yip is a VPN — its whole job is security — but it is pre-1.0 software with one maintainer and no external audit. Treat it accordingly. If you rely on it in a threat-model where compromise is dangerous, read the honest accounting below first.

Report privately via GitHub's private vulnerability reporting, or email the maintainer at the address in the repo profile. Include enough to reproduce (affected version/commit, config, and a description or PoC). We'll acknowledge, work a fix, and credit you unless you'd rather stay anonymous. There is no bounty program.

There are no releases yet; main is the only supported branch. Fixes land there.

• Confidentiality + integrity of tunneled traffic: ChaCha20-Poly1305 AEAD over a Noise-IK session, per-direction keys, a replay window, and ~120 s rekey. Keys are classical today.
• Identity: mesh membership requires a CA-signed certificate; handshake admission checks the peer's static key; a TAI64N timestamp closes an endpoint-hijack-via-replay vector; endpoint roaming follows only authenticated packets; rendezvous registration is signed on the UDP path.
• Content-DPI resistance: no fixed bytes or constant offsets (property-tested); nDPI classifies obfuscated traffic as Unknown; an Xray-REALITY transport parrots real TLS for hostile networks.

• Traffic analysis. There is no timing/padding (DAITA-style) defense yet. The FEC burst shape, packet-size distribution, and constant-interval idle cover traffic are observable to a statistical/flow classifier, and the high-entropy-UDP anomaly is not gated. Use the REALITY/TLS transport, not plain obf_psk, where the network is actively hostile.
• The control plane has known hardening gaps — the TLS relay-front registration is not yet signed, the resolver can be steered by an unauthenticated address, and a couple of registration/handshake replay windows exist. These are tracked openly in the issue tracker (search security).
• No independent audit. The crypto reuses audited primitives (snow, ring), but the protocol composition and the control plane have not been externally reviewed.

If a documented limitation above matters to your threat model, that's a reason to wait, not a vulnerability to report — but a new way to break the stated guarantees is exactly what we want to hear about privately.