Hodari ships from main. Security fixes go to main only. There is no
long-term support branch.
Do not open a public issue.
Report a vulnerability in one of these two ways:
- GitHub private reporting (preferred). Go to the Security tab of this repository and choose Report a vulnerability. This opens a private advisory that only the maintainers can read.
- Email. Send the report to
pacymugisho@gmail.comwith the subjectSECURITY: <short summary>.
Please include:
- The affected component, such as the agent runtime, the Next.js client, or an MCP sidecar.
- The steps that reproduce the problem.
- What an attacker gains.
- Your suggested fix, if you have one.
| Step | Target time |
|---|---|
| We confirm we received your report | 3 working days |
| We give you our first assessment | 10 working days |
| We release a fix for a confirmed high severity problem | 30 days |
We credit you in the advisory unless you ask us not to.
In scope
- The agent runtime in
agents/. - The Next.js client in
client/, including the API route handlers. - The MCP sidecars in
agents/community_mcp/andinfra/mongodb-mcp/. - The Cloud Run service templates in
infra/. - Prompt injection that makes an agent leak another user's data or call a tool it must not call.
Out of scope
- Vulnerabilities in Google Maps Platform, Vertex AI, MongoDB Atlas, or Stripe. Report those to the vendor.
- Denial of service through plain request volume.
- Missing security headers with no shown impact.
- Results from an automatic scanner with no proof of exploitation.
Hodari holds real user data and calls paid APIs. If you run your own instance:
- Keep every secret in a secret manager. The templates in
infra/use Google Secret Manager references, not plain values. - Rotate your Google Maps key, your Gemini key, and your MongoDB password if you ever commit them, even for one push.
- Restrict your Google Maps browser key by HTTP referrer and your server key by IP address.
- Do not expose the agent service to the public internet. Give the frontend
service account the
run.invokerrole and removeallUsers. - Read docs/SECURITY_HARDENING.md for the full checklist.