay": timedelta(minutes=5),
                "retry_exponential_backoff": True, "max_retry_delay": timedelta(minutes=30)}

Alerting

# Dagster: freshness policies
@asset(freshness_policy=FreshnessPolicy(maximum_lag_minutes=120))
def fct_orders(stg_orders): ...
# Configure Slack/PagerDuty alerts in Dagster Cloud or via dagster-slack

# Airflow: failure callbacks
def on_failure_callback(context):
    send_slack_alert(channel="#data-alerts",
        message=f"Task {context['dag'].dag_id}.{context['task'].task_id} failed on {context['ds']}")

Partitioned Backfills

# Dagster: daily partitions
daily_partitions = DailyPartitionsDefinition(start_date="2024-01-01")

@asset(partitions_def=daily_partitions)
def daily_orders(context):
    date = context.partition_key
    orders = fetch_orders(date=date)
    load_to_warehouse(orders, partition=date)
# Backfill via UI (select range → Materialize) or CLI:
# dagster asset materialize --partition "2024-01-01...2024-01-31"

---

Security Posture

This skill generates orchestration code including DAG definitions, asset configurations, and scheduling logic. See Security & Compliance Patterns for the full security framework.

Credentials required: Warehouse connections, API keys, secrets manager access, alerting webhooks Where to configure: Dagster EnvVar resources, Airflow Connections, environment variables Minimum role/permissions: Orchestrator service account with scoped warehouse access

Capability	Tier 1 (Cloud-Native)	Tier 2 (Regulated)	Tier 3 (Air-Gapped)
Execute pipelines	Against dev/staging	Generate code for review	Generate code only
Configure schedules	Deploy to dev	Generate configs for review	Generate configs only
Manage connections	Configure dev connections	Generate templates	Document requirements
Backfill data	Execute against dev	Generate plans for review	Generate plans only

---

Reference Files

• Dagster Patterns — Assets, resources, sensors, schedules, partitions, backfills, I/O managers, asset checks
• Dagster Integrations — dagster-dbt, dagster-dlt, trifecta example, dagster-k8s, dagster-cloud
• Airflow Patterns — TaskFlow API, operators, connections, dynamic task mapping, cosmos, MWAA/Composer
• Embedded Orchestration — When NOT to add an orchestrator: dbt Cloud, Databricks Workflows, Snowflake Tasks, Prefect
• Consulting Orchestration — Recurring engagement patterns, file-drop sensors, client-specific scheduling, per-client dashboards