Do not open a public issue for a security problem.
Report it privately through GitHub's advisory flow:
https://github.com/Epichlo/Public-Intelligence/security/advisories/new
If that is not available to you, email atharvdesh200713@gmail.com with
SECURITY in the subject.
Please include what you can — affected file and line, the version or commit, what an attacker gains, and a way to reproduce it. A rough report is more useful than none; do not sit on something because it is not written up.
This is a pre-alpha project maintained by one person. There is no security team and no on-call. Setting expectations honestly rather than copying a template:
| Acknowledgement | within 7 days |
| Assessment | within 30 days |
| Fix or a stated decision not to fix | best effort, no guarantee |
You will be credited unless you ask not to be. There is no bounty programme.
In scope — anything in packages/scheduler, packages/node,
packages/website, the installers (install.sh, install.ps1), and the gate
scripts under scripts/.
Out of scope — docs/historical/: superseded documents, kept as history, not
shipped code.
Code for features cut from v1 (split inference, Raft consensus, KV-cache
checkpointing) is currently still inside packages/ and is in scope while it
sits there. ROADMAP C2 moves it to experimental/; once it has moved, reaching it
from a running Scheduler or node is itself the vulnerability, because it is
supposed to be unreachable.
This project keeps its known weaknesses in the open rather than in a private tracker.
VERIFY.md step 3 lists the auth bypasses and default credentials present in the
tree, with file:line. They are pre-existing, tracked in ROADMAP.md, and not
things you need to report.
Anything not on that list is worth reporting.
There are no releases yet. main is the only supported version. When there are
releases, this table gets filled in rather than deleted.
Running a node or a coordinator carries risks that are not vulnerabilities in the
code — you become the egress point for generated content, prompts you serve are other
people's data, and Zenoh mesh links are plaintext unless you configure TLS. Those are
documented in docs/OPERATING.md, which you should read before
running this for anyone but yourself.