claudecoder — Claude Code R

claudecoder is a lightweight helper that scaffolds a new R project ready for use with Claude Code Web. It creates a clean project skeleton, drops in the right CLAUDE.md instruction files, initialises git and pushes the repository to GitHub — and it ships helpers to anonymise your real data into a safe, shareable form.

Installation

# install.packages("remotes")
remotes::install_github("ogdtg/claudecoder")

Create a project

library(claudecoder)

create_ccr_repo(
  name                = "my_analysis",
  directory           = "~/projects",
  template            = "normal",        # or "dashboard", "package"
  github_organisation = NULL,             # NULL = your own account
  private             = TRUE
)

This will:

1. Create an R project at ~/projects/my_analysis.
2. Add the sub-directories R/, anonym_data/ and real_data/.
3. Copy the template CLAUDE.md files to the project root, R/ and anonym_data/.
4. Write an R-style .gitignore that ignores the .Rproj file and the real_data/ directory.
5. Run usethis::use_git() and commit everything.
6. Run usethis::use_github() and push the repository.

Templates

Each template is a directory under inst/templates/ holding three CLAUDE.md files (project root, R/, anonym_data/). Available templates:

• normal — standard R data analysis project
• dashboard — Shiny dashboard project (template = "dashboards" also works)
• package — R package development

list_templates()

Templates are fully editable and expandable: edit the files under inst/templates/<name>/, or add a new sub-directory with its own three CLAUDE.md files, and it becomes available to create_ccr_repo().

Every CLAUDE.md tells Claude Code that the data to work with lives in anonym_data/ and the functions and code live in R/. The anonym_data/CLAUDE.md additionally states that only anonymised data is present and that the real data cannot be shared with Claude Code for data-security reasons.

Anonymising data

Real data lives in real_data/ (git-ignored) and must never be shared. Use the helpers to produce anonymised samples (max 20 rows) in anonym_data/:

# Permute the values within each column (breaks row linkage)
anonymize_dir("real_data", "anonym_data", method = "shuffle")

# Replace each column with type-matched random values
anonymize_dir("real_data", "anonym_data", method = "replace")

The building blocks are also exported:

• shuffle_dataset(data) — independently shuffles each column.
• replace_columns(data) — replaces each column with random values of the same data type (numeric → numeric, character → character, …).

Both are capped to 20 rows by anonymize_dir() before writing.