Skip to content

Security: lelabdev/svforge

SECURITY.md

Security Policy

SVForge ships boilerplate that scaffolds authentication, uploads, jobs, and other security-sensitive surfaces. If you found a vulnerability, this document explains how to report it privately and what to expect.

Supported versions

SVForge packages are versioned independently. Support depends on the package's maturity:

Package state Supported
pre-1.0 versions (0.x) published to npm ❌ evaluation only — no security fixes
main branch (current development) ✅ all fixes land here first
1.0.0 and later: latest published version of the package ✅
1.0.0 and later: any older version ❌ upgrade instead

While packages are pre-1.0, only main is supported: consumers tracking npm releases must upgrade to the fixed version when it ships. Once a package reaches 1.0.0, its latest published version receives security fixes; there is no backport policy for older lines.

Reporting a vulnerability

Do NOT open a public GitHub issue for security reports. Public issues can expose exploitation details before a fix exists.

Report privately through GitHub's private vulnerability reporting:

https://github.com/lelabdev/svforge/security/advisories/new

This channel is owned by the maintainers and is the only supported reporting route. It requires the repository's private vulnerability reporting feature to be enabled by the maintainers (Settings → Code security) — it is currently enabled. If that route is ever unavailable, contact a maintainer directly through their established private channels (maintainer handles are listed in the repository) and reference "SVForge security". Never use public issues, discussions, or pull requests for vulnerability details.

What a useful report contains

  • affected package name(s) and version(s), or the main commit SHA;
  • the affected surface (addon code, generated project code, sv add flow, CLI);
  • a minimal proof of concept or reproduction steps;
  • impact assessment (what an attacker gains);
  • any known mitigation.

What to expect

  • Acknowledgement: within 72 hours of your report.
  • Status updates: at least every 7 days until resolution.
  • Fix targets (from acknowledgement): critical ≤ 7 days, high ≤ 30 days, medium/low ≤ 90 days. Complex dependency-chain issues may take longer; you will be told.
  • Credit: reporters are credited in the advisory and changelog entry unless they prefer anonymity.

Coordinated disclosure

We practice coordinated disclosure. Please give us a reasonable window — default 90 days — to ship a fix before public disclosure. We will keep you informed of progress and agree on a disclosure date together. Avoid automated scanning against hosted demos, and never access, modify, or exfiltrate data that is not yours.

Safe harbor

We consider security research conducted in good faith against this policy to be authorized: we will not pursue legal action for accidental, respectful vulnerability discovery and disclosure. This safe harbor does not cover denial of service, social engineering, spam, physical attacks, or access to data you are not explicitly allowed to process.

Leaked credentials and secrets

If you discover credentials, tokens, or secrets committed in this repository or in a published package:

  1. Do not publish them — report privately through the channel above.
  2. If they are yours, rotate them immediately; rotation is the primary fix.
  3. We will revoke, rotate, and purge the secret from history when applicable.

Vulnerabilities affecting generated projects

SVForge scaffolds projects with third-party foundations (Better Auth, Drizzle, Skeleton, SvelteKit…).

  • Vulnerability in SVForge templates or generated code patterns (insecure default, wrong middleware order, unsafe example) → report here.
  • Vulnerability in a third-party dependency of a generated project → report to that project upstream; also tell us if a pinned version needs a bump.
  • Vulnerability in YOUR generated application's custom code → this is not an SVForge issue; handle it in your own repository. The boilerplate cannot audit application logic written on top of it.

There aren't any published security advisories