Skip to content

Security: dills122/session-chat

SECURITY.md

Security policy

Session Chat 2.0 is a protocol laboratory. It has no production release and is not suitable for protecting real conversations. The current code proves only the properties listed as implemented in docs/INDEPENDENT_AUDIT_BRIEF.md.

Supported scope

Security reports should target the current master branch or an open pull request. The retired v1 implementation is preserved under the legacy-v1 tag for historical evidence and is not supported.

Especially useful reports include:

  • a bypass of an implemented parser, size bound, signature check, or invitation lifecycle invariant;
  • a contradiction between code, tests, the threat model, and an accepted ADR;
  • secret disclosure through an error, log, fixture, serialized object, or CI artifact;
  • a dependency or build-workflow compromise path; or
  • evidence that a proposed security contract cannot be implemented safely.

Design omissions that are already marked unimplemented or deferred are welcome as architecture feedback, but are not vulnerabilities in a deployed product.

Reporting

Use GitHub's private vulnerability-reporting form under Security → Advisories → Report a vulnerability when it is available. Do not publish exploit details in a public issue. If the private form is unavailable, contact the repository owner through GitHub to arrange a private channel and share only a non-sensitive summary until that channel is established.

Include the inspected commit, affected files, preconditions, impact, a minimal reproduction, and whether the issue applies to implemented code or a future contract. Never include real credentials, invitation capabilities, mailbox rights, private keys, or conversation data.

Receipt and remediation timelines are not yet promised. A production release is blocked on establishing an operated private reporting channel and a published response policy.

There aren't any published security advisories