Kaggle Solution Write-up Writer

    2

    Draft, structure, and review Kaggle solution write-ups with a section contract distilled from 5,700+ public write-ups.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Kaggle Solution Write-up Writer

    Example session with this skill installed

    Draft my 5th place solution write-up for BirdCLEF 2024. Notes: EfficientNet-B0 backbone, 5-fold CV, pseudo-labeling on the unlabeled soundscapes gave +0.012 CV. Final CV 0.81, private LB 0.82.

    • Read your context and instructions
    • Compiled the kaggle solution write-up

    Summary

    I drafted the write-up from your notes using only the numbers you gave me. Several things are missing, and I marked them rather than guess. Answer the questions at the end and I will fill them in.


    5th Place Solution

    Thanks to the hosts for another well-run BirdCLEF, and congratulations to everyone who took part. [Credit by @handle: which public notebooks or discussion posts did you build on?]

    TL;DR

    A single-architecture solution: EfficientNet-B0 trained with 5-fold cross-validation, plus pseudo-labeling on the unlabeled soundscapes. Pseudo-labeling was the change that mattered most, worth +0.012 CV. Final scores: CV 0.81, public LB not reported, private LB 0.82.

    Data & preprocessing

    [Needed: sample rate, spectrogram parameters, chunk length, and whether you used external data. A reproducer has to match these exactly.]

    Validation

    5-fold cross-validation.

    [Needed: how the folds were split, and why.] Final CV was 0.81 against a private LB of 0.82. Public LB was not reported, so I make no claim about how closely CV tracked the leaderboard during the competition.

    Model

    EfficientNet-B0. [Needed: anything you changed from the off-the-shelf version.]

    Training

    [Needed: loss, schedule, epochs, and the hyperparameters that mattered.]

    Pseudo-labeling

    Pseudo-labels generated on the unlabeled soundscapes and added to training: +0.012 CV.

    Results

    ChangeCVPublic LBPrivate LB
    Pseudo-labeling on unlabeled soundscapes+0.012not measurednot measured
    Final submission0.81not measured0.82

    There is no baseline row. You did not report a pre-pseudo-labeling score, and subtracting the delta from the final CV would present a number nobody measured.

    What didn't work

    [Needed: each dead end and why it failed. This section is required, and I will not invent experiments.]

    Code & artifacts

    [Needed: absolute links to the repo, inference notebook, dataset, and model weights.]


    Before you post

    1. What was your public LB score?
    2. Did you measure a score before pseudo-labeling? If so I will add a baseline row.
    3. What did you try that failed, and why do you think it failed?
    4. Whose public work did you build on? I need @handles and links.
    5. BirdCLEF caps inference runtime. What was your measured runtime against the cap?

    The draft is about 300 words. Once the gaps are filled it should land near the 700 to 900 words the most-upvoted write-ups run to, and a pipeline diagram would help.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Turn your experiment notes into a Kaggle solution write-up people actually read. This skill is not a style opinion: it was distilled from every public solution write-up linked from a Kaggle leaderboard (5,700+ write-ups across 540+ competitions) by measuring what the most-upvoted posts contain that typical posts do not, and turning those differences into a section contract your agent follows.

    What it does

    Ask for a write-up in your own words and the skill triggers on its own. It gathers the competition's facts and the other teams' write-ups, asks you for any score deltas you left out, and produces a draft that follows the contract. It also reviews and revises drafts you already have.

    What it enforces

    • A section contract. The sections real top write-ups use, in the order they actually use them (measured median ordering, not invented).
    • Numbers on every claim. CV, public LB, and private LB stated explicitly; a delta attached to each technique, or the technique marked unmeasured.
    • No invented scores. The most common failure when an agent drafts from incomplete notes is back-computing a baseline from final minus delta and presenting it as a measurement. The results-table row schema blocks this; unmeasured cells read not measured.
    • A real "what didn't work" section. With reasons, so a reader can skip or retry each dead end.
    • Credit and artifact links. @-handles for the public kernels and discussions you built on, absolute links to code, notebook, dataset, and model.

    It also knows Kaggle's own pre-filled write-up template (Context / Overview of the Approach / Details of the submission / Sources) and when to use it instead, such as prize-eligible winners' documentation.

    What's inside

    • SKILL.md: the contract, the rules, and a pre-posting checklist.
    • references/corpus-evidence.md: the measurements, cut by finishing position and by community votes. Every recommendation traces to a row in it.
    • references/exemplars.md: worked examples of results tables, "what didn't work" sections, and openings from real write-ups, with links to the originals.

    Who it's for

    Kaggle competitors posting an "Nth place solution", filling in the write-up form linked from a leaderboard, or submitting winners' documentation for prize eligibility.

    Good to know

    The skill works on its own from your notes. If the nvidia-kaggle skill is installed, it uses that skill's fetch scripts to pull competition facts and other teams' write-ups automatically.

    You must supply the measured CV, public LB, and private LB scores and per-technique deltas; the skill will not estimate or back-compute them. It does not generate figures or diagrams. It ships no scripts of its own: automatic fetching of competition facts and other teams' write-ups relies on the separate, free nvidia-kaggle skill, and without it you paste that context in yourself.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Permissions

    No special permissions declared or detected

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    A generic prompt tends to transcribe your notes and fill gaps with plausible-looking numbers. This skill follows a section contract measured from 5,700+ public Kaggle write-ups, attaches a delta to every technique, and writes `not measured` wherever you did not report a score instead of inventing one.

    Any agent that supports the open SKILL.md standard, including Claude Code, Codex CLI, Cursor, Copilot CLI, and Gemini CLI. The skill is plain markdown with no code to run.

    No. The skill organizes results you already measured. If a delta or score is missing it asks you for it, and if you do not have it the cell reads `not measured`.

    Three markdown files: SKILL.md (the section contract, rules, and pre-posting checklist), references/corpus-evidence.md (the measurements behind every recommendation), and references/exemplars.md (worked examples from real write-ups, with links to the originals). No scripts are included.

    It follows Kaggle's own four-heading write-up template (Context, Overview of the Approach, Details of the submission, Sources) when you are filing prize documentation. Each competition's host sets the actual requirements, so check the draft against your competition's rules before submitting.

    New in Technical Documentation