Add the "reprexes" skill (.claude/skills/reprexes)

[d-morrison]

Summary

Adds a project-level Claude Code skill at .claude/skills/reprexes/SKILL.md.

reprexes captures a debugging workflow: when a technical problem is hard to pin down, isolate it into a minimal reproducible example ("reprex") and iterate candidate fixes on that fast, clean loop instead of inside the full application — then port the fix back. Includes the R/Quarto reprex package workflow (reprex::reprex(), venue / session_info / std_out_err).

References: https://r4ds.hadley.nz/workflow-help.html#making-a-reprex, https://reprex.tidyverse.org/.

Vendoring it here makes it available to collaborators and the @claude CI bot in this repo (previously it lived only in a personal user-level config).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds a repository-level Claude Code skill definition for creating “reprexes” (minimal reproducible examples) to support faster, cleaner debugging workflows in this repo.

Changes:

• Introduces .claude/skills/reprexes/SKILL.md describing when/how to build a minimal reproducible example and iterate fixes outside full-context runs.
• Includes R/Quarto-specific guidance for using the {reprex} package and clean-session execution patterns.

[claude]

Claude finished @d-morrison's task in 1m 41s —— View job

---

Code Review — PR #239

• Read changed files and gather context
• Review SKILL.md content
• Check issues flagged by prior reviewers
• Post review feedback

---

Summary

This PR adds a well-structured Claude Code skill for building minimal reproducible examples ("reprexes"), updates .Rbuildignore to exclude .claude/ from the R package build, and adds a NEWS.md entry. The CI failures from the initial submission were fixed in 525375d. The overall quality is high — the skill is clear, practical, and R-aware. A few small issues worth noting:

---

.Rbuildignore

The regex ^\.claude$ is correct. No issues.

---

NEWS.md

Entry is concise and accurate. No issues.

---

.claude/skills/reprexes/SKILL.md

Strengths:

• Clear frontmatter (user-invocable: true, correct allowed-tools)
• Good scoping of when to use vs. skip the workflow ("Don't bother for a one-line obvious fix...")
• Cross-platform path guidance (tempfile(fileext = ".R") / tempdir() for non-Unix)
• The tidyverse_update() concern from the Copilot review is handled correctly: the text explicitly says to run it outside the reprex and not include it inside, which is the right advice.
• The .lintr concern from the Copilot review no longer applies — the current text says "Respect the repo's lint config" without naming a specific file.

Issues:

1. Namespace inconsistency (line ~94)
   The paragraph introduces reprex::reprex() (qualified), but the same sentence immediately drops to reprex(input = ...) (unqualified). This is a small inconsistency in documentation style:

   Copy the code and call `reprex::reprex()` (reads the clipboard by default),
   or point it at a file with `reprex(input = "/tmp/reprex.R")`
   

   Either qualify both or neither. Fix this →

2. reprex_invert() — worth verifying (line ~111)
   The skill lists reprex_invert() as a companion helper "exported in reprex 2.x" and describes it as "the inverse of reprex() — recover the input code from a rendered reprex." This function does not appear in the current CRAN reprex reference (2.1.x). reprex_clean() and reprex_rescue() are real, but reprex_invert() may not be. If it doesn't exist, the description should be removed or corrected. Fix this →

3. Working-tree note (non-blocking)
   SKILL.md shows as an unstaged deletion in the working tree (a known CI-environment artifact noted in the prior bot comment). The committed file is intact; this does not affect the PR.

---

Verdict

Ready to merge pending the two minor fixes above (namespace consistency and verifying reprex_invert()). Both are low-risk documentation-only changes if they need addressing.
• Branch: add-reprexes-skill

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

[d-morrison]

@claude fix failing workflows; use .rbuildignore

[claude]

Claude finished @d-morrison's task in 6m 8s —— View job

---

Fixing failing workflows — done

All four failing checks diagnosed and fixed in commit 525375d.

• R-CMD-check — check-r-package runs with error-on: "note", and the new top-level .claude/ directory triggers a NOTE. Added ^\.claude$ to .Rbuildignore so it's excluded from the build tarball.
• Version increment check — branch and main were both 0.0.0.9054. Bumped DESCRIPTION → 0.0.0.9055.
• Check NEWS.md Changelog — the PR didn't touch NEWS.md. Added a changelog entry for the skill.
• Spellcheck — failed on UCD / SeRG from NEWS.md:39 (a pre-existing line, not the skill). Added both to inst/WORDLIST.

Notes:

• The spellcheck action scans package files (NEWS.md, R/, vignettes, etc.) but not .claude/, so SKILL.md itself wasn't the cause there.
• The working-tree showed .claude/skills/reprexes/SKILL.md as deleted (the CI-environment artifact called out in the earlier review). I left that deletion unstaged, so the committed SKILL.md is untouched — the commit changes only .Rbuildignore, DESCRIPTION, NEWS.md, and inst/WORDLIST.

These changes should clear all four workflows on the next run. I can't re-run CI myself, but the new commit will trigger it.
• Branch: add-reprexes-skill

[d-morrison]

@claude review

[claude]

Claude finished @d-morrison's task in 1m 41s —— View job

---

Review — PR #239 (Add the "reprexes" skill)

Verdict: Looks good to merge. The skill content is accurate, the CI-fix commit (525375d) correctly addresses all four failing checks, and both Copilot reviewer concerns appear to have been addressed in subsequent commits (0cc8e3e, 265edad).

---

What I checked

CI-fix commit (525375d) verified end-to-end:

• .Rbuildignore: ^\.claude$ appended in the existing dir-name style (^\.github$, ^\.devcontainer$, ^\.quarto$). ✓
• DESCRIPTION: bumped to 0.0.0.9055. ✓
• NEWS.md: changelog entry added at the top of the dev version section. ✓
• inst/WORDLIST: contains UCD (line 20, 73) and SeRG (line 18, 72). ✓

Copilot reviewer comments — both already addressed in later commits:

1. ".lintr vs .lintr.R may mislead readers" — the current SKILL.md does not name either file. The R/Quarto section just says "Respect the repo's lint config if the reprex code will be ported back" (SKILL.md:98). The earlier commit 265edad corrected this. For reference, this repo's .Rbuildignore ignores both ^\.lintr$ and ^\.lintr\.R$, and only .lintr.R exists on disk.

2. "tidyverse_update() is an update helper, not a way to record versions" — the current text (SKILL.md:126–130) already makes that distinction explicit:

   If you suspect stale packages are the cause, tidyverse::tidyverse_update() outside the reprex can rule that out — but it updates packages, it doesn't record versions, so don't put it in the reprex itself.

   And the doc separately recommends session_info = TRUE for version-dependent bugs (SKILL.md:113–115, SKILL.md:124) — which is exactly what the reviewer asked for.

Content spot-check (R/Quarto specifics):

• reprex::reprex() runs in a fresh session via callr since 2.0 ✓
• venue options (gh, so, ds, slack, r, html, rtf) ✓
• session_info, std_out_err, wd arguments ✓
• Companion helpers reprex_clean(), reprex_rescue(), reprex_invert() ✓ (all exported in reprex 2.x)
• Rscript /tmp/reprex.R, quarto render … --to html, tempfile(fileext = ".R") ✓
• dput() for inline-serializing minimal data slices ✓

Nits (optional — not blocking)

• SKILL.md:124 says "Capture sessionInfo() (or set session_info = TRUE above)" — minor point, but session_info = TRUE actually emits sessioninfo::session_info() output when the sessioninfo package is installed, falling back to base sessionInfo() otherwise. The doc already notes both upstream at SKILL.md:113, so this is fine as-is.
• .Rbuildignore pattern ^\.claude$ is consistent with how the file ignores other top-level directories (e.g. ^\.github$). Works because R CMD build skips the whole subtree under a matching directory entry — no need for a trailing /.

Re: the working-tree deletion of SKILL.md

git status on the PR-bot's fresh checkout shows .claude/skills/reprexes/SKILL.md as unstaged-deleted — this is the CI-environment artifact already noted in the prior @claude comment. The committed file on add-reprexes-skill is intact (git show HEAD:.claude/skills/reprexes/SKILL.md returns the full 148-line skill). Nothing to fix on the PR.

---

Todo

• Read changed files
• Check repo conventions (.lintr.R, .Rbuildignore style, no root CLAUDE.md)
• Verify CI-fix claims (DESCRIPTION bump, WORDLIST entries, NEWS entry)
• Cross-check Copilot reviewer's concerns
• Post review
  • Branch