# Security Policy We take the security of ReviewBench seriously. This document explains how to report vulnerabilities and what is — and isn't — in scope. ## Supported versions ReviewBench is pre-1.0 and ships from `main`. Security fixes are applied to `main` only; there are no maintained back-branches. | Version | Supported | | ---------- | ------------------ | | `main` | :white_check_mark: | | < `0.1.0` | :x: | ## Reporting a vulnerability **Please do not open public GitHub issues or pull requests for security problems.** Public reports give attackers a head start before we can ship a fix. Instead, use **GitHub Private Vulnerability Reporting**: 1. Go to 2. Fill in the advisory form with: - A clear description of the issue and its impact - Steps to reproduce (or a proof of concept) - The affected commit / version - Any suggested mitigations you've identified 3. Submit. The maintainers will be notified privately. You don't need a CVE in advance — we'll request one if the issue warrants it. ### What to expect - **Acknowledgement:** within **7 days** of your report. - **Triage:** we'll confirm whether we can reproduce the issue and classify severity (low / medium / high / critical). - **Fix timeline:** depends on severity. Critical issues get immediate attention; lower-severity issues are scheduled into the normal release flow. - **Coordinated disclosure:** we'll work with you on a disclosure timeline. By default we ask for 90 days before public disclosure, but we're flexible. - **Credit:** if you'd like to be credited, we'll name you in the advisory and release notes. Anonymous reports are also welcome. ## Scope ### In scope - **Code in this repository** — the CLI, fetchers, parsers, processors, and prompts under `src/` and `cli.py`. - **Database integrity** — ReviewBench's Cloud SQL Postgres instance is publicly readable by design (see "Data Accessibility" in [`README.md`](README.md)). However, the following **are** in scope and worth reporting: - The `user` role having more privileges than read-only (`INSERT`, `CREATEROLE`, `CREATEDB`, etc.) - Any path that lets an unauthenticated user mutate `papers`, `comments`, `claims`, `assessments`, or `issues` - Privilege-escalation paths from the public `user` account to higher-privileged roles - **Supply-chain risks** — malicious or compromised dependencies declared in `pyproject.toml`, GitHub Actions used in `.github/workflows/`, or referenced LLM API endpoints. - **Secrets accidentally committed** to the repo. ### Out of scope - The fact that the public Cloud SQL credentials are documented in `README.md` — that is **intentional** read-only access for reproducibility, not a misconfiguration. (See above re: privileges that exceed read-only — those *are* in scope.) - Denial-of-service attacks against the public database. Please don't try to exhaust it. - Vulnerabilities in third-party services we link to (Google Cloud SQL itself, GitHub, Gemini API, OpenAI API). Report those upstream. - Social-engineering attacks against maintainers or contributors. - Issues requiring physical access to a maintainer's machine. ## Safe-harbor We will not pursue legal action against researchers who: - Make a good-faith effort to follow this policy - Report vulnerabilities promptly via the channel above - Avoid privacy violations, data destruction, and service degradation - Give us reasonable time to respond and fix the issue before public disclosure Thank you for helping keep ReviewBench and its users safe.