SkillGoblin is a self-hosted homelab learning platform maintained as an open-source side project. There is no commercial entity behind it, no security team, and no SLA. That said, security issues are taken seriously — please follow the guidance below.
Only the main branch (latest commit) is supported. There are no LTS releases, and patches are not back-ported. If you're running an older snapshot, the upgrade path is to pull the latest image and restart the container.
For low-impact, non-exploitable issues (CVE-style dep advisories that Dependabot already flags, hardening recommendations, configuration suggestions): open a regular GitHub issue and tag it security. Public discussion is fine — these are usually obvious from the codebase anyway.
For exploitable issues (auth bypass, RCE, privilege escalation, data exfiltration, anything where a working PoC exists or is straightforward to construct): use GitHub's private vulnerability reporting. This routes the report directly to the maintainer without exposing it publicly until a fix lands. Please include:
- A clear description of the issue
- Steps to reproduce, or a minimal PoC
- Affected versions / commits if known
- Your suggested severity (CVSS or just gut feeling — both are fine)
Acknowledgement is best-effort within a few days. Fix turnaround depends on severity and complexity.
SkillGoblin is designed for trusted local networks (a homelab, a small classroom, a family server). The defaults assume the operator controls the network and reverse proxy. The auth model protects against:
- Account takeover within the deployed instance (auth bypass, session forgery, credential leakage, role escalation)
- Server-side data exposure to unauthenticated clients
- Persistent XSS in user-controllable content
- Path traversal and other input-validation bugs
Out of scope (or addressed by deployment, not the app):
- TLS termination — runs as a plain HTTP service. Operators put it behind nginx / Caddy / Traefik for HTTPS.
- DDoS / volumetric attacks — the app has basic rate limiting on auth, but bulk traffic is the reverse proxy's job.
- Reports based on running the container with default test credentials (
ADMIN_NAME/ADMIN_PASSWORDenv-var bootstrap is documented as required first-run setup — leaving these as defaults is operator error, not a vulnerability). - Reports based on exposing the service to the public internet without a reverse proxy / authentication layer in front of it. The threat model assumes a trusted LAN.
Upgraded installations can contain active non-admin accounts created by older releases without a password or PIN. Their first visitor may set a credential and claim the account because no prior secret exists to verify ownership. This recovery flow is retained for trusted-LAN deployments and is hardened as follows:
- Admin accounts are never claimable this way. A credential-less admin (possible after upgrading from a pre-auth-hardening release) is refused by the self-claim endpoint; only another administrator can set its credentials from the Admin Panel. This closes the unauthenticated admin-takeover path.
- Claiming is atomic. A compare-and-swap ensures only one racing request can ever claim a given account; the loser gets a 409 and no session.
If the wrong person claims a non-admin account, an administrator can reset its password or PIN from the Admin Panel. Do not rely on this flow on an untrusted network.
The defaults suit a plain-HTTP trusted LAN. When exposing the service through a reverse proxy (or to a less-trusted network), set these so the app's defenses line up with your topology:
| Env var | Default | When to set |
|---|---|---|
TRUST_PROXY_HOPS |
0 (ignore X-Forwarded-For) |
Set to the number of trusted reverse proxies in front of the app so login rate-limiting keys on the real client IP instead of a spoofable header. With 0, a client-supplied X-Forwarded-For is ignored and the transport peer is used — safe, but every request behind a proxy shares one IP bucket. |
COOKIE_SECURE |
auto-detect | Set to true when TLS is terminated by a proxy that does not forward trusted X-Forwarded-Proto=https. This also enables HSTS. false forces both off for intentional plain-HTTP LAN use. |
Unsafe browser requests with an Origin header must be same-origin. CLI and other non-browser clients that omit browser origin metadata remain supported. Responses include baseline CSP, clickjacking, MIME-sniffing, referrer, permissions, and cross-origin resource headers on both HTTP and HTTPS; HSTS is emitted only when HTTPS is explicitly detected or configured.
The app also caps request bodies (256 KiB for JSON APIs) to blunt memory-
exhaustion attempts. As defense in depth, configure a body-size limit at your
reverse proxy too (e.g. nginx client_max_body_size).
When fixes ship, the commit message will credit the reporter (with permission) by name or handle. There is no bug bounty program.