nlmixr2llm

nlmixr2llm

LLM-facing documentation for the nlmixr2 pharmacometric modeling ecosystem in R, distributed as an R package and as a Claude Code plugin.

The same content — a single combined nlmixr2verse agent plus per-package skill files for rxode2, nlmixr2, nonmem2rx, monolix2rx, and babelmixr2 — is shipped two ways:

• As an R package, with accessor functions for use as a system prompt with any LLM client (ellmer, the Anthropic SDK, the OpenAI SDK, ...) and installers that write the files into the locations expected by Claude Code, OpenAI Codex CLI, Positron Assistant, and other tools that follow the AGENTS.md convention.
• As a Claude Code plugin via the .claude-plugin/ directory at the repo root, so users who already use the Claude Code plugin marketplace can install without touching R.

Coverage

A single combined nlmixr2verse agent spans the whole ecosystem (the orchestration layer), and each package ships a skill for on-demand depth:

Package	Skill	Purpose
rxode2	✓	ODE-based PK/PD modeling and simulation
nlmixr2	✓	Population PK/PD parameter estimation (SAEM, FOCEi, ...)
nonmem2rx	✓	Convert finished NONMEM runs into rxode2 / nlmixr2 objects
monolix2rx	✓	Convert Monolix projects into rxode2 / nlmixr2 objects
babelmixr2	✓	Fit nlmixr2 models via NONMEM, Monolix, or PKNCA backends

The nlmixr2verse agent covers all five packages and the end-to-end workflow that connects them (author in rxode2 → fit in nlmixr2 → run on other engines with babelmixr2 → import legacy runs with nonmem2rx / monolix2rx).

Install (R package)

# install.packages("remotes")
remotes::install_github("john-harrold/nlmixr2llm")

Usage

Use as a system prompt with any LLM client

library(nlmixr2llm)

prompt <- system_prompt(packages = c("rxode2", "nlmixr2"))

# Example: with ellmer
chat <- ellmer::chat_anthropic(system_prompt = prompt)
chat$chat("Write a one-compartment PK model with first-order absorption in rxode2.")

Install into Claude Code

# User-level (~/.claude/agents/, ~/.claude/skills/)
install_claude_code(scope = "user")

# Or project-local (.claude/agents/, .claude/skills/) under the current project
install_claude_code(scope = "project")

Install into OpenAI Codex CLI

# Project AGENTS.md at the repo root
install_codex(scope = "project", packages = c("rxode2", "nlmixr2"))

# Global ~/.codex/AGENTS.md
install_codex(scope = "user", mode = "append")

Codex enforces a default 32 KiB cap on combined AGENTS.md content. The full corpus (~62 KiB) exceeds that. The combined nlmixr2verse agent is ~29 KiB on its own and is included whole whenever agents are requested, so packages = ... only subsets the skills. For Codex, install include = "agents" (just the agent, ~29 KiB) or include = "skills" with a packages = ... subset.

Install into Positron Assistant

# Project-root agents.md (also picked up by Codex, Cursor, Aider, Zed, ...)
install_positron(workspace = ".", style = "agents_md")

# Per-package .github/instructions/*.instructions.md with applyTo: "**/*.R"
install_positron(workspace = ".", style = "instructions")

Install for any tool that reads AGENTS.md

install_agents_md(path = ".", packages = c("rxode2", "nlmixr2"))

Covered by the agents.md cross-tool spec: Codex, Cursor, Aider, GitHub Copilot, Zed, Warp, Jules, Devin, and others.

Keeping installs in sync

The installers write independent copies, so re-run them after upgrading the package (pass overwrite = TRUE / mode = "write" to refresh existing files). install_claude_code() and install_positron(style = "instructions") keep a manifest (.nlmixr2llm-manifest) and, by default (prune = TRUE), remove files they installed in an earlier version but no longer ship — so a renamed agent or dropped package doesn't leave an orphan behind. Only files nlmixr2llm created are ever removed.