Przejdź do treści
Biblioteka skilli
Nauka i badania · Budowanie i kod

release-research-artifacts

/release-research-artifacts

Pakuje kod i dane do pobrania dla czytelnika lub recenzenta: wydanie kodu artykułu, zanonimizowane repozytorium do podwójnie ślepej recenzji, archiwum z DOI, metadane cytowania, data card i model card.

Licencja
Apache-2.0
Ostatnia zmiana repozytorium
23.09.2026

Ocena i bezpieczeństwo

85/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść20 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
15
Skrypty
2
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 23.09.2026, bez odwołań do wycofanych modeli i API.

Licencja

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

Co sprawdzić przed instalacją
  • 2 skrypty w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.

Skrypty: examples/release-package/src/reproduce.py, scripts/check_release_package.py

Co robi i kiedy

Pakuje kod i dane do pobrania dla czytelnika lub recenzenta: wydanie kodu artykułu, zanonimizowane repozytorium do podwójnie ślepej recenzji, archiwum z DOI, metadane cytowania, data card i model card.

Oryginalny opis (EN)

Use when packaging the code and data a reader or reviewer will download, including releasing the code for a paper, building an anonymized repository for double-blind submission, preparing an archive with a DOI for a journal, writing citation metadata, producing a data card or model card, or preparing a submission for artifact evaluation. Use once the results are frozen, so that what ships is what produced them.

Jak działa

  1. Select one mode
  2. Resolve the installed skill and components
  3. Resolve the file set by allowlist
  4. Stage, then scan every byte
  5. Test inside the built package
  6. Record what shipped
  7. Boundaries
Sekcje z nagłówków SKILL.md, w oryginale · 81 linii, 653 słów
Podgląd SKILL.md
---
name: release-research-artifacts
description: Use when packaging the code and data a reader or reviewer will download, including releasing the code for a paper, building an anonymized repository for double-blind submission, preparing an archive with a DOI for a journal, writing citation metadata, producing a data card or model card, or preparing a submission for artifact evaluation. Use once the results are frozen, so that what ships is what produced them.
---

# Release Research Artifacts

Build the package someone else downloads, then verify the package rather than the
workspace it came from. Almost every release defect is invisible from inside the working
directory, because the working directory contains the missing file.

## Select one mode

| Mode | Responsibility | Read first |
|---|---|---|
| `anonymized-submission` | Build a package that carries no author or institution identity, for double-blind review | `references/anonymized-submission.md` |
| `named-archive` | Build the citable, deposited version with persistent identifiers | `references/named-archive.md` |
| `data-card` | Describe the data and any model well enough to be used without asking you | `references/cards.md` |
| `manifest-verification` | Prove the built package is complete, runnable and unchanged | `references/verification.md` |

## Resolve the installed skill and components

Resolve `SKILL_DIR` as the directory containing this `SKILL.md`. Every runtime resource
is below that directory; never resolve through a package parent.

## Resolve the file set by allowlist

List what goes in. Do not build the package by excluding what should stay out.

An exclusion list is a claim about everything you did not think of, and the things nobody
thinks of are exactly the ones that carry identity: editor backup files, notebook
checkpoints, cached credentials, environment dumps naming a home directory, log files with
absolute paths, and files with no extension that a text scan skips by default.

Write the allowlist as paths and patterns, resolve it into an explicit file list, and
review that list before staging. The reviewed list is the artifact; the command that
produced it is not.

## Stage, then scan every byte

Copy the allowlisted files into an empty staging directory. Scan the staged tree, not the
source tree, for author names, institutions, email addresses, absolute home paths,
internal hostnames, project codenames, credentials and tokens.

Scan file contents and file names. Scan inside container formats that hold text: office
documents, notebooks, archives. A scanner that reads only files with known text
extensions will pass a tree whose cache files are the leak.

Seed a known string that should be caught and confirm the scan catches it. A scan
reporting zero findings on a clean tree and a scan that is silently not running produce
the same output.

Plant the seed in the real formats the tree actually holds, not in a convenient one. A
seed placed in a plain text file proves only that the scanner reads plain text files; the
formats that leak are the notebook output cell, the archive member, the office document
body and the file name itself. A control built in a format the tool already handles
validates the control, not the tool.

## Test inside the built package

Unpack the built archive into an empty directory and run the project's own test or
reproduction command there, with no access to the working repository.

This is the step that catches the file that was never committed, the import that resolves
only because of the working directory, the data path that points outside the package, and
the dependency that is installed on your machine and nowhere else.

A test run in the workspace proves nothing about the archive.

## Record what shipped

Write a listing of the released files with their sizes and a cryptographic hash for each,
computed over the built package. Record the commit the package was built from, the
interpreter and dependency versions used for the verification run, and the date.

Someone who downloads the package later needs to answer two questions from that record
alone: is this the same package, and what produced it.

## Boundaries

This skill builds and verifies the package. It does not write the availability statements
that point at it, does not grant rights you do not hold over third-party data, and cannot
make a restricted dataset shareable. Where the data cannot be released, the package
carries the access route and the exact identifiers instead, and says which.
Pliki w paczce (15)
  • SKILL.md 4 KB
  • agents/openai.yaml 1 KB
  • examples/release-package/CITATION.cff 1 KB
  • examples/release-package/LICENSE 1 KB
  • examples/release-package/README.md 1 KB
  • examples/release-package/RELEASED_FILES.json 2 KB
  • examples/release-package/requirements.txt 1 KB
  • examples/release-package/data/README.md 1 KB
  • examples/release-package/data/scores.csv 1 KB
  • examples/release-package/src/reproduce.py 1 KB
  • references/anonymized-submission.md 2 KB
  • references/cards.md 2 KB
  • references/named-archive.md 2 KB
  • references/verification.md 2 KB
  • scripts/check_release_package.py 12 KB

Wersje

W bazie nie ma innych skilli o nazwie release-research-artifacts.

Migawka bazy z 13.11.2026 · wersja paczki c87158fa3678 (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100prepare-artifacts
    Przygotowuje artefakt reprodukowalności (kod i dane) do złożenia i otrzymania odznak, w tym ACM, USENIX, SIGMOD i NeurIPS, z anonimizacją repozytorium i DOI w Zenodo.
  • 100anonymize-repo
    Eksportuje zanonimizowaną migawkę repozytorium badawczego bez historii do recenzji podwójnie zaślepionej i sprawdza, czy się buduje.
  • 100submit-paper
    Prowadzi przez zgłoszenie artykułu naukowego do recenzji podwójnie ślepej: artefakt, profil HotCRP, anonimowość i uprawnienia autora.
  • 100paper-research
    Wyszukuje artykuły naukowe i literaturę z różnych źródeł, pobiera pliki PDF i eksportuje cytowania, na przykład do BibTeX, według DOI.
  • 100release-doi
    Prowadzi procedurę wydania repozytorium badawczego z rejestracją DOI (Zenodo, Software Heritage) oraz poprawia metadane już opublikowanego rekordu.
  • 100prepare-artifacts
    Przygotowuje paczkę artefaktów badawczych do oceny powtarzalności: plan pakowania, README, matrycę twierdzeń i poleceń oraz opis do dodatku.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.