Skip to content

v2.4.0 — Verbatim demos, retrieval QC, docs refresh

Latest
Latest

Choose a tag to compare

View all tags
@shepsci shepsci released this 04 Jul 17:48
v2.4.0
f1c0a81
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

First release since v2.1.0 — rolls up the unreleased 2.2.0/2.3.0 milestones plus the retrieval-QC and screencast-refresh work. See CHANGELOG.md for the full history.

Highlights

Leaderboard writeup retrieval (PR #25)

  • leaderboard_writeups.py handles the current Kaggle competition JSON shape, joins teams[].solutionWriteUpUrl to private/public leaderboard ranks and scores, supports --preview (title/meta-description extraction with boilerplate filtering) and --fallback-search for competitions whose leaderboards expose no writeup URLs.
  • New live integration tests prove Kaggle assets and writeups are retrieved and wrapped as untrusted content — never refused: kagglehub dataset download, Vesuvius Challenge top-3 writeup previews, and the Kaggle CLI 2.2.x forums/topics suite.
  • README demo switched from ARC-AGI to the Vesuvius Challenge (new competitions expose leaderboard writeups); the demo cast is the verbatim output of the demoed command.
  • Badges standalone self-test fixed (REPO_ROOT export + submission_titanic.csv template, with the .gitignore exception it needed to actually ship).

Screencast & docs refresh (PR #26)

  • Every demo GIF re-rendered from its .cast source — the previously committed renders were broken (1-frame blanks or clipped frames showing no output).
  • Three casts rebuilt from verbatim live command output; fabricated flags, invented CLI strings, and placeholder writeup IDs are gone. Credential lines are redacted by policy, everything else is a real capture.
  • Dead VHS recording pipeline removed; asciinema + agg is the single documented pipeline.
  • New guard tests: manifest-wide version consistency and GIF animation validity.
  • Added CHANGELOG.md.

Earlier unreleased milestones (2.2.0 / 2.3.0)

  • Module reorganization by workflow, truthfulness audit of docs claims, direct Claude marketplace install path, Antigravity CLI documentation, platform-compatibility updates, and the Kaggle affiliation disclaimer.

Verification

  • Full test suite: 376 passed, 47 skipped. Live suites (retrieval, MCP, CLI 2.2.3) green.
  • Claude and Codex plugin validators pass.
Assets 2
Loading