Use GitHub's private vulnerability reporting (Security tab → Report a vulnerability) — reports reach the maintainer privately, with no email required. If you prefer email: sagar@sagarwaidande.org.
Please do not open public GitHub issues for suspected vulnerabilities.
We commit to:
| Stage | Target |
|---|---|
| Acknowledge receipt | within 2 business days |
| Initial triage + severity assessment | within 5 business days |
| Patch landed in main | within 30 days for high/critical, 90 days for medium, best-effort for low |
| Public disclosure (with credit) | after a fixed release is available, coordinated with the reporter |
Hardcount is pre-1.0. Only the latest release gets security fixes.
- A fix lands in
mainand ships in the next release, a patch or a minor release. - Older releases get no backported fixes. To get a fix, upgrade to the latest release.
- 0.9.0 is the first release named Hardcount (the project was called Velox). It cannot take over a database Velox made, from 0.8.x or from a build of
main. So on 0.8.x, every later fix needs a fresh install. - Before 1.0, an upgrade can need more than a restart: read Before upgrading in the CHANGELOG first. See Versioning before 1.0.
- The Hardcount Go binary (
cmd/hardcount,cmd/hardcount-bootstrap,cmd/hardcount-bench-seed,cmd/hardcount-doctor) - Code under
internal/andpkg/ - The migration runner and schema in
internal/platform/migrate/ - The web-v2 dashboard (
web-v2/) - Docker Compose deployment config under
deploy/compose/ - Outbound webhook signing, inbound Stripe webhook verification, API key handling, session/cookie auth, RLS (row-level security) policy enforcement, AES-GCM encryption-at-rest, HMAC blind index for email (a keyed hash that allows equality lookup without storing the plaintext address)
- Vulnerabilities in the deployment environment of the person who runs the server (Kubernetes cluster, managed Postgres, load balancer, secrets store, IAM)
- Vulnerabilities in third-party services that Hardcount integrates with (Stripe, cloud providers, email providers, S3, KMS) — report those to the vendor
- Configuration mistakes by the person who runs the server (e.g., running with
HARDCOUNT_ENCRYPTION_KEYunset in a non-production env, or exposing the dashboard without TLS termination) - DoS via traffic flooding (a property of the load balancer and WAF of the person who runs the server, not Hardcount)
- Self-XSS, social engineering, physical attacks
- Reports against forks or vendored copies of Hardcount — please reproduce on the canonical
mainbranch first
Hardcount has had no external security audit or penetration test yet.
In place:
- Account isolation: row-level security on most API requests; the exceptions are listed in Architecture. Hardcount refuses to start if its request-time database user can bypass it.
- Secrets at rest: AES-256-GCM encryption for Stripe keys, webhook signing secrets and customer personal data, required in production; an HMAC blind index (a keyed hash) to find a customer by email without storing the address in plain text.
- Credentials: passwords as bcrypt (cost 12); sessions and one-time tokens stored only as SHA-256 hashes, and API keys as salted SHA-256 hashes.
- Sign-in (who can sign in):
- about 30,000 common leaked passwords are refused when a password is set;
- repeated wrong passwords slow down instead of locking the account;
- the sign-in cookie travels only over HTTPS in production.
- Requests: CSRF checks on the dashboard, security headers and HSTS, rate limiting. TLS itself is ended by a proxy run by the person who runs the server. The dashboard's pages, the hosted invoice page among them, refuse to be shown in a frame on another site.
- Logs: the token in a link, such as a hosted invoice or invite link, is redacted from the API's access log and traces, and from the Compose stack's nginx access logs.
- Webhooks: outbound events signed with HMAC-SHA256; inbound Stripe events verified.
- Audit log: append-only; a database trigger refuses any update or delete.
- Supply chain: CI runs
govulncheck, scans the code with CodeQL and scans the images with Trivy; release images carry an SBOM and are signed (Verify a release). GitHub secret scanning and push protection are on, so a commit that holds a known secret format is refused before it is public.
Known gaps:
- No two-factor sign-in. It is built when a team running Hardcount asks for it, or when a customer's security review requires it (ADR-095). Until then, a team can put the dashboard's sign-in behind its own SSO or VPN.
- No built-in rotation of
HARDCOUNT_ENCRYPTION_KEYorHARDCOUNT_EMAIL_BIDX_KEY. - No external penetration test on record.
- No SOC 2 control mapping until a customer requires one (Later in the roadmap).
The attacks Hardcount is built to stop, and your part as the person who runs it: Threat model. What data it stores, and what leaves your server: What data Hardcount stores.
If you can help close any of these, open an issue: CONTRIBUTING says how a proposed change is taken up and credited.
We will not pursue legal action against good-faith security research that:
- Does not access, modify, or destroy data belonging to anyone other than the researcher
- Does not degrade availability for other users
- Stays within the in-scope list above
- Coordinates disclosure with us before going public
If a researcher inadvertently violates this policy in good faith, please tell us promptly and we'll work it out.