shepsci/kaggle-skill / skills/kaggle
kaggle
FlaggedUnified Kaggle skill. Use when the user explicitly mentions Kaggle, kaggle.com, a Kaggle URL, Kaggle competitions, Kaggle datasets/models/notebooks, Kaggle forums/discussions/writeups, Kaggle benchmarks, hackathons hosted on Kaggle, Kaggle badges, or Kaggle account setup. Do not use for generic ML, GPU/TPU, notebook, dataset, benchmark, or data-science tasks unless the user clearly ties them to Kaggle.
modules/competitions/references/competition-operations.md
View on GitHub# Competition Operations
What to do before and after a submission, so that a slot or a GPU hour is not
spent on something that could have been caught first. Every step that changes
the account needs the user's go-ahead.
Commands are run as `python3 scripts/kaggle_skill.py <command>`; only the
command is written below.
## Before any submission
1. **Know the rules that apply.** `brief <slug>` gives the metric, the
deadline, the team size and the daily limit. Read the pages the question
needs:
```bash
python3 scripts/kaggle_skill.py pages <slug> --page rules
python3 scripts/kaggle_skill.py pages <slug> --page evaluation
```
2. **Check where you stand.**
```bash
python3 scripts/kaggle_skill.py status <slug>
```
It shows how many submissions are left today, what is still being scored,
your best and latest scores, and the GPU hours left this week. The rules
page and the count can disagree. Plan on the smaller number unless the
host has confirmed the larger one.
3. **Look at what was already submitted**, so the same file is not sent
twice. `status` lists the latest submissions, `ledger` lists what went
through this skill, and the dry run in step 5 warns when the same file was
sent before.
4. **Validate the file locally.**
```bash
python3 scripts/kaggle_skill.py validate <slug> ./submission.csv
```
It compares the file with the sample submission: same columns, same ids,
same row count, no empty values, finite numbers. An empty value in a text
column is a warning, not a failure: check on the evaluation page whether
an empty prediction is allowed. It cannot check a format
rule that only the evaluation page states, or the range the metric
expects: read that page. A submission that errors on Kaggle can still use
one of the day's slots.
5. **Tell the user** which file, which competition, what message, and how
many slots remain. The dry run prints exactly that, and runs step 4's
checks again (`--sample PATH` names the sample):
```bash
python3 scripts/kaggle_skill.py submit <slug> ./submission.csv -m "what changed" --expect 0.81
```
Wait for a yes.
## A file submission
```bash
python3 scripts/kaggle_skill.py submit <slug> ./submission.csv -m "what changed" --expect 0.81 --yes
python3 scripts/kaggle_skill.py watch <slug>
```
`submit --yes` sends the file and adds a line to `./.kaggle-skill/ledger.jsonl`
with the file's SHA-256, the message and the expected score. A rejected
submission gives a non-zero exit status and is not recorded.
`watch` checks every 30 seconds until Kaggle has scored the submission, for
up to 30 minutes (`--timeout`). It prints the public score, records it, and
reports the difference from `--expect`. Exit status 124 means it is still
being scored: run `watch` again. Exit status 4 means the last check could not
read the submission, so its state is not known.
`--expect` is the score your own validation predicts. A large difference
between it and the public score is worth telling the user about: it points at
a leak, a mismatch between the validation and the test data, or a bug.
## A code competition
The submission is a notebook version, and Kaggle reruns it on hidden data.
`brief` says "submit with: a notebook" for these.
1. **Check the quota.** `status <slug>` shows this week's GPU and TPU hours
for the whole account. Only two GPU runs can be active at a time.
2. **Check the notebook's settings** in `kernel-metadata.json`: internet off
if the rules require it, the competition listed under
`competition_sources`, datasets and models listed as sources, and the
accelerator the rules allow. The dry run shows them:
```bash
python3 scripts/kaggle_skill.py notebook-run ./notebook-dir
```
3. **Push, which starts a run.**
```bash
python3 scripts/kaggle_skill.py notebook-run ./notebook-dir --timeout 7200 --yes
```
It stops if the push is rejected, waits for the run, downloads the output,
and prints the end of the log when the run fails.
4. **Read the output** before submitting. Check that the submission file
exists and passes `validate`.
5. **Submit that version.** The version number is on the notebook's page and
in the push output.
```bash
python3 scripts/kaggle_skill.py submit <slug> --notebook <owner>/<notebook> --version <N> -m "what changed" --yes
```
This uses a slot. Ask first.
6. **Watch the result.** A code submission can fail during the rerun on
hidden data even when the public run passed: time limit, memory, or a
missing file. `watch <slug>` reports the state and exits 1 on an error.
## A simulation competition
- Only your latest two submissions stay in play. A new one replaces the older
of the two.
- Ratings settle over many episodes. Do not judge a submission from its first
few games.
- `episodes <submission-id>` lists the games with rewards and opponents.
`episodes --replay <episode-id>` and `episodes --logs <episode-id>` save a
game and your agent's log. See
[episode-endpoints.md](../hackathons/references/episode-endpoints.md).
## The leaderboard
```bash
python3 scripts/kaggle_skill.py leaderboard <slug>
```
Your row, the gap to the leader, and the score at each medal line. The medal
lines follow Kaggle's table (kaggle.com/progression/competitions) at today's
team count; they move as teams join, and the private leaderboard decides the
result. Run it again later to see what moved.
## Final submissions
Which submissions count at the end is chosen on the competition's website.
No command or MCP tool shows or changes the selection. Remind the user to
check it before the deadline.
## What cannot be done from here
Accepting the rules, joining or merging teams, identity checks, and choosing
final submissions are done on kaggle.com. Say so, and give the address.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| deprecation-and-migrationaddyosmani/agent-skillsManages deprecation and migration. Use when removing old systems, APIs, or features. Use when migrating users from one implementation to another. Use when migrating a database schema in production, such as renaming or dropping a column without downtime (expand/contract). Use when deciding whether to maintain or sunset existing code. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| performance-optimizationaddyosmani/agent-skillsOptimizes application performance across frontend, backend, queries, and databases. Use when performance requirements exist, when you suspect performance regressions, when Core Web Vitals or load times need improvement, when N+1 query patterns need fixing, or when profiling reveals bottlenecks. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| agent-cost-reportthedotmack/claude-memBelievable agent cost report for any period, default the last 7 full days PT, not counting today. Measured tokens from Claude Code transcripts priced at OpenRouter list prices (ESTIMATED), measured provider spend when a sanctioned source exists, note-taker cost separate, Timing-style HTML/PDF plus report.json, line-items.csv, evidence.json. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |
| agent-cost-reportthedotmack/claude-memBelievable agent cost report for any period, default the last 7 full days PT, not counting today. Measured tokens from Claude Code transcripts priced at OpenRouter list prices (ESTIMATED), measured provider spend when a sanctioned source exists, note-taker cost separate, Timing-style HTML/PDF plus report.json, line-items.csv, evidence.json. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |