Email support@vyttle.com, or use GitHub private vulnerability reporting.

Please don't open a public issue for a security bug.

We'll acknowledge within 3 business days and aim to give you a fix timeline within 10 business days. We'll credit you in the advisory unless you'd rather we didn't.

Reviso reads untrusted input by design. A code review tool ingests diffs, commit messages, branch names, code comments, and repository configuration — all of it potentially attacker-controlled if the repo takes outside contributions. This class of tool has already produced a real incident: a remote code execution in CodeRabbit that reached write access on roughly a million repositories.

We take that seriously and you should hold us to it.

• Prompt injection through diffs, commit messages, branch names, code comments, or .reviso/ files — anything that makes Reviso suppress a real finding, fabricate one, or act outside its report-only contract
• Any path by which Reviso modifies files — it is report-only, so any write is a vulnerability by definition
• Any outbound network write beyond the feedback contract in docs/feedback.md — the only permitted send is a feedback report the user explicitly approved; any other destination, or a payload carrying content the user never saw, is a vulnerability
• Command injection or path traversal in the detectors or context assembly
• Exfiltration of repository contents, environment variables, or credentials
• Escaping the repository sandbox

• Reviso missing a bug in your code (that's a missed finding)
• False positives (file a false positive report)
• Vulnerabilities in Claude Code itself — report those to Anthropic
• Issues requiring an already-compromised local machine

Pre-1.0: the latest release only.