kaggle-skill

agent
Security Audit
Pass
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 20 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose
This tool is an integration skill for AI coding agents that provides full access to the Kaggle platform. It automates account setup, dataset downloads, notebook execution, competition submissions, and badge collection.

Security Assessment
The overall risk is Medium. The tool inherently accesses sensitive data because it manages your Kaggle API tokens and account credentials, which must be stored locally for it to function. It regularly makes external network requests to the Kaggle API. While the automated code scan passed with no dangerous patterns, hardcoded secrets, or dangerous permissions found, the tool requires Playwright for browser scraping. Giving an AI agent the ability to automate browser actions and execute notebooks inherently expands its system access.

Quality Assessment
The project is actively maintained, with its most recent code push happening today. It uses the highly permissive and standard MIT license. The automated code scan successfully reviewed 12 files and found no malicious code. However, with only 20 GitHub stars, the community trust and peer review base is relatively small, meaning widespread public auditing is currently limited. The documentation is straightforward and clearly outlines installation, prerequisites, and modules.

Verdict
Safe to use, but users should remain cautious about securely storing their API tokens and granting an AI agent access to their Kaggle account.
SUMMARY

Complete Kaggle integration skill for AI coding agents — account setup, competition reports, dataset/model downloads, notebook execution, submissions, and badge collection. Works with Claude Code, Gemini CLI, Cursor, Codex, and 35+ agents via skills.sh.

README.md

kaggle-skill

skills.sh
ClawHub
License: MIT
GitHub

An agent skill for everything Kaggle: account setup, competition landscape reports, dataset/model downloads, notebook execution, competition submissions, hackathon writeup retrieval and grading, badge collection, and general Kaggle questions.

Works with any AI coding agent that supports the SKILL format — including Claude Code, OpenClaw, Gemini CLI, Cursor, Codex, and 35+ more agents via skills.sh.

Available On

Platform Link Install Command
skills.sh skills.sh/shepsci/kaggle-skill npx skills add shepsci/kaggle-skill
ClawHub clawhub.ai/skills/kaggle clawhub install kaggle
Claude Code Marketplace shepsci/claude-marketplace /plugin marketplace add shepsci/claude-marketplace then /plugin install kaggle-skill@shepsci

Modules

  • Registration — Account creation, API token generation, credential storage
  • Competition Reports — Landscape reports with API + Playwright scraping
  • Kaggle Interaction (kllm) — kagglehub, kaggle-cli, MCP Server (66 tools), UI workflows
  • Hackathon — Writeup retrieval, overview/rubric extraction, role-aware grading bundles
  • Badge Collector — Systematic badge earning across 5 phases (~38 automatable)

Installation

Via skills.sh (all agents)

Installs to Claude Code, OpenClaw, Codex, Cursor, Gemini CLI, and 35+ other agents:

npx skills add shepsci/kaggle-skill

Via ClawHub (OpenClaw)

clawhub install kaggle

Via Claude Code Plugin Marketplace

Add the catalog once, then install:

/plugin marketplace add shepsci/claude-marketplace
/plugin install kaggle-skill@shepsci

Or load directly from a local clone:

claude --plugin-dir /path/to/kaggle-skill

Manual

git clone https://github.com/shepsci/kaggle-skill.git
pip install kagglehub kaggle python-dotenv requests

Then copy skills/kaggle/ into your agent's skills directory.

Prerequisites

  • Python 3.11+
  • pip install kagglehub kaggle python-dotenv requests
  • Kaggle API token (the skill walks you through setup)
  • Optional: Playwright for browser badges and competition report scraping

Credential Setup

  1. Go to kaggle.com/settings
  2. Under API Tokens (Recommended), click Generate New Token
  3. Save the token:
mkdir -p ~/.kaggle
echo 'YOUR_TOKEN' > ~/.kaggle/access_token
chmod 600 ~/.kaggle/access_token

Or set the environment variable:

export KAGGLE_API_TOKEN=YOUR_TOKEN

Legacy credentials (~/.kaggle/kaggle.json) are also supported. Run the credential checker for details:

python3 shared/check_all_credentials.py

Usage

Once installed, your agent automatically detects the skill when you mention anything Kaggle-related:

  • "Set up my Kaggle credentials"
  • "Generate a Kaggle competition report"
  • "Download the Titanic dataset"
  • "Earn Kaggle badges"
  • "Enter a Kaggle competition"
  • "What competitions are running right now?"

Bundled MCP Server (Claude Code)

When installed as a Claude Code plugin, this skill includes a .mcp.json that configures the official Kaggle MCP server, giving direct access to 66 Kaggle tools (verified against the live server in the shepsci/kmcp-tools 2026-04-22 audit):

  • Searching and listing competitions, datasets, models, notebooks
  • Downloading competition data and datasets
  • Submitting predictions to competitions
  • Pushing and executing notebooks on Kaggle Kernels
  • Publishing datasets and models
  • Hackathon writeup retrieval — overview pages, submission rosters, full writeup bodies
  • Benchmark task creationcreate_benchmark_task_from_prompt
  • Episode/simulation data — agent logs, replays, per-submission episode listings

See skills/kaggle/modules/kllm/references/mcp-reference.md for the full inventory with status flags (PASS / KNOWN_FAIL / role-gated).

The MCP server requires KAGGLE_API_TOKEN to be set.

Security

  • No automatic persistence: No cron jobs or launchd plists are auto-installed
  • No dynamic code execution: All imports are explicit and static (no __import__(), eval(), exec())
  • Untrusted content handling: Scraped content is wrapped in <untrusted-content> boundary markers
  • Credentials stored in ~/.kaggle/access_token (chmod 600) — never logged or echoed

Project Structure

kaggle-skill/
├── .claude-plugin/plugin.json    # Claude Code plugin manifest
├── .mcp.json                     # Bundled Kaggle MCP server (Claude Code)
├── PRIVACY.md                    # Privacy policy
├── skills/kaggle/
│   ├── SKILL.md                  # Main skill definition (all agents)
│   ├── shared/                   # Unified credential checker
│   └── modules/
│       ├── registration/         # Account & credential setup
│       ├── comp-report/          # Competition landscape reports
│       ├── kllm/                 # Core Kaggle interaction
│       └── badge-collector/      # Badge earning automation
└── README.md

Compatibility

Platform Status
Claude Code (CLI, VS Code, JetBrains, Desktop) Tested
OpenClaw Tested
Codex Compatible
Gemini CLI Compatible
Cursor Compatible
GitHub Copilot Compatible
Cline Compatible
Amp Compatible
35+ agents via skills.sh Compatible

Network requirements: outbound HTTPS to api.kaggle.com, www.kaggle.com, and storage.googleapis.com.

License

MIT — see LICENSE

Privacy

See PRIVACY.md — this skill collects no data. All credentials and processing remain local.

Reviews (0)

No results found