VARRD handles financial data, authentication credentials, and payment processing. We take security seriously.
For quickest results do not open a public issue. Email security findings to:
Include:
- Description of the vulnerability
- Steps to reproduce
- Impact assessment (what an attacker could do)
- Any proof-of-concept code
| Area | Examples |
|---|---|
| Authentication | JWT handling, passkey system, agent provisioning, session management |
| Billing | Credit deduction, payment replay, balance manipulation, Stripe webhook handling |
| MCP Server | Tool injection, parameter manipulation, rate limit bypass, unauthorized data access |
| API | Endpoint auth bypass, IDOR, injection, information disclosure |
| Data Exposure | Edge formulas/methodology leaking at unpaid tiers, user data cross-contamination |
- Rate limiting thresholds (we're aware, actively tuning)
- Self-XSS or attacks requiring physical device access
- Denial of service (please don't load-test production)
- Social engineering of VARRD staff
- Vulnerabilities in third-party services (Stripe, AWS, Anthropic)
| Step | Target |
|---|---|
| Acknowledgment | 24 hours |
| Initial assessment | 72 hours |
| Fix deployed | 7 days for critical, 30 days for moderate |
| Disclosure | Coordinated, after fix is live |
We've run multiple rounds of penetration testing. Current protections include:
- Atomic idempotency on all payment paths (DynamoDB ConditionExpression)
- Cross-user payment replay blocked (PaymentIntent metadata verification)
- Stripe webhook signature verification
- Per-agent billing isolation (balance deduction with ConditionExpression)
- IP rate limiting on agent provisioning and API endpoints
- JWT expiry enforcement, brute-force lockout (15 fails in 5 min)
- Dormant edges completely blocked from all tiers
- Request body size limits (1MB) and batch limits (10)
- No secrets in client-facing responses or documentation
- Lookahead bias verification on all validated edges
We'll credit researchers who report valid vulnerabilities (with permission) in our release notes. No bug bounty program at this time, but significant findings will be acknowledged.