Skip to content

v3.0.0 — One entry point, competition operations, a dry run for every write

Choose a tag to compare

View all tags
@shepsci shepsci released this 03 Oct 00:26
v3.0.0
6cd56f8
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

The first run works with nothing installed, the output is short enough to read, and the skill can run a competition as well as research one. Commands were renamed, so read "Changed" in CHANGELOG.md before you update.

  • One entry point: python3 scripts/kaggle_skill.py <command>, with short text by default and --json on the read commands.
  • Public reads need Python only: a competition's facts and pages (brief, pages), solution writeups, and discussions work with no package and no credential.
  • Competition operations: status, leaderboard (public, or --private after the deadline), validate, submit, watch, ledger, episodes, download.
  • Every write is a dry run until --yes, badge phases included, and the agent asks before it adds --yes. KAGGLE_SKILL_READ_ONLY=1 refuses every write. The cli runner lets through only commands it knows to read and never prints the token.
  • The shell scripts are Python, requests is no longer needed, and CI covers Python 3.11 to 3.14.
  • The README's demos are recorded Claude Code sessions, made with no Kaggle credential.

Verification

  • 1390 offline tests pass on Python 3.11, 3.12, 3.13 and 3.14.
  • 168 read-only live tests pass against Kaggle CLI 2.2.4 and the MCP server.
  • The eval suite (11 cases) passes with the plugin in Claude Code 2.1.286.
Assets 2
Loading