kaggle-automation
by ComposioHQkaggle-automation helps agents automate Kaggle tasks through Composio’s Rube MCP by discovering current tools, checking Kaggle connection status, and using live schemas before execution.
Score: 67/100. This is acceptable for listing, but as a limited utility skill rather than a rich Kaggle automation playbook. Directory users get enough evidence to understand that it helps agents route Kaggle requests through Composio/Rube MCP with connection checks and dynamic tool discovery, but they should expect to rely on live Rube tool schemas rather than detailed built-in Kaggle workflows.
- Valid frontmatter declares the skill name, description, and Rube MCP requirement, making the intended trigger and dependency clear.
- Prerequisites and setup explicitly tell agents to verify RUBE_SEARCH_TOOLS, manage a Kaggle connection with RUBE_MANAGE_CONNECTIONS, and confirm ACTIVE status before running workflows.
- The skill repeatedly instructs agents to call RUBE_SEARCH_TOOLS first for current tool schemas, reducing schema guesswork when Kaggle tools change.
- No support files, scripts, references, README, or install command are present beyond SKILL.md, so adoption depends on users already understanding how to install skills and configure MCP.
- The guidance is mostly a generic Rube MCP discovery/connection pattern; the evidence shows little Kaggle-specific operational detail or practical examples beyond searching tools and managing the Kaggle connection.
Overview of kaggle-automation skill
What kaggle-automation does
kaggle-automation is a Claude skill for automating Kaggle operations through Composio’s Kaggle toolkit using Rube MCP. It is designed for agents that need to discover Kaggle tools, verify authentication, and execute Kaggle-related workflows without hard-coding stale API schemas.
The key behavior is simple but important: before doing any Kaggle action, the agent should call RUBE_SEARCH_TOOLS to find the current available tools, schemas, execution plan, and pitfalls.
Best-fit users and workflows
The kaggle-automation skill is most useful if you already use Claude with MCP and want an agentic workflow for Kaggle tasks such as dataset discovery, competition-related operations, notebook or resource management, or other actions exposed by Composio’s Kaggle toolkit.
It fits users who care less about writing one-off Kaggle API scripts and more about letting an AI agent choose the correct Rube tool after checking the live schema. This matters because MCP tool schemas can change, and the skill’s main value is forcing discovery before execution.
What makes this skill different
Unlike a generic “help me use Kaggle” prompt, kaggle-automation encodes a specific operating pattern:
- Use Rube MCP as the execution layer.
- Confirm the Kaggle connection is active before running workflows.
- Search tools first instead of assuming tool names or parameters.
- Use returned tool schemas and recommended plans as the source of truth.
That makes it better for Workflow Automation where reliability depends on current tool metadata, not memory of older API examples.
How to Use kaggle-automation skill
kaggle-automation install and setup context
Install the skill from the repository path:
npx skills add ComposioHQ/awesome-claude-skills --skill kaggle-automation
Then configure Rube MCP in your client by adding:
https://rube.app/mcp
The upstream skill states that no API keys are needed for the MCP endpoint itself, but you still need an active Kaggle connection through Rube. Verify that RUBE_SEARCH_TOOLS is available, then use RUBE_MANAGE_CONNECTIONS with toolkit kaggle. If the connection is not ACTIVE, follow the returned auth link and check again before attempting Kaggle actions.
Inputs the skill needs from you
For reliable kaggle-automation usage, give the agent a concrete Kaggle goal, the object you want to act on, and any limits. Weak prompts such as “do my Kaggle task” force the model to guess the use case and may lead to unnecessary discovery loops.
Better input:
Use kaggle-automation to find the current Rube MCP Kaggle tools for searching datasets about housing prices. Check that my Kaggle connection is active first. Return the available actions, required fields, and ask before downloading or modifying anything.
Even stronger input includes:
- Target type: dataset, competition, notebook, file, or account action.
- Desired result: search, list, upload, download, submit, inspect, or summarize.
- Safety limits: read-only, no submissions, no downloads over a size limit.
- Output format: table, plan, tool call summary, or next-step checklist.
Practical workflow for agents
A good kaggle-automation guide should follow this sequence:
- Call
RUBE_SEARCH_TOOLSwith a use case such as"Kaggle dataset search"or"Kaggle competition submission". - Reuse the returned session ID for subsequent calls.
- Call
RUBE_MANAGE_CONNECTIONSfor toolkitkaggle. - If inactive, stop and request authentication.
- Use the exact returned tool slug and input schema.
- Execute the action only after confirming destructive or public actions.
This skill is especially sensitive to schema drift. Do not ask the model to invent fields. The repository explicitly instructs agents to search tools first because Rube returns the current schema and known pitfalls.
Repository files to read first
This skill is compact: the main source is composio-skills/kaggle-automation/SKILL.md. There are no visible helper scripts, references, rules, or metadata files in the provided tree preview, so most install-decision value comes from understanding the MCP prerequisites and workflow pattern inside SKILL.md.
Read the source if you need the exact example calls for:
RUBE_SEARCH_TOOLSRUBE_MANAGE_CONNECTIONS- Session handling
- Connection checks before execution
kaggle-automation skill FAQ
Is kaggle-automation only for Kaggle competitions?
No. The skill is framed around Composio’s Kaggle toolkit, not only competitions. The available actions depend on what RUBE_SEARCH_TOOLS returns for your use case. It may cover datasets, competitions, notebooks, or other Kaggle operations exposed by the current toolkit.
Why not just ask Claude to use the Kaggle API?
Generic prompts can explain Kaggle concepts, but they do not automatically discover the current Rube MCP tool schema. The kaggle-automation skill is useful when the agent must operate through Composio/Rube and should not rely on remembered API names or parameters.
Is this beginner-friendly?
It is beginner-friendly for users comfortable adding an MCP server and completing an OAuth-style connection flow. It is not ideal for someone who only wants a local Python Kaggle API tutorial. The main requirement is understanding that Rube MCP is the execution layer and Kaggle must be connected before tasks run.
When should I not use this skill?
Do not use kaggle-automation if you need a standalone Kaggle CLI replacement, offline scripting, or custom data science modeling logic. It is also a poor fit when your environment cannot use MCP tools or when you need deterministic CI/CD behavior without an interactive authentication step.
How to Improve kaggle-automation skill
Improve prompts for kaggle-automation
The biggest quality improvement is giving the agent a narrow use case for tool discovery. Instead of:
Search Kaggle.
Use:
Use kaggle-automation to discover current Kaggle tools for finding public datasets related to credit card fraud. Keep the workflow read-only, show required fields before calling any execution tool, and return the top candidate actions with risks.
This helps the skill choose the right search query, preserve safety boundaries, and avoid unnecessary Kaggle actions.
Prevent common failure modes
Common issues are usually setup-related:
RUBE_SEARCH_TOOLSis unavailable because Rube MCP is not connected.- Kaggle auth is incomplete or not
ACTIVE. - The agent skips tool discovery and guesses schemas.
- The prompt does not say whether actions may modify, upload, download, or submit.
You can reduce failures by requiring the agent to report connection status and the discovered tool schema before execution.
Iterate after the first output
After the first tool discovery result, refine the workflow using the returned tool names and fields. Ask the agent to convert the discovery output into a short execution plan:
Based on the discovered Kaggle tools, list the exact tool calls needed, required inputs, optional inputs, and any irreversible actions. Do not execute until I approve.
This turns kaggle-automation from a broad automation helper into a controlled operational workflow.
Add local project guardrails
For team use, pair the skill with your own rules about Kaggle data handling: download locations, dataset size limits, competition submission approval, credential handling, and logging. The upstream skill provides the Rube/Kaggle workflow pattern, but your environment should define what the agent is allowed to do once the connection is active.