Vibe coding with skills¶

If you want your AI coding assistant to understand Kedro conventions out of the box (correct dataset naming, catalog patterns, layer structure), install kedro-skills. It writes contextual guidance files into your project that activate automatically when you edit matching files. No server, no configuration, works in any IDE.

Quick install¶

pip install kedro-skills

Quick start¶

# Inside a Kedro project:
kedro skills list                      # see what's available
kedro skills install catalog-config    # install the catalog guidance skill

After running this, your AI assistant will give better answers when you edit catalog files in conf/.

How it works¶

kedro-skills writes managed files into your project (committed to git). When you edit a file matching the skill's patterns (for example, conf/**/*.yml), your IDE loads the relevant guidance into the AI assistant's context. The assistant then responds using current Kedro conventions instead of outdated training data. No MCP server is needed. Skills work offline, in any IDE.

Supported editors¶

IDE	How it activates
Cursor	`.cursor/rules/*.mdc`, fires on matching `globs:`
GitHub Copilot	`.github/instructions/*.instructions.md`, fires on matching `applyTo:`
Claude Code	`.claude/skills/<id>/SKILL.md`, always discoverable
Codex CLI	`AGENTS.md` block, always active

Available skills¶

Skill	Description	Activates on
`catalog-config`	Correct dataset naming, docs lookup, factory patterns, credentials, layers	`conf/*/.yml`, `conf/*/.yaml`

More skills are planned. Run kedro skills list to see what's available in your installed version.

Usage¶

After installing a skill, open any matching file and ask your AI assistant a question. The skill guidance loads automatically, no special commands needed.

Example: adding a catalog entry

Open conf/base/catalog.yml and ask your assistant:

Add a parquet dataset for my preprocessed_companies table stored in S3.

With catalog-config installed, the assistant will follow current Kedro conventions for dataset type, filepath pattern, layer assignment, and credentials handling instead of guessing from outdated training data.

Example: reviewing an existing catalog

Review my catalog.yml and flag anything that doesn't follow Kedro best practices.

The assistant will check naming, layers, factory patterns, and credential usage against the guidance provided by the skill.

For the full CLI reference and authoring guide, see the kedro-skills documentation.

Managing skills¶

kedro skills update          # re-render after upgrading the package
kedro skills uninstall <id>  # clean removal of all managed files

If you hand-edit a managed file, update will warn you and suggest --force.
All managed files should be committed to git so the whole team benefits.
Run kedro skills update after pip install --upgrade kedro-skills to pick up new skill content.

Skills vs MCP: when to use what¶

Tell the agent something → kedro-skills. Let the agent do something → kedro-mcp.

	Skills (`kedro-skills`)	MCP (`kedro-mcp`)
What it provides	Ambient knowledge (conventions, patterns, best practices)	Active tools (project migration, notebook conversion)
Runtime	None (static files on disk)	Long-running MCP server
Activation	Automatic on file globs	Explicit (user invokes `/mcp.Kedro.…`)
IDE reach	All assistants (Cursor, Copilot, Claude Code, Codex CLI)	MCP-capable clients
Failure mode	Slightly worse suggestion	Server crash / tool error
When on the line	Default to skills: cheaper, broader reach	Use when the agent needs to execute something

Install both. They're complementary: skills provide passive knowledge (what a catalog entry should look like), while MCP provides active capabilities (migrating your project to Kedro 1.0). You can use one without the other, but together they give your assistant the best Kedro experience.

See Vibe coding with Kedro-MCP for MCP setup and usage.

Contribute a skill¶

We welcome community-contributed skills. If you have an idea for a new skill or want to improve an existing one, check the contributing guide in the kedro-skills repository to get started. You can also reach out to us on the Kedro Slack to discuss your ideas before opening a PR.