contextd holds context you write and, in server mode, mediates access to it for a team. A bug in it can expose files, tokens or an audit trail. Reports are welcome and taken seriously.
Report privately, through GitHub's advisory form:
https://github.com/orkcom-tech/contextverse/security/advisories/new
That opens a private thread with the maintainers. Please do not open a public issue, and please do not post proof-of-concept code publicly before a fix is available.
Include what you would want if you were fixing it: the version (contextd version), the mode (solo, client or server), what an attacker can reach, and the smallest reproduction you have.
This is a small project, so the honest version rather than a service-level promise: you will get a human reply, and if the report is valid you will be credited in the advisory and the changelog unless you would rather not be.
In scope — anything that lets someone read or change context they should not, escape a path ACL, forge or break the audit chain, extract a token, drive the local console from another origin, or make a client trust a server it should not.
Out of scope — findings that require an attacker who already has your shell or your filesystem. contextd protects a space from the network and from other users; it does not defend against someone who is already you on your own machine.
Two known and deliberate properties, so you need not report them:
- The local web console (
contextd ui) binds to loopback and grants full read/write to the space it serves. That is what it is for, and why it is off by default and prints its own warning. - A model handed your context can ignore it.
contextdguarantees delivery, not obedience.