Top Claude Skills for Data Engineers (2026 Guide)

What Claude skills are (and what they are not)

A skill is a folder with a SKILL.mdfile inside it. That file teaches Claude how to do one job well: structure a dbt model the way your team actually structures models, or debug a failed Airflow task without six rounds of back and forth. The clever part is progressive disclosure. Claude only loads the skill's name and one-line description until your task actually matches it. So you can install a dozen skills and pay almost no context cost until the moment one is needed. If your team is still working out which skills to standardise on, that is exactly what we cover in our Claude Code training.

People confuse skills with two neighbouring things, so it is worth being exact. Skills teach. MCP servers connect. Slash commands trigger.A skill is knowledge and procedure. An MCP server is a live connection to an external system: your Airflow instance, your warehouse, your lakehouse catalogue. It gives Claude real data and actions. A slash command is a prompt template you fire manually. Astronomer's own repo is a clean example of pairing skills with a connection: Airflow skills for the procedure, an MCP server (and a CLI called af) for the live DAG state.

How to install any of these (three routes, not one)

Not every skill below installs the same way, so check which route the repo actually documents rather than assuming. dbt Labs and Astronomer both ship proper Claude Code plugins, which need a .claude-plugin/marketplace.json file in the repo. Inside Claude Code:

# dbt Labs
/plugin marketplace add dbt-labs/dbt-agent-skills
/plugin install dbt@dbt-agent-marketplace

# Astronomer (the plugin, not the standalone skills installer,
# bundles the Airflow MCP server the operational skills need)
claude plugin marketplace add astronomer/agents
claude plugin install astronomer-data@astronomer

TerminalSkills and the three generalist repos (borghei, travisjneuman, antigravity-awesome-skills) do not ship a marketplace file, so there is no /plugin install for them. Install by hand instead: clone or copy the skill folder into .claude/skills/<name>/ in your project, or ~/.claude/skills/<name>/to make it available everywhere, then reload Claude Code. Each repo's own README gives the exact path, so check it before copying rather than guessing.

The 7 best Claude skills for data engineers

Ranked by a blend of trust (first-party beats anonymous), usefulness for real pipeline work, and how actively maintained the skill is as of October 2026. We checked every repo below for a real, recent commit before including it, and flagged licensing honestly.

1. dbt-agent-skills: dbt from the source

dbt Labs publishes its own skill collection for Claude Code, and for dbt work it is the obvious starting point. The set covers analytics engineering (building and modifying models, writing tests, exploring unfamiliar sources), the semantic layer (metrics, dimensions and MetricFlow), dbt Mesh for multi-project setups with contracts and versions, platform operations (troubleshooting job failures, configuring the dbt MCP server), and a separate migration pack for moving a project from dbt Core to Fusion (dbt Labs' own skill calls this "v2" internally; the product itself is named Fusion). One detail worth knowing: the analytics-engineering skill will stop and route you to the Mesh skill if it detects a breaking change to a model with downstream consumers, rather than letting Claude edit it in place. That is the kind of guard rail you want baked in, not bolted on. The limitation: it only covers dbt. If your pipeline is Airflow-orchestrated and dbt is one task inside it, you still need a separate orchestration skill alongside this one.

2. agents: Airflow from the people who run it commercially

Astronomer builds managed Airflow for a living, and this repo is its skill set plus an MCP server and an af CLI, all Apache-2.0 and compatible with open-source Airflow, not just Astro. The core airflow skill handles day-to-day operations: listing and triggering DAGs, reading task logs, diagnosing import and parse errors, checking connections and pools. Sibling skills handle the rest of the lifecycle: testing-dags for iterative test-debug-fix cycles, checking-freshness and tracing-downstream-lineage for data observability, and cosmos-dbt-core / cosmos-dbt-fusionfor turning a dbt project into an Airflow DAG via Astronomer Cosmos. If you run Airflow and dbt together, this is the skill that actually bridges them. The limitation: the operational skills lean on Astronomer's own af CLI and MCP server for live state. Without those installed, the DAG-authoring skills still work, but triggering runs and reading task logs through Claude loses most of its value.

3. data-engineering-skills: the dbt and Snowflake specialist pack

AltimateAI (the team behind the dbt-focused Datapilot and Turntable tooling) publishes seven dbt skills (model creation, debugging, testing, documentation, migration, refactoring, incremental models) and three Snowflake skills (cost analysis, and query optimisation either by query ID or by pasted SQL text), plus a delegation skill for its own altimate-code CLI. The README reports benchmark results against ADE-bench, a public 43-task dbt benchmark, and against a 62-query TPC-H Snowflake test; we have not independently reproduced those numbers, so treat them as the vendor's own claim rather than a verified figure, but the skill content itself is genuinely dbt- and Snowflake-specific rather than generic advice. There is also a Databricks skill in the same pack, optimizing-databricks-sql, which layers Delta, Photon and Unity Catalog knowledge (Z-ordering, table statistics, Photon UDF blocking) on top of the same query-analysis pipeline the Snowflake skills use.

4. great-expectations: data quality that does not care which tool you use

TerminalSkills maintains an open catalogue of Agent Skills following the agentskills.io standard, and its Great Expectations skill is a properly current one: it correctly flags that GX 1.0 was a ground-up rewrite (the old great_expectations init CLI-driven workflow is gone, replaced by an in-code context), and targets the current GX Core 1.23 API. It walks through the Data Source to Data Asset to Batch Definition to Batch flow, then Expectation Suite, Validation Definition and Checkpoint, so Claude generates code against the framework as it actually works today rather than a tutorial from two major versions ago. The limitation: it only validates what you explicitly define as an expectation. It will not tell you what checks you forgot to write, and wiring up the Data Source, Batch and Checkpoint chain for a brand new project takes real setup time before your first validation run.

5. senior-data-engineer: one skill, broad coverage

If you want a single generalist skill instead of several specialist ones, this is a solid pick: it generates Airflow, Prefect and Dagster pipeline code, validates data quality with profiling and anomaly detection (including Great Expectations suite generation), and analyses SQL and Spark performance with partition and cost recommendations. One thing to flag before you install it commercially: the licence is MIT with the Commons Clause added, which restricts selling the software itself as a service. That does not affect using it as a Claude skill inside your own team, but read the exact terms if your use case is unusual. This is the same repo as the runbook-generator skill we feature in the Platform Engineers and SREs guide, a different skill from the same author covering a different job, which is why it belongs in both lists.

6. data-engineering: a fast reference for the big architecture calls

A lighter, cleanly MIT-licensed skill that is less about generating code and more about getting the architecture decision right first: when to use ETL versus ELT, which orchestrator fits (Airflow, Dagster or Prefect), warehouse schema design across BigQuery, Snowflake and Redshift, and streaming patterns with Kafka or Spark Streaming. Good complement to a code-generation skill like dbt-agent-skills or senior-data-engineer rather than a replacement for one. The limitation: it is reference material, not a generator. It is good for talking through an architecture decision, but you will still reach for one of the other skills on this list when you want Claude to actually write the pipeline code.

7. data-engineer (antigravity-awesome-skills): useful, but read the warning label first

A genuinely substantial skill covering lakehouse architecture with Delta Lake, Apache Iceberg and Apache Hudi, alongside the usual batch and streaming pipeline guidance. It earns the last spot for an honest reason rather than a content problem: the registry that catalogues it tags its own metadata with risk: critical, the same self-rating the registry applies across its wider community collection. That is not proof of anything malicious, but it is exactly the kind of signal the security section below is about: read the SKILL.md yourself before you decide whether that rating matters for your setup.

Worth watching (and what we left off)

A few things nearly made the cut, or are worth knowing about even though we did not rank them. dbt-labs' dbt-migration plugin (same repo as #1, installed separately) is a one-off tool for moving a project from dbt Core to Fusion rather than a day-to-day skill, which is why we covered it as part of #1 instead of listing it on its own. Astronomer's warehouse-init and analyzing-data skills(same repo as #2) are genuinely useful for cold-starting a new warehouse connection and running SQL analytics on Airflow's own metadata tables, but are narrower than the core airflowskill. And if you want to browse further, majiayu000's claude-skill-registry is a searchable catalogue of thousands of community skills, including several more data-warehousing options, but it is a mirror of other people's work rather than an author in its own right. Follow its source links and read the original repo before installing anything from it.

A word on security before you install anything

Skills are code, and this warning carries extra weight for this audience. A SKILL.mdcan contain prompt-injection instructions, and any script a skill bundles runs with your agent's permissions: for a data engineer that can mean Snowflake, BigQuery or Redshift credentials and a live line into production tables. Snyk's 2026 ToxicSkills research catalogued thousands of skills and found 36.8% had at least one security flaw and 13.4% had a critical-level issue, with roughly one in nine containing hardcoded or exposed secrets and 91% of confirmed malicious skills using prompt injection.

How to actually use these together

You do not need all seven. The strongest loadout for most data engineers is three or four skills covering transform, orchestrate and validate:

1. dbt-agent-skills for model building, testing and the semantic layer if you run dbt.
2. Astronomer's agents skills for Airflow DAG authoring, testing and debugging, and the Cosmos skill if dbt and Airflow need to talk to each other.
3. great-expectations for data quality checks that sit outside either tool.
4. AltimateAI's Snowflake skills added on top if that is your warehouse and query cost is a live concern.

Reach for one of the three generalist skills (senior-data-engineer, travisjneuman's data-engineering, or antigravity-awesome-skills' data-engineer) only if you specifically want one broad skill instead of several narrow ones, for example on a smaller team without a dedicated orchestration or transformation tool yet.

The short version

Skills are the difference between an AI that produces a dbt model which runs once on your laptop and one that produces a pipeline you would trust with a production warehouse. If you install nothing else today, install dbt-agent-skills or the Astronomer agents skills, whichever matches your stack, and read their SKILL.md files first. Validate every AI-generated model and DAG against a staging warehouse before it touches production. The ecosystem moves fast; the principle does not. Constrain the model with good skills and you get fewer pipelines that looked fine in testing and quietly broke a downstream dashboard.

If your pipelines feed a model rather than a dashboard, our companion guide covers the top Claude skills for ML/AI engineers, and if the API layer sits in front of the warehouse rather than the pipeline itself, see the top Claude skills for backend engineers. If your team also owns the infrastructure these pipelines run on, see the top Claude skills for platform engineers and SREs, and if it owns provisioning and deployment more broadly, see the top Claude skills for DevOps engineers.