Please report security issues privately rather than as a public issue, either through GitHub's private vulnerability reporting or by email to boj@19pine.ai. I'll acknowledge within a few days. This is a research project maintained by one person, so please be patient with timelines.

OneBarrier is a research prototype. It hasn't been audited, and its threat model is fail-stop crashes, not a malicious actor. Nodes are assumed to fail by stopping. Byzantine faults are out of scope by design.

Some sharp edges worth knowing about if you plan to run it:

The determinism stack deliberately destroys entropy. librngdet.so replaces getrandom(2) with a deterministic stream and the stack disables RDRAND. Assume any process running under it has predictable randomness. Don't generate keys, session tokens, or nonces in such a process and treat them as secret. This isn't a flaw, determinism is the point, but it's easy to forget.

Captures contain request bytes. capture.log holds the raw bytes of every intercepted request, which means credentials and personal data if your workload carries them. Sessions live in ~/.onebarrier/ with normal user permissions. Treat them as sensitive and scrub before attaching one to a bug report.

setarch -R disables ASLR for the process it launches, which removes a standard exploitation mitigation. It's needed for byte-identical V8 randomness and you can skip it with --no-rng.

Reports that go beyond what's documented here are welcome, particularly a way to escape the intended scope of the interposition, or a case where a capture gets written somewhere it shouldn't.