Security First - Protecting Client Data

Critical security configuration before using AI - setting up .cursorignore and protecting sensitive data

The Problem

When you use Cursor with @codebase or Composer, AI can read your project files. If you're not careful, you'll accidentally share:

Client API keys
Database credentials
Private certificates
Customer data
Internal infrastructure details

This is a breach of trust and potentially illegal (GDPR, etc.).

The Solution: Configure .cursorignore

Create .cursorignore in project root (FIRST THING, before any AI work):

# Environment files
.env
.env.*
*.env
.env.local
.env.production
.env.staging

# Credentials & Secrets
**/secrets/
**/credentials/
*.key
*.pem
*.p12
*.pfx
*.cert
*.crt
config/secrets.yml
config/credentials.yml.enc

# Database dumps
*.sql
*.dump
*.backup
dumps/

# Logs (may contain sensitive data)
*.log
logs/
npm-debug.log*
yarn-debug.log*
yarn-error.log*

# Cloud provider configs
.aws/
.azure/
.gcloud/
terraform.tfvars
terraform.tfstate
*.tfstate

# IDE & Editor configs that might contain paths
.vscode/settings.json
.idea/

# Test data that might be real
test-data/real/
fixtures/production/

# Any file with "secret", "password", "token" in name
*secret*
*password*
*token*
*auth*key*

# Client-specific (add your own)
client-data/
prod-backup/

Verify It Works

Test before using AI:

# 1. Create .cursorignore with above content

# 2. Test it
echo "SECRET_KEY=test123" > .env

# 3. Ask Cursor
"Search @codebase for the word SECRET_KEY"

# 4. AI should say: "I cannot find SECRET_KEY in the codebase"
# If AI finds it → your .cursorignore is not working

Git Ignore vs Cursor Ignore

Important: They serve different purposes!

.gitignore → Prevents committing to repository
.cursorignore → Prevents AI from reading files

Always have both:

# .gitignore (prevent commits)
.env
*.log
node_modules/

# .cursorignore (prevent AI access)
.env
*.log
# Note: node_modules is OK for AI to read (it's public code)

Team Practice

In every new project:

## Security Checklist (before AI work)

- [ ] Created .cursorignore
- [ ] Tested with dummy .env file
- [ ] Verified AI can't see secrets
- [ ] Added client-specific paths
- [ ] Reviewed with team lead
- [ ] Documented in project README

Red Flags

Stop immediately and check .cursorignore if:

AI suggests actual API keys in responses
AI mentions real database connection strings
AI references production URLs you didn't share
AI knows customer names from data files

For Limestone Digital Team

Our standard .cursorignore template is in: /templates/cursorignore-template

Copy it to every new client project:

cp /templates/cursorignore-template .cursorignore
# Then add client-specific paths

Code review requirement: Every project must have .cursorignore before first AI-assisted PR.

What If I Already Shared Secrets?

If you accidentally shared secrets with AI (via chat or @codebase):

Immediate actions:

Assume compromised - AI providers store conversation history

Rotate all secrets immediately:

# Database passwords
# API keys
# JWT secrets
# OAuth client secrets
# Any credentials mentioned

Notify client - Be transparent, explain what happened
Document incident - What was exposed, what was rotated
Update .cursorignore - Prevent it happening again

Do NOT:

Hope nobody noticed
Delete chat history and pretend it didn't happen
Wait to see if something bad happens

Example incident response:

## Security Incident Report

**Date:** 2024-12-04
**Severity:** High
**Status:** Resolved

**What happened:**
Developer used `@codebase` before configuring .cursorignore.
File `.env.production` was accessible to AI.

**What was exposed:**

- Database connection string
- AWS S3 access keys
- SendGrid API key

**Actions taken:**

1. Rotated database password (10 minutes)
2. Rotated AWS keys (15 minutes)
3. Rotated SendGrid key (5 minutes)
4. Configured .cursorignore
5. Notified client
6. Updated team training

**Prevention:**

- Added .cursorignore to project template
- Added security checklist to onboarding
- Required .cursorignore in PR reviews

Cursor AI Data Policy

What Anthropic (Claude) stores:

Conversation history (for improving model)
Code snippets you share in chat
Files you reference with @ mentions

What they DON'T store:

Files blocked by .cursorignore
Your entire codebase (only what you explicitly share)

Best practice: Assume anything shared with AI could be seen by:

Anthropic employees (for model training)
Potential data breaches
Subpoenas / legal requests

Golden rule: Never share what you wouldn't put in a public GitHub repo.

Security First - Protecting Client Data

On this page