Organization Rules

PumasAide's knowledge base covers pharmacometric concepts and API usage, but not organization-specific thresholds or methodology preferences. Instead of assuming defaults, the AI asks you directly when a decision depends on acceptance criteria, statistical thresholds, or methodology choices.

Organization rules let you codify your team's standards as skill files that the AI discovers automatically — so it applies your criteria consistently without asking each time.

What Can Be Configured

Rules span 9 categories:

Category	Examples
Statistical Thresholds	ΔOFV cutoffs, RSE% limits, shrinkage bands, condition number
NCA Quality	Terminal slope R² minimum, AUC extrapolation limit, minimum λz points
Regulatory Framework	Default agency (FDA/EMA), BE limits, NTI handling, BLQ methods
Methodology	Covariate method (FME/SCM), estimation progression, allometric scaling
Clinical Relevance	Covariate effect size bands, PTA targets, dose proportionality criteria
Parameter Defaults	Typical CL/V/Ka ranges, IIV starting values, error model defaults
Study Design	Sample sizes by phase, washout rules, dropout assumptions
Data Handling	Imputation methods, missing data thresholds, outlier detection, units
Naming Conventions	Model naming, parameter naming style, file naming rules

You don't need to configure all categories. For any category without rules, the AI asks you during the analysis.

Setting Up Rules

Ask the AI to run the org rules workflow:

Configure organization rules for our team

Set up org rules for my-org

The AI walks through each category interactively, collecting your values one category at a time. You can skip categories or individual items. At the end, it generates skill files and tells you where to place them.

Start from existing documents

If your team already has SOPs, style guides, or regulatory templates, paste or attach the plain text and ask the agent to generate skill files from them. The agent extracts the relevant thresholds and conventions, so you don't need to re-enter values manually.

Where Rules Live

Rules are stored as skill files — markdown with YAML frontmatter. The AI generates them into a skills/ directory in your project. From there, move them to the location your agent expects:

Agent	Location
Claude Code (project)	`.claude/skills/`
Claude Code (global)	`~/.claude/skills/`
Cursor	`.cursor/rules/`
Other agents	Agent-specific location

Project-level locations (.claude/skills/, .cursor/rules/) hold team standards — commit them to version control so everyone on the team shares the same rules. Global locations (~/.claude/skills/) hold personal preferences that apply across all your projects.

For broader distribution, you can also package skill files as a plugin or extension through the marketplace mechanisms that Claude Code and other agents support, ensuring consistency across teams or an entire organization.

Example

After running the org rules workflow for NCA quality, you might get a file at skills/my-org-nca-quality/SKILL.md:

---
description: Organization NCA quality acceptance criteria. Consult when evaluating terminal slope fits and AUC reliability.
---

# NCA Quality Rules

## Terminal Slope

- Adjusted R² minimum: 0.8
- Minimum points for λz: 3
- AUC extrapolation limit: 20%

## Sample Coverage

- Minimum sampling duration: 3 half-lives for AUC∞ reliability

With this in place, the AI applies these thresholds automatically instead of asking.

How Rules Interact with Analyses

Without org rules, PumasAide's knowledge base explains what each metric means and why it matters, but defers threshold decisions to you:

"Check rsqadjkel against your organization's acceptance criteria."

With org rules, the AI applies your standards directly:

"Subject 7 has rsqadjkel = 0.72, below your threshold of 0.8. Flagging for review."

This applies across all analyses — NCA quality checks, model diagnostics, covariate screening, and more.

Walkthroughs use example thresholds

The NCA Analysis Walkthrough and Population PK Walkthrough show specific threshold values (e.g., rsqadjkel >= 0.8, shrinkage < 20-30%) for illustration. In practice, the AI uses your organization's rules if configured, or asks you directly if not.

Updating Rules

Skill files are living documents — update them as your standards evolve. You can edit them directly in any text editor, or ask the AI:

Update our org rules for NCA quality

The AI reads your current skill files, shows the existing values, and collects updates. Since skill files are plain markdown, you can also refine wording, add notes, or reorganise sections by hand.