Quasnooze runs shell commands on your machine for a living, so a real vulnerability here is worth reporting privately rather than in an issue.

Use GitHub's private reporting: Security → Report a vulnerability. That opens a draft advisory only you and the maintainer can see.

If you would rather not use GitHub, open an issue that says only "security report, please contact me" with a way to reach you — no details — and a channel will be arranged.

Please include what you did, what happened, and the version or commit. A probe script is better than a sentence.

What to expect: an acknowledgement within a few days, an honest assessment of severity, and credit in the advisory once a fix ships. If the answer is "that is the documented trade-off", the relevant section of docs/AUDIT.md will be pointed at — the residual-risk register there is deliberately public.

Read docs/AUDIT.md before reporting. It documents six real defects that were found by attacking the product, their fixes, and the risks that remain on purpose:

• the command screen is a denylist, and obfuscated commands still reach the approval gate;
• there is no OS-level sandbox — the approval gate and workspace confinement are the controls;
• prompt injection through tool output is possible, which is why mutating tools are gated;
• the WhatsApp channel is unofficial, against WhatsApp's terms, and off by default.

Reporting one of those is still welcome if you can show it is worse than described — that is exactly how the six got fixed.

The latest commit on main. See docs/SUPPORT.md for which Node versions are supported and until when.

No CVE will be requested for a vulnerability that requires an attacker to already have local code execution, or that needs the user to approve the destructive action in the approval gate. That is the threat model, not an oversight.