Estuary Agent Skills

Estuary Agent Skills let you create and operate Estuary pipelines from an AI coding assistant. Each skill is a small instruction set that teaches your assistant how to run a specific flowctl workflow, such as creating a Postgres capture, materializing into Snowflake, or diagnosing a failing task.

Skills are distributed from the open-source estuary/agent-skills repository and follow the open SKILL.md standard. They work with Claude Code, Cursor, OpenAI Codex, GitHub Copilot, Gemini CLI, and other compatible tools.

Prerequisites

Before installing the skills, make sure you have:

• An Estuary account.
• The flowctl CLI installed and authenticated. See the flowctl installation guide. The estuary-flowctl-setup skill can also guide your assistant through installation and authentication on demand.
• A supported AI coding assistant, such as Claude Code, OpenAI Codex, or Cursor.

What's included

The skills are grouped into three plugins: captures, materializations, and operations.

Captures (sources)

Capture data from databases, APIs, and webhooks into Estuary collections. The generic skill can configure any connector in the Estuary connector catalog.

Skill	Description
capture-postgres-create	PostgreSQL CDC (vanilla, RDS, Aurora, Cloud SQL, Supabase, Neon)
capture-mysql-create	MySQL CDC via binlog replication (RDS, Aurora, Cloud SQL, Azure)
capture-mongodb-create	MongoDB CDC (Atlas, DocumentDB, self-hosted)
capture-sqlserver-create	SQL Server CDC (RDS, Azure SQL, Cloud SQL)
capture-http-ingest-create	HTTP webhook capture (GitHub, Shopify, Stripe, or any JSON source)
capture-generic-create	Any of 150+ source connectors via dynamic schema discovery

Materializations (destinations)

Stream Estuary collections into downstream databases and warehouses.

Skill	Description
materialize-postgres-create	PostgreSQL destination
materialize-snowflake-create	Snowflake destination (JWT auth)
materialize-bigquery-create	BigQuery destination (GCS staging)
materialize-redshift-create	Amazon Redshift destination (S3 staging)
materialize-databricks-create	Databricks destination (Unity Catalog)
materialize-generic-create	Any destination connector via dynamic schema discovery

Operations

Manage and troubleshoot running pipelines.

Skill	Description
estuary-flowctl-setup	Install, authenticate, and update the flowctl CLI
estuary-task-health	End-to-end health check for a task: status, data flow, errors, and history
estuary-catalog-status	Check the control-plane status of a task with flowctl catalog status
estuary-task-stats	Inspect data volume, document counts, and hourly throughput for a task
estuary-catalog-history	Review publication history and recent spec changes
estuary-logs	Search and analyze task logs with jq filtering
estuary-connector-restart	Pause and restart connectors via shard management
estuary-ssh-tunnels	Diagnose and fix SSH tunnel connection issues

Installation

Installation steps depend on which assistant you use. After installing, restart your assistant if it doesn't pick up the new skills automatically.

Claude Code

Claude Code includes a plugin marketplace. Add the Estuary marketplace:

/plugin marketplace add estuary/agent-skills

Then install the plugins you want:

/plugin install estuary-captures@estuary
/plugin install estuary-materializations@estuary
/plugin install estuary-operations@estuary

You can also run /plugin and browse skills from the Discover tab. Installed skills update automatically when the marketplace refreshes.

To verify, ask Claude: "What Estuary skills do you have?" and confirm the relevant skills are listed.

OpenAI Codex

Codex reads skills from a .codex/skills/ directory in your project, or globally from ~/.codex/skills/. Clone the repo and copy the skills:

git clone https://github.com/estuary/agent-skills.git
mkdir -p .codex/skills
cp -r agent-skills/skills/* .codex/skills/

You can also install everything with the cross-tool Skills CLI:

npx skills add estuary/agent-skills

Start a new Codex session and ask: "capture Postgres into Estuary". Codex should load the matching skill.

Cursor

Cursor reads skills from a .cursor/skills/ directory in your workspace. Clone the repo and copy the skills:

git clone https://github.com/estuary/agent-skills.git
mkdir -p .cursor/skills
cp -r agent-skills/skills/* .cursor/skills/

Or use the Skills CLI:

npx skills add estuary/agent-skills

Reload Cursor and ask: "set up an Estuary capture from MySQL" to confirm the skills are wired up.

Other tools

The skills are plain SKILL.md files. Any tool that supports the open SKILL.md standard can use them. Common directory conventions:

Tool	Skills directory
Claude Code	.claude/skills/
Cursor	.cursor/skills/
GitHub Copilot / VS Code	.github/skills/
OpenCode	.opencode/skills/
Codex	.codex/skills/

Install across any of them with the Skills CLI:

npx skills add estuary/agent-skills

Or clone the estuary/agent-skills repo and copy skills/* into your tool's skills directory.

Walkthrough: your first pipeline

Once a skill group is installed and flowctl is authenticated, you can drive Estuary from natural-language prompts. A typical first run looks like this.

1. Confirm flowctl is ready

Ask your assistant:

"Check that flowctl is installed and I'm authenticated to Estuary."

This invokes the estuary-flowctl-setup skill, which checks your CLI version, refreshes your auth token, and points you at the flowctl installation docs if anything is missing.

2. Create a capture

Describe the source:

"Capture my Postgres database into Estuary. The host is db.example.com, database analytics, and I want all tables in the public schema."

The matching capture-*-create skill discovers your schema, drafts a capture spec, guides you through replication slot setup, and publishes the capture with flowctl.

3. Materialize into a destination

"Materialize the collections I just captured into Snowflake."

The materialize-snowflake-create skill collects your warehouse, account, and role, generates the materialization spec, and publishes it. The same pattern works for BigQuery, Redshift, Databricks, Postgres, or any other destination through materialize-generic-create.

4. Troubleshoot and operate

When something goes wrong, ask:

"Why is my materialization to Snowflake failing?"

"Is data flowing through my Postgres capture?"

"Show me what changed on this task in the last week."

These prompts route to the operations skills (estuary-task-health, estuary-task-stats, estuary-logs, estuary-catalog-history), which pull data from flowctl and summarize the results.

If a skill itself misbehaves, file an issue on the agent-skills repo or reach out in the Estuary Slack community.

Example prompts

Use the skills for any task you would normally run by hand:

• "Capture from MySQL and materialize into Redshift."
• "Set up a webhook capture for Shopify orders."
• "Run a health check on my acme/postgres-prod capture."
• "My capture is failing with an SSH tunnel error. Help me debug it."
• "Restart the materialization to BigQuery."

For docs-aware Q&A from the same assistants (without the action skills), see MCP Integration.