Last updated: 2026-10-04

Mycelium is built and maintained by one person. This page says plainly what that means for reporting a security problem. There is a private way to reach me, I answer it, and I do not promise round-the-clock cover.

Mycelium runs on your own machine, inside your own coding agent. It has no servers, no accounts and collects nothing (see PRIVACY.md). So the problems worth reporting here are about what runs locally and how it reaches you:

• A hook or script in the plugin that can be made to do something harmful on the machine it runs on.
• The supply chain: anything that would let someone other than me publish code that installers then run.
• Prompt injection through files Mycelium reads into the agent's context, such as canvas or memory files.
• Mycelium leaking something it should not, for example a file or a secret written where it does not belong.

Not here: vulnerabilities in Claude Code, Codex, Cursor or opencode themselves (report those to their makers), and the general behaviour of the AI model.

Use "Report a vulnerability" on the Security tab of this repository. The report stays private between you and me until a fix is out.

Please do not open a public issue for a vulnerability.

• I acknowledge within 7 days. Best effort, since this is a solo project and not a staffed security team.
• I aim to fix or mitigate serious issues within 30 days, and will tell you if it will take longer and why.
• Only the latest release is supported. Fixes ship as a new release.
• Credit is yours if you want it in the advisory and the changelog.
• If you hear nothing within 14 days, open a public issue that says only "security report pending", with no details. That reaches me without disclosing anything.

There is no bug bounty.

Pin the plugin to a release tag rather than following main. Every release is tagged at the commit that set its version, and main only accepts signed commits that have passed CI.