What problem does it solve?
Kaggle competitions often fail due to slow iteration, opaque submission errors (like 400s or invalid kernel mode), and fragile replication of top notebooks. This Skill helps you systematically diagnose failures, replicate winning approaches with dependencies, and automate monitoring so you spend less time guessing and more time improving.
Core Features & Use Cases
- Score stabilization guidance: Prevents premature decisions by teaching you when leaderboard scores become reliable.
- Submission troubleshooting: Guides fixes for common Kaggle submission problems such as ZIP formatting issues and incorrect ID handling.
- Kernel workflow for reliable replication: Shows how to pull/push kernels with dependency metadata and use the correct Kaggle execution mode for valid results.
- Spec-driven development + delegation: Helps you document a clear SPEC.md plan and delegate implementation to coding agents while enforcing constraints (time, GPU, internet rules).
Quick Start
Ask your agent: "Use the agentic-data-science-competition skill to replicate a top Kaggle kernel and troubleshoot why my submission is returning a 400 error."