Do not open a public issue for a security problem.
Email problems@mwater.co with a description, the affected component, and steps to reproduce. If you have a proof of concept, attach it. We acknowledge reports within 5 working days and aim to ship a fix or a documented mitigation within 90 days, sooner for anything actively exploitable.
You may also report privately through GitHub Security Advisories on this repository.
Security fixes land on main and in the latest release. Deployments should
run the latest release and update with deploy/update.sh.
Deployments are self-hosted, so the operator owns the host, the network, and the Postgres instance. Reports about the code in this repository are in scope, in particular:
- Sandbox escape. Integration code runs in rootless podman + gVisor with an egress policy that permits public destinations only. Anything that reaches the host, another container, or a private network address from inside a job is the highest-severity class here.
- Agent-driven code execution outside the sandbox. The designer agent writes and runs integration code. Any path where agent output executes in the server process rather than in a job container.
- SQL execution boundaries. Dashboards run frozen, parameterized SQL;
Explore (the query assistant) runs SQL the model writes for each question.
Both run as the read-only
sql_enginerole with statement timeouts and row caps. Escapes from that role, the parameterization, or those limits are in scope. - Authentication and authorization. Session handling, personal access tokens, and the admin/editor/viewer role checks.
- Secret handling. Integration secrets are AES-256-GCM encrypted at rest
under
INTEGRATOR_SECRET_KEY, and their values are masked in stored run output. Disclosure of plaintext secrets to logs, the API, job containers that did not declare them, or the agent transcript. - Unauthorized data exposure through the query API, raster tiles, or COG downloads.
Out of scope:
- Findings that require an operator to have already misconfigured the
deployment against the documented guidance, such as running the server as
root, exposing Postgres publicly, or reusing
INTEGRATOR_SECRET_KEYacross deployments. - Denial of service from a legitimately authorized user's own job hitting its configured resource limits.
- Vulnerabilities in third-party dependencies with no exploitable path through this code. Report those upstream; tell us if you want the pin bumped.
We prefer coordinated disclosure. Tell us when you intend to publish and we will work to have a fix out first. We credit reporters in the advisory unless you ask us not to.