How to use the PK/PD Modeling Assistant agent skill

What an agent skill is and what this one does

An agent skill is a folder of instructions and supporting files that an AI agent can read when a task matches it. This skill's main file, SKILL.md, describes a workflow for PK/PD work: checking units and definitions, computing exposure measures such as AUC, Cmax and half-life from concentration-time data, fitting simple individual compartmental models, and recording assumptions.

The skill also includes reference notes on topics such as non-compartmental analysis conventions, bioequivalence, population PK, PK/PD and exposure-response, drug-drug interactions and antimicrobial PK/PD, plus two templates: an NCA reporting checklist and a population PK analysis plan. It is explicitly framed as exploratory. It does not give clinical dosing advice or certify regulatory submissions.

• Non-compartmental analysis (NCA) with AUC, Cmax, lambda_z and clearance-type outputs
• Individual 1-, 2- and 3-compartment model fitting, with comparison by AIC and BIC
• Population PK dataset checks for NONMEM-style data
• Reporting checklists and analysis-plan templates

Install the skill from the ZIP download

Download the skill ZIP from the item page. Then choose one of two routes. For a personal setup that applies to all your projects, unzip the download into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. Either way, the folder should be named pkpd-modeling so the agent can find SKILL.md inside it.

If you use the Claude apps rather than a local folder, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/ (available across projects)
• Project install: .claude/skills/ inside your project folder
• App install: Settings → Capabilities → Skills, then upload

Install the Python requirements for the scripts

The scripts need Python 3.12 or newer, NumPy 2 or newer, and SciPy. The skill's compatibility notes say it needs no network access and no proprietary software at runtime. Use an isolated environment for these packages. Commands in the skill are run from the scripts folder inside the skill.

• Python 3.12 or newer
• NumPy 2 or newer
• SciPy
• No network access and no proprietary engine needed at runtime

How to trigger the skill with natural requests

You do not need to name the skill. The agent is more likely to load it when your request uses the terms the skill is written around. Useful trigger terms include pharmacokinetics, NCA, AUC, Cmax, half-life, clearance, volume of distribution, compartmental model, popPK, exposure-response, bioequivalence, DDI and therapeutic drug monitoring.

Be specific about your data and goal. State the dose, route, units and what you want computed. The skill expects you to record units and definitions first, so include them in your request.

• "Run an NCA on this concentration-time file. Dose 100 mg, oral, report AUC0-24 and lambda_z half-life."
• "Fit 1, 2 and 3 compartment IV bolus models to these data and compare AIC and BIC."
• "Check this popPK dataset for NONMEM-style ordering problems before I fit it."
• "Draft a population PK analysis plan for a bioequivalence study, with the intended use stated."

Run the non-compartmental analysis script

For NCA, the main script reads a CSV with the columns id, time and conc (id is optional). Optional columns include dose, tau, tinf and blq. The script accepts a route flag and can report partial AUC between chosen times. Output goes to stdout as a table, TSV or JSON, and notes go to stderr.

Check the script's findings even when it exits cleanly. Exit code 0 means no findings, 1 means findings, and 2 means invalid input. Terminal-phase fits, extrapolation percentage and BLQ handling all change the answer, so the skill asks the agent to record those choices.

• Times must be non-negative and distinct within each profile
• BLQ handling is one global rule: zero, half-lloq or missing
• Choose the AUC method: linear, linup-logdown or log
• Automatic lambda_z selection is not the same as Phoenix Best Fit

Know what the skill will not do

The skill is built with several limits, and they matter for how you use the output. Its individual fitting is not nonlinear mixed-effects (NLME) population estimation. Its outputs are not clinical evidence and it does not recommend patient doses. Analytical checks and synthetic tests confirm numerical behaviour only, not model validity in a patient population.

Because the skill runs local scripts, review them before use in a shared environment, as you would with any code an agent executes.

• Not for patient-specific dosing decisions
• Not for regulatory submissions; results need validation by qualified specialists
• Not for production population PK estimation; fitting is individual-only
• Estimation tools such as NONMEM, Monolix, Phoenix and Simcyp are licensed separately and are never called by these scripts