Security Scrubbing Knowledge

LLM Provider API Keys

Key Patterns by Provider

Provider	Pattern	Environment Variable
OpenAI	`sk-...`, `sk-proj-...`, `sk-svcacct-...`	`OPENAI_API_KEY`
Anthropic	`sk-ant-...`	`ANTHROPIC_API_KEY`
Google AI	`AIza[0-9A-Za-z-_]{35}`	`GOOGLE_API_KEY`, `GEMINI_API_KEY`
Cohere	40-char alphanumeric	`COHERE_API_KEY`
Mistral	32-char alphanumeric	`MISTRAL_API_KEY`
Groq	`gsk_...`	`GROQ_API_KEY`
Perplexity	`pplx-...`	`PERPLEXITY_API_KEY`
Hugging Face	`hf_...`	`HF_TOKEN`, `HUGGINGFACE_API_KEY`
Replicate	`r8_...`	`REPLICATE_API_TOKEN`
Together AI	64-char hex	`TOGETHER_API_KEY`
xAI (Grok)	varies	`XAI_API_KEY`

Secure API Key Usage

Never do this:

# BAD - hardcoded key
client = OpenAI(api_key="sk-abc123...")

Do this instead:

# GOOD - environment variable
import os
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

Or use dotenv:

from dotenv import load_dotenv
load_dotenv()  # Loads from .env file
client = OpenAI()  # Auto-reads OPENAI_API_KEY

Other Common Secret Patterns

Service	Pattern	Example
AWS Access Key	`AKIA[0-9A-Z]{16}`	`AKIAIOSFODNN7EXAMPLE`
AWS Secret Key	40 char base64	`wJalrXUtnFEMI/K7MDENG...`
GitHub Token	`ghp_[a-zA-Z0-9]{36}`	`ghp_xxxx...`
GitHub PAT	`github_pat_[0-9a-zA-Z_]{82}`	`github_pat_...`
GitLab Token	`glpat-[0-9a-zA-Z-]{20}`	`glpat-xxxx...`
Stripe Live	`sk_live_[a-zA-Z0-9]{24}`	`sk_live_...`
Stripe Test	`sk_test_[a-zA-Z0-9]{24}`	`sk_test_...`
Slack Token	`xox[baprs]-...`	`xoxb-...`
Slack Webhook	`https://hooks.slack.com/...`	webhook URL
Twilio	`SK[0-9a-fA-F]{32}`	`SK...`
SendGrid	`SG.[a-zA-Z0-9]{22}.[a-zA-Z0-9]{43}`	`SG....`
Discord Bot	`[MN][A-Za-z0-9]{23,}...`	Bot token
Firebase	`AAAA[A-Za-z0-9_-]{7}:...`	FCM key
Supabase	`sbp_[a-f0-9]{40}`	`sbp_...`
JWT	`eyJ...` (3 base64 segments)	`eyJhbGc...`
Private Key	`-----BEGIN...PRIVATE KEY-----`	PEM format

Credential Guard

The scrub plugin includes a PreToolUse hook that blocks Claude from reading files containing credentials.

How It Works

Hook intercepts Read operations
Scans target file for 40+ credential patterns
Blocks read if secrets detected
Allows if file is in allowlist

Allowlist

Add trusted files to ~/.claude/plugins/scrub/.guard-allowlist:

# Allow test fixtures
tests/fixtures/
*.test.js
.env.example

Commands

/scrub:guard status      # Check if enabled
/scrub:guard on          # Enable guard
/scrub:guard off         # Disable guard
/scrub:guard allowlist   # Manage allowlist

Where Secrets Hide

Configuration files: .env, config.js, settings.py
Test files: Hardcoded test credentials
Git history: Removed from HEAD but still in commits
Logs: Accidentally logged tokens
Comments: "Temporary" credentials
CI/CD files: .github/workflows/, Jenkinsfile
Notebooks: Jupyter notebooks with inline keys
Docker files: Dockerfile, docker-compose.yml

PII Categories

Identifiable Data

Direct identifiers: Name, SSN, passport number
Contact info: Email, phone, address
Network info: IP address, MAC address
Financial: Credit card, bank account
Biometric: Fingerprints, face data

Log Sanitization

Before sharing logs:

Remove IP addresses
Redact email addresses
Mask user IDs
Strip session tokens
Remove query parameters with PII

Git History Security

When to Scrub History

Accidentally committed credentials
Pushed .env file
Included private keys
Added database dumps with PII

Tools

git-filter-repo (recommended): Fast, safe, feature-rich
git filter-branch: Built-in but slower
BFG Repo Cleaner: Java-based, good for large repos

Post-Scrub Checklist

Force push all branches
Delete all tags and recreate
Notify all collaborators
Rotate ALL exposed credentials
Clear CI/CD caches
GitHub: Contact support to clear caches

File Permissions

Secure Defaults

File Type	Recommended	Why
Private keys	600	Owner only
.env files	600	Contains secrets
Scripts	755	Execute for all, write for owner
Config files	644	Read for all, write for owner
Directories	755	Standard access

Dangerous Permissions

World-writable (o+w): Anyone can modify
World-readable keys: Credentials exposed
SetUID on scripts: Privilege escalation risk

.gitignore Security

Must-Have Entries

# Environment
.env
.env.*
.env.local

# Keys and certificates
*.pem
*.key
*.p12
*.pfx
*.keystore
id_rsa
id_ed25519

# Credentials
credentials.json
secrets.json
*secret*
*credential*

# Logs
*.log
logs/

Best Practices

Never commit secrets - Use environment variables
Use credential guard - Block secrets from reaching AI
Pre-commit hooks - Scan before commit
Least privilege - Minimal file permissions
Rotate regularly - Even if not leaked
Audit trails - Know who accessed what
Defense in depth - Multiple protection layers
.env.example - Document required vars without values

security-scrubbing