Skip to contents

reproducibleai

Description

{reproducibleai} is an R package of tools to help teams adopt a reproducible workflow for AI-assisted projects. The package provides reusable instruction modules, recipe-based composition, and supporting scaffolds for durable AI-assisted development workflows.

We hold the opinionated view that the ethical use of AI requires data scientists to make well-informed, conscious decisions throughout the analysis process. These tools are designed to help teams experiment with repeatable methods for AI-assisted workflows while keeping those workflows reviewable and auditable.

Core ideas:

  • make chat sessions reproducible by recording the exact instruction modules used,
  • keep instruction content reviewable as static markdown,
  • support recipe-based composition of reusable instruction modules,
  • provide lightweight handler-backed installation where some modules require supporting repository structure,
  • support an auditable workflow as teams adopt AI tooling.

Instruction model

reproducibleai is instruction-first.

Canonical instruction text is stored as static markdown in inst/instructions/. Public workflows compose instruction modules by name and install them into a target repository with use_instructions().

Internally, installation is now handler-backed:

  • every public instruction module has a corresponding handler,
  • simple modules install canonical text only,
  • config-aware modules can also scaffold supporting repository structure.

This preserves static, reviewable instruction text while allowing modules such as development-governance to establish the repository structure they depend on.

Authors

Installation

You can install the development version from GitHub with:

# install.packages("pak")
pak::pak("MVR-GIS/reproducibleai")

Quick start

1) See which instruction modules are available 

library(reproducibleai)
instructions_available()
#> [1] "chat-manual"            "development-governance" "goals"                  "parameterized-help"    
#> [5] "python-package"         "quarto-book"            "r-package"              "shiny-golem"           
#> [9] "user-manual"
# returns the public module names you can compose into recipes
recipes <- instructions_recipes()
recipes$r_package
#> [1] "chat-manual" "goals"       "r-package"

3) Install instructions into your project 

use_instructions(recipes$r_package)
# installs into dev/instructions/ by default
# also writes dev/instructions/CHAT_INSTRUCTIONS.md

4) Use development-governance when you want repository scaffolding

The development-governance module is the first config-aware module. In addition to installing its instruction text, it scaffolds a standard development-governance structure under dev/, including:

  • dev/05_plan.md
  • dev/10_design.md
  • dev/40_schemas.md
  • dev/decisions/
  • dev/instructions/
  • dev/sessions/

Example:

use_instructions(c("chat-manual", "development-governance"))

This is useful for repositories that want durable AI-assisted development artifacts such as plans, design notes, schema contracts, decision records, reusable instructions, and archived sessions.

Notes on installation behavior

use_instructions() is designed around the repository layout:

  • dev/instructions/ for installed instruction modules
  • dev/instructions/CHAT_INSTRUCTIONS.md as the session entrypoint

Modules are installed through internal handlers. For most modules this means installing static instruction text. For selected modules, handlers may also scaffold required supporting files or directories.

The package currently treats canonical instruction markdown as the authoritative module content; handlers support installation and scaffolding but do not dynamically rewrite substantive instruction text.

Vignettes 

# What vignettes exist for this package?
vignette(package = "reproducibleai")

Bug reports

Please open an issue at: https://github.com/MVR-GIS/reproducibleai/issues