Email security@canonry.ai. Do not open a public issue, pull request or discussion for a suspected vulnerability.
Include what you can of:
- the affected version (
cnry --version) and how Canonry runs (localcanonry serve, Docker, Cloud Run or the MCP gateway); - the component (API route, CLI command, MCP tool, dashboard page or integration);
- steps to reproduce, or a proof of concept against your own install;
- the impact you expect, such as data exposure, privilege escalation or server-side request forgery.
Do not send real API keys, OAuth tokens, customer data or production database files. Redact them, or reproduce against a fresh install with test credentials.
Fixes ship in the next release of @canonry/canonry on npm. Only the latest
published version is supported, so upgrade before reporting and confirm the
issue still reproduces.
In scope: the @canonry/canonry package (CLI, server, MCP adapter, built-in
Aero agent and dashboard), apps/api, apps/mcp-gateway, and the agent
plugins and skills in this repository.
Out of scope:
- vulnerabilities in third-party services Canonry connects to (AI engines, Google, Bing, Cloudflare, Vercel, WordPress, ad platforms); report those to the vendor;
- findings that need an attacker who already holds a full-instance API key or shell access to the host;
- running one instance for unrelated tenants. Canonry is single-tenant by design: one trust boundary per instance (see "Deployment Posture" in AGENTS.md).
Good-faith research that follows this policy, stays within your own installation and accounts, and avoids privacy violations, data destruction and service disruption is welcome. Give us a reasonable chance to fix an issue before you disclose it.