Skip to content

v3.0.1 - #32

Merged
shepsci merged 5 commits into
mainfrom
improve-demos-3.1
Oct 3, 2026
Merged

shepsci merged 5 commits into
mainfrom
improve-demos-3.1

Conversation

@shepsci

@shepsci shepsci commented Oct 3, 2026 •

Copy link
Copy Markdown
Owner

Version 3.0.1: the steps the recorded demo sessions had to work around are now one command each, the README opens with a demo of what is running on Kaggle now, and the demo GIFs are easier to read.

Found by watching the recorded sessions

  • "Which competitions are running, and what are their metrics?" took a listing plus one lookup per competition. competitions now shows each competition's metric and the last day to join (or "submissions closed").
  • To answer "what's the metric and the deadline?", the agent had to read the evaluation and timeline pages after brief. brief now quotes the start of the evaluation page and lists the host's timeline, and takes several competitions at once.
  • To read the top teams' writeups, the agent piped each one through head -c. writeup now takes several ids or URLs, and cuts each body at 8,000 characters with a note.
  • solutions --preview often showed only a writeup's title. Previews now come from the writeup body.
  • To submit, the agent ran validate, then the submit dry run, which then told it to validate. The dry run now runs those checks itself.
  • KAGGLE_SKILL_HIDE_ACCOUNT=1 leaves your entries and ranks out of listings and briefs, for screen sharing and recordings.

The demos

  • Four real Claude Code sessions, recorded on a subscription, read-only: what is running now (the README's first screen), one competition's deadline and metric, what the top teams did, and a submission that stops at the dry run and asks.
  • Every answer was checked against what the commands printed; the recorder now lists any number in an answer that no command printed. The "running now" answer was also compared row by row with a separate lookup of each competition and with the Kaggle CLI's listing.
  • Answers are shown a screen at a time with time to read; Markdown is drawn as plain text and table rows as list items.

Checked: 1,420 offline tests on Python 3.11–3.14, ruff, the skill and plugin validators, and 169 read-only live tests.

🤖 Generated with Claude Code

shepsci and others added 5 commits October 2, 2026 20:10
…ound

- brief quotes the start of the evaluation page next to the metric.
- writeup takes several ids or URLs, one block each, and cuts a body after
  8,000 characters with a note (--max-chars 0 for all).
- solutions --preview reads how each writeup's body starts from the MCP
  server, with no credential; the page itself often showed only the title.
- The submit dry run runs validate's checks on the file (--sample PATH).
- Demo GIFs: answers drawn as prose and shown a screen at a time, long
  commands typed faster, block tags without their attributes, a longer hold
  on the last screen.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The three sessions were recorded again (on a Claude subscription, read-only,
no Kaggle credential). Each now takes the short path the 3.1.0 commands
allow: one brief, one submit whose dry run checks the file, and one writeup
call for the three top teams. The GIFs show an answer a screen at a time with
time to read it, without Markdown marks; the casts were rebuilt for the
version and the new rendering.

SKILL.md says that long text is already cut with a note, so a command needs
no `| head`. Live tests cover the evaluation line, several writeups in one
call and the file check in the submit dry run.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nst Kaggle

- competitions shows each competition's metric and the last day to join (or
  "submissions closed"), in two lines per competition.
- brief lists the dated lines of the Timeline page (entry, team merger and
  final deadlines in the host's words), takes several competitions at once,
  and says "join by" where it said "entry closes".
- KAGGLE_SKILL_HIDE_ACCOUNT=1 leaves your entries and ranks out of the
  listing and the brief, for screen sharing and recordings.
- A new recorded session, now the README's first screen: which competitions
  with prize money are running, with deadline, prize and metric. It needs a
  credential to list competitions, so it was recorded with one and with the
  account hidden. The single-competition session was recorded again.
- record_session.py lists every number in an answer that no command printed.
  All four sessions pass that check; each row of the new answer was also
  compared with a separate lookup of the competition and with the Kaggle
  CLI's own listing.
- The GIFs draw a table row as a list item.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… traced

- The session was recorded again with a tighter question, so the whole
  answer fits in the GIF: two lines for each of the top three teams.
- solutions says which leaderboard its ranks and scores come from. The
  recorded answer called them private-leaderboard scores, which was true but
  not something the command had said.
- Answer pages in the GIFs end where a paragraph ends, so a heading stays
  with its lines, and each screen is held a little longer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The smallest version step that still reaches people who installed 3.0.0:
Claude Code refreshes an installed plugin only when the version changes, and
ClawHub accepts new content only under a new version.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@shepsci shepsci changed the title v3.1.0 v3.0.1 Oct 3, 2026
@shepsci
shepsci merged commit a47bff9 into main Oct 3, 2026
7 checks passed
@shepsci
shepsci deleted the improve-demos-3.1 branch October 3, 2026 15:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant