Kronn is in active pre-1.0 development. Security fixes are applied to:
- The latest minor release: the minor line of the most recent release tag on GitHub, whatever its number.
- The
mainbranch.
Older minor versions do not receive backported fixes. Upgrade to the latest tag for security updates.
| Version | Supported |
|---|---|
Latest minor release (newest x.y.* tag) |
✅ |
| Any older minor release | ❌ |
Please do not open a public GitHub issue for security reports. Public issues are indexed by search engines and notify everyone watching the repo before a fix is available.
Instead, use GitHub's private security advisories:
This route is private, read by the maintainers only, and lets us coordinate a fix and a coordinated disclosure with you.
A useful report contains:
- Affected version (output of
kronn --versionor the release tag). - A minimal reproduction or steps to trigger the issue.
- The impact you observed (information disclosure, RCE, auth bypass, data loss, denial of service…).
- Your suggested severity (Low / Medium / High / Critical), if you have one.
- Whether the issue is already public anywhere.
- Acknowledgement within 5 business days.
- Initial assessment within 10 business days, including a target fix window.
- Fix + advisory published together once the patch ships. The advisory will credit you (unless you ask to remain anonymous).
We aim to ship security fixes in the next patch release. For Critical issues with a working exploit, we may ship an out-of-band patch.
In scope:
- The Kronn backend binary (Rust / Axum).
- The Kronn frontend (React, served by the backend).
- The desktop bundles (Tauri / packaged builds).
- The default MCP servers and templates shipped with Kronn.
Out of scope:
- Third-party AI providers (Anthropic, OpenAI, Google…) — report to the vendor.
- Third-party MCP servers the user installs themselves.
- Self-hosted misconfigurations where a user deliberately exposes the backend port without authentication on a public network.
- Findings from automated scanners without a working proof of concept.
Kronn is designed to run locally or on a trusted network. If you expose
it beyond localhost:
- Always enable authentication (
auth_enabled) and put it behind a reverse proxy with TLS. - Treat the encrypted vault directory (
~/.kronn) as a secret — back it up but don't store it on shared drives. - Audit the MCP servers you install: an MCP server runs with the same filesystem access as the agent it serves.
These are not vulnerabilities in Kronn itself, but they're the most common ways a Kronn deployment ends up exposed.
- Windows desktop build — project writes are check-then-write. On
macOS and Linux every write Kronn makes inside a project (docs, native
agent files, templates, seeds) goes through directory file descriptors
opened with
O_NOFOLLOW, so a folder swapped for a symlink while Kronn works is never written through. Windows has no such primitive in use: Kronn checks the path for links, then writes it. A folder swapped for a link in that window is not excluded. Creating a symlink on Windows needs elevated rights or developer mode, so Kronn accepts this rather than refusing to write there. A rooted primitive for Windows is tracked as KT-1609. Any other platform that is neither unix nor Windows refuses these writes. Details:docs/operations/fs-guard-rooted-writes.md.