Enforcing Standards

Lab-wide consistency through Claude Code

Modified

February 1, 2026

Claude Code / Advanced / Enforcing Standards

Advanced 8 min read Phase 4: Advanced

← MCP Servers

Prerequisites

  • Skills (custom /commands)
  • Hooks (automation triggers)

This guide shows how to configure Claude Code to automatically enforce lab standards across all projects.

The Consistency Stack

Multiple layers work together to maintain consistency:

Global CLAUDE.md          # Lab standards (all projects)
    ↓
Project CLAUDE.md         # Project-specific rules
    ↓
globals.yml               # Configuration values
    ↓
/review skill             # Code review with checklist
    ↓
Pre-commit hooks          # Block non-compliant commits
    ↓
review-checklist.yml      # Known issue patterns

Setting Up New Projects

1. Copy from Templates

Lab templates include pre-configured .claude/:

# Clone a template
gh repo create rashidlab/my-project \
  --template rashidlab/template-research-project --private

# Or copy .claude/ from template
cp -r ~/rashid-lab-setup/template-research-project/.claude .

2. Customize Project CLAUDE.md

Edit .claude/CLAUDE.md for project-specific context:

# Project: My Research Project

## Purpose
This project implements a Bayesian adaptive trial design...

## Key Files
- _targets.R: Pipeline definition
- R/tar_functions.R: Core functions
- config/globals.yml: All configurable parameters

## Project-Specific Rules
- Never hardcode fidelity values (use globals.yml)
- Use canonical names: total_n (not nmax), type1 (not alpha)
- All simulations must preserve seeds

3. Configure globals.yml

# config/globals.yml
simulation:
  n_iterations: 1000
  n_bootstrap: 500
  seed: 42

model:
  prior_mean: 0
  prior_sd: 1

output:
  figure_format: "pdf"
  table_format: "csv"

4. Create Review Checklist

# .claude/review-checklist.yml
patterns:
  - name: "Hardcoded fidelity"
    regex: "fidelity_low\\s*=\\s*\\d"
    severity: critical
    message: "Use globals$simulation$fidelity_low instead"

  - name: "Wrong parameter name"
    regex: "\\bnmax\\b"
    severity: warning
    message: "Use total_n instead of nmax"

  - name: "Math notation"
    regex: "\\\\mathbf\\{"
    severity: warning
    message: "Use \\boldsymbol{ instead"

file_triggers:
  critical:
    - "globals.yml"
    - "_targets.R"
    - "R/tar_functions.R"
  warning:
    - "*.csv"
    - "manuscript/*.qmd"

5. Enable Hooks

// .claude/settings.json
{
  "permissions": {
    "allow": [
      "Bash(Rscript *)",
      "Bash(git *)",
      "Bash(make *)"
    ]
  },
  "hooks": {
    "pre-commit": {
      "command": "Rscript scripts/validate_pipeline_consistency.R"
    },
    "session-stop": {
      "agent": "lab-reviewer"
    }
  }
}

Enforcement Points

At Code Review

The /review skill checks:

  1. Logic errors — Bugs, incorrect algorithms
  2. Security issues — Hardcoded secrets, injection vulnerabilities
  3. Data integrity — Seed preservation, NA handling
  4. Lab standards — data.table not tidyverse, globals.yml usage
  5. Project patterns — Checklist from review-checklist.yml

At Commit Time

Pre-commit hooks validate:

  • globals.yml values are used correctly
  • No hardcoded values in manuscripts
  • Critical files reviewed
  • Tests pass

At Session End

Stop hooks trigger final review:

  • Summary of changes made
  • Issues found in session
  • Recommendations for next session

Integration with Project Consistency Framework

Claude Code works with the lab’s Project Consistency Framework:

Reading Configuration

Claude automatically reads:

config/globals.yml              # Simulation parameters
docs/plans/*.md                 # Design decisions
docs/DATA_PROVENANCE.md         # Data sources

Checking Consistency

The /validate skill verifies:

# /validate checks:
1. globals.yml values used in R code
2. Manuscript uses dynamic values
3. Figures match targets outputs
4. Data provenance documented
5. Claims match analysis results

Documenting Changes

Claude updates consistency files:

> Add an entry to DATA_PROVENANCE.md for the new dataset
> Update the changelog with today's changes
> Document this design decision in docs/plans/

Customizing Enforcement

Project-Specific Patterns

Add patterns to review-checklist.yml:

patterns:
  - name: "Project-specific check"
    regex: "your_pattern_here"
    severity: warning
    message: "Explanation of issue"

Disable Specific Checks

In project CLAUDE.md:

## Exceptions

For this project, we use tidyverse because [reason].
Do not flag tidyverse usage as a lab standard violation.

Add Custom Validation

Create project-specific validation scripts:

# scripts/validate_pipeline_consistency.R

# Load project-specific validation rules
source("R/validation_rules.R")

# Run checks
results <- run_all_validations()

# Report
if (any_failures(results)) {
  report_failures(results)
  quit(status = 1)
}

Lab-Wide vs Project-Specific

Enforcement Level Example
Use data.table Lab-wide In global CLAUDE.md
Use globals.yml Lab-wide In global CLAUDE.md
Specific variable names Project In project CLAUDE.md
Review checklist patterns Project In review-checklist.yml

Verification Workflow

Before Starting Work

> Check that this project is properly configured for enforcement

Claude verifies: - CLAUDE.md exists (global and project) - globals.yml exists and is valid - review-checklist.yml has patterns - Hooks are configured

During Work

Claude automatically: - References globals.yml for values - Follows lab coding standards - Warns about potential issues

Before Committing

> /review

Claude reviews all changes against the full checklist.

Before Submitting

> /validate

Claude runs complete consistency validation.

Troubleshooting

“Claude ignores my standards”

Check configuration is loaded:

> What CLAUDE.md files are you reading?
> What's in my review checklist?

“Hook keeps failing”

Test the hook manually:

Rscript scripts/validate_pipeline_consistency.R
echo $?  # Should be 0 for success

“Patterns don’t match”

Test regex patterns:

grep -E "your_pattern" R/*.R

See Also