Imported from rweekly/rweekly.org (.claude/skills/curate/SKILL.md). Install upstream with npx skills add rweekly/rweekly.org --skill curate. Copyright stays with the author.

R Weekly Release Curation

Live Context

• Current draft: !cat draft.md

Your Task

You are helping the R Weekly curation team prepare the weekly issue. Work through the candidate content from curatinator_latest.md and populate draft.md.

Classification Rules

For each RSS POST, decide:

1. Is it R-related? Fetch and read the actual page content for every item before deciding — do not infer relatedness from the title, author, or domain alone. A post on a known R blogger's site or a stats-adjacent title can still be off-topic (e.g. a general statistics/data-ethics essay with no R content). If not R-related, skip it entirely.
2. Is it already in the draft or flagged as a duplicate? If so, skip it.
3. Which section does it belong to?

For CRANberries NEW entries: add under ### New Packages -> **CRAN**. For CRANberries UPDATED entries: add under ### Updated Packages. RSS posts announcing a package release (e.g. "RcppArmadillo 15.6.0-1 on CRAN") also go under ### Updated Packages. Bioconductor and GitHub packages go under ### New Packages -> **Bioconductor** / **GitHub or Bitbucket or GitLab**.

Do NOT add anything to ### Highlight — editors vote for that section separately. Do NOT touch ### Quotes of the Week — it uses raw social-embed HTML, added by hand.

Link Format

+ [Title](URL)

For packages, keep the CRANberries description, and the diffify link for updated packages. Some descriptions wrap onto a second line in curatinator_latest.md: join them back into one line.

+ [{pkgname} version](https://cran.r-project.org/package=pkgname): Description - [diffify](https://diffify.com/R/pkgname)

Images (optional, editor discretion):

![Alt text](https://raw.githubusercontent.com/...)

Fetching Page Content

To inspect a URL's content before classifying it, use any of:

• WebFetch tool — built-in; use it directly.
• curl https://r.jina.ai/https://www.example.com — fallback if WebFetch fails.
• curl https://defuddle.md/example.com — second fallback.
• npx defuddle parse https://www.example.com --markdown — third fallback if defuddle.md itself is down or rate-limited; runs the same extraction locally with no dependency on the hosted proxy.

Steps

1. Check open PRs with gh pr list --repo rweekly/rweekly.org. If any are open, tell the user and wait for their go-ahead before continuing.
2. Collect content — only if curatinator_latest.md is missing or stale (the Saturday GitHub Action normally refreshes it): Rscript -e 'source("scripts/curatinator.R")'. This scrapes the network and overwrites the file, so skip it when the file is already current.
   • Stale = its CRAN packages already appear in the last _posts/ issue (a recent mtime proves nothing).
   • Outside the Nix env, first check Rscript -e 'library(tidyRSS); library(RCurl); library(pkgsearch)' and that OPENAI_API_KEY is set: otherwise the RSS section comes out empty without any error.
3. Read curatinator_latest.md (the current draft.md is already loaded in Live Context above). Optionally parse it with Rscript -e 'source("scripts/parse_curinator.R")' for a tidy view.
4. Process RSS POSTS: for each link, classify and add to the correct section in draft.md. Skip non-R content.
5. Process CRANberries NEW: add new CRAN packages under ### New Packages -> **CRAN**.
6. Process CRANberries UPDATED: add updated packages under ### Updated Packages.
   • For both lists, keep only the most popular packages (about 40 new, 25 updated), ranked by last-week downloads: curl -s 'https://cranlogs.r-pkg.org/downloads/total/last-week/pkg1,pkg2,...'.
7. Edit draft.md using the Edit tool to insert the new links into the appropriate sections.
8. Check duplicates (after editing, so new links are checked too): Rscript -e 'source("scripts/find_duplicates.R"); get_dups()'; remove any flagged URLs.
9. Report a summary:
   • How many RSS posts were added (and to which sections)
   • How many CRAN new/updated packages were added
   • How many items were skipped (not R-related or duplicates)
   • Any items that were ambiguous and need editor review

Keep the existing content and structure of draft.md intact. Add new links at the bottom of each section, above any closing HTML comments.