RhubarbTart builds and manages hardened, provenance-verified VMs for authorized security work. Security is the product, so a flaw in the build/verify/harden pipeline, the provenance chain, or how secrets and clone state are handled is a serious bug. Reports are welcome.

Please do not open a public issue for a security problem.

Report it through GitHub's private vulnerability reporting: the repository's Security tab → Report a vulnerability, or this form. Only the maintainer can see the report.

Please include:

• What the issue is and which component (a tools/ module, a Packer template, a guest/ script, the rhubarbtart CLI, the provenance/verification logic, secret handling, or the network model).
• How to reproduce it, and what an attacker gains.
• The commit or version you're looking at, and your host OS.

We aim to acknowledge a report quickly and to fix confirmed issues before any disclosure. This is an early-stage project maintained on a best-effort basis; there is no paid bounty.

The things this project promises, and therefore wants to hear about if they break:

• Provenance bypass: a way to get an unpinned, unverified, or tampered input (OS image, package, or host tool) into a build without the hash/signature checks catching it, on the host or inside the guest.
• Verification weaknesses: flaws in the GPG, apt-repo, Tailscale distsign/Ed25519, NAR-hash, or macOS codesign/notarization checks (tools/rhubarb/).
• Hardening escapes: a built image that ships with a default/weak password, auto-login, passwordless sudo, password SSH, a disabled firewall, or shared identity, without the seal step or scripts/smoke-test.sh catching it.
• Secret leaks: a path where a VM password, VPN token, or enrollment secret reaches an image, a log, a committed file, argv, or another clone. Secrets should only move keychain → subprocess stdin → shred.
• Clone-record integrity: a way to make the rhubarbtart CLI act on the wrong VM or keychain entry via a crafted record (see the record model in tools/rhubarb/clones.py).
• Isolation gaps: a clone reaching another clone without a declared engagement link, or a lab target reaching the internet or the host (see Host-side boundaries).
• Evidence and vault integrity: a way to change, drop or reorder evidence journal entries, or alter a sealed vault, without rhubarbtart evidence verify or rhubarbtart vault verify noticing.
• Control-plane and agent boundaries: a way to reach the control-plane socket without file permission to it, to make rbt-range act on a clone other than its assigned one, or to run a tiered command without a single-use approval.

The Trust model and Security posture pages, and the threat model in docs/PLAN.md, describe the intended guarantees.

• Misuse by an authorized operator. This is a tool for permitted testing; a trusted operator pointing it at systems they're allowed to test is by design. Using it against systems you are not authorized to test is misuse, not a vulnerability in this project.
• Vulnerabilities in the target systems an operator tests with these VMs.
• Vulnerabilities in upstream software (Tart, Packer, the guest OSes, the bundled tools): report those to their maintainers. We will, however, update pins and re-verify in response.
• A fully root-compromised host. The host is the trust anchor; keep it hardened.

RhubarbTart is for authorized security research, scoped penetration testing, and CTFs. Reports, issues, and contributions should assume that authorized-use context.