Security
How Creo finds, fixes and reports security issues.
What the code and our accounts do today. Anything not yet in place says so.
Last reviewed November 20, 2026. Next review due by November 20, 2027. The same policy is SECURITY.md in the repository.
Reporting a vulnerability
Use the contact form. Choose the topic “Something else” and start the message with “Security”. Include the URL or route, the steps to reproduce, and what an attacker could do with it. Please don't put access tokens or other people's data in the report.
There is no bug bounty, and we don't pay for reports. We will tell you whether we could reproduce the issue and when it is fixed. We won't take legal action against good-faith research that avoids other users' data, doesn't degrade the service, and gives us time to fix the issue before it's disclosed.
Vulnerability management
- Dependabot alerts and security updates are on. A known-vulnerable dependency gets its own pull request as soon as GitHub publishes an advisory.
- Dependabot version updates run weekly for Python packages and GitHub Actions.
- pip-audit runs in CI on every push and fails the build on any known vulnerability in a locked dependency. Accepted exceptions are listed, with a date and a reason, in an allowlist file that stays empty unless no fix exists yet.
- Bandit static analysis runs in CI on the Python code and fails on medium or high findings. Reviewed false positives are recorded in a baseline file.
- gitleaks scans the full git history in CI for committed credentials.
Not yet in place: GitHub secret scanning, push protection and CodeQL, which GitHub doesn't offer for private repositories on our free plan. gitleaks covers secret scanning but only after a push; it can't block one.
Patch deadlines, from when we learn of an issue: critical within 7 days, high within 30 days, medium and low in the next routine dependency update. If no fix exists we mitigate (by disabling the affected feature or pinning around it) and apply the fix once it's released.
Access control
Creo is run by one operator. The accounts that can change production are GitHub (a push to main deploys), Render (hosting and production secrets), Neon (the database), Stripe (payments), and Plaid and SnapTrade (brokerage connections).
Every account with access to production systems or customer data must have multi-factor authentication. We use authenticator apps or passkeys where the provider supports them and are moving remaining SMS-based factors to them.
Not yet in place: the GitHub organization doesn't enforce 2FA as a setting, and main has no branch protection, which needs a paid GitHub plan for a private repository.
Production secrets are set only in Render's environment and are never in the repository. API keys are stored only as a SHA-256 hash and can be revoked at any time. Admin routes need a separate token and answer 404 without it.
Encryption
- In transit: itscreo.com accepts only TLS 1.2 and TLS 1.3 and sends HSTS for one year. Calls to Neon, SEC EDGAR, Plaid, SnapTrade and Stripe use TLS with certificate verification.
- Brokerage credentials: every Plaid access token and SnapTrade user secret is sealed with AES-256-GCM before it reaches the database, with a fresh random nonce and the row's identity bound in, so a sealed value can't be moved to another row. The key lives only in Render's environment.
- At rest: Neon encrypts database storage. Holdings are never written to the database.
Logging
- Request and response bodies are never logged.
- The access log records the client address, method, path with the query string removed, and status. Query strings are removed because the analyzer sends ticker symbols in some of them.
- Brokerage holdings are fetched live, held in memory for up to 10 minutes, and never written to the database or logs. The browser analyzer sends tickers only, never quantities.
- Keyed API calls keep time, route, status, credits and request ID for 30 days.
- Render keeps its own platform request logs under its own policies.
Incident response
- Contain: rotate the affected secrets in Render and redeploy, revoke affected API keys, and if brokerage tokens may be exposed, revoke every connection (Plaid
/item/remove, SnapTrade user deletion) and rotate the token key. Not yet in place: a single command that revokes every connection; today it is the per-connection disconnect run over every stored row. - Investigate with Render's logs, the request log and the git history.
- Notify affected account holders by email, post a notice on the site and the status page, and notify Plaid, SnapTrade or Stripe when their credentials or users are involved.
- Fix and review: fix the cause, add a check that would have caught it, and update this policy.
Review
This policy is reviewed at least once a year, and whenever hosting, a data processor or the handling of brokerage data changes.