Skip to content

Generated at build time from SECURITY.md

Security

agemon is maintained on a best-effort basis. This policy describes what you can expect. It does not set deadlines.

Only the latest release gets security fixes. Fixes are not backported to older minors.

Please report privately. Do not open a public issue with details.

  • Preferred: GitHub private vulnerability reporting, under the repository’s Security → Report a vulnerability.
  • If that is unavailable: open a public issue that says only that you have a security report and would like a private contact. Leave out every detail.

Include the agemon version (agemon --version), your OS, steps to reproduce, and the impact you expect.

These are best-effort targets, not guarantees:

  • acknowledgement within about 7 days;
  • an assessment and a rough plan once the report has been reproduced;
  • coordinated disclosure: details are published together with the fix, and you are credited unless you prefer not to be.

There is no bug bounty.

In scope:

  • the agemon CLI (this repository’s src/ and bin/);
  • install.sh and the self-update path;
  • release artifacts published from this repository;
  • the vendored skill copies under assets/skills/. Please also report problems that come from the original skill to its upstream project. Each skill’s NOTICE.md names it.

Out of scope (report these upstream):

  • third-party tools that agemon installs or runs, such as the skills CLI, code-review-graph, agnix, pipx and npm;
  • the content of skills that agemon installs from remote sources and does not vendor. How agemon fetches, pins or verifies them is in scope.

agemon treats repository contents (agemon.toml, .agemon/ state and plan files, existing instruction files) and network responses as untrusted. It trusts the local user, meaning the environment variables and CLI arguments of the person running it. The clearest reports show untrusted input crossing that boundary, for example a cloned repository’s plan file that makes agemon write outside the repository. Dangerous handling of the local user’s own input, such as a destructive default, is still worth reporting.