Title: mdr-kernel security and privacy-bug reporting Type: Security Policy Context: How to report a vulnerability or a bug that could leak private memory content, privately and before public disclosure.

This pattern's core promise is that bin/setup-private.ps1 and bin/remote-privacy.ps1 refuse to let your memory land anywhere but a private destination you approved. A bug that breaks that promise — a way to push private content to a public or unintended remote, bypass a hook, or leak a secret into a tracked file — is a privacy bug and should be reported privately, not as a public issue.

Use GitHub's private vulnerability reporting: this repository's Security tab → Report a vulnerability. That opens a private draft advisory only you and the maintainer can see until a fix ships. Do not open a public issue or PR that demonstrates the bypass.

If you cannot use GitHub Security Advisories, open a regular issue that says only "privacy/security report, details sent separately" and asks for a contact method — do not include exploit details or example private-repo URLs in the public issue itself.

• Setup, doctor, close-session, or the pre-push hook accepting a public repo, a fork, or an unverifiable-visibility remote as an approved private destination.
• A code path that pushes before remote-privacy.ps1 has verified the destination, or that skips verification on retry/recovery (sync-private.ps1, -OfflineCheckpoint recovery).
• A secret-shaped string (API key, token, credential) that the pre-commit scan or bin/secret-audit.ps1 fails to catch.
• The exporter (bin/export-brain-template.ps1) writing outside its intended output root, or shipping studio-private content into the public template.

Ordinary correctness bugs (a doctor check reports the wrong wording, a test is flaky, a doc dangles) belong in a normal public issue — see CONTRIBUTING.md.

This is a pattern of scripts, not a hosted service: there is no server-side attack surface, no telemetry, and no account system. The threat model is a cooperating multi-agent workflow protecting against accidental leaks (see the "Known Limits" section in README.md), not a defense against a malicious actor with write access to your machine or repo.

Before opening a PR that touches privacy/security-relevant code (bin/remote-privacy.ps1, bin/setup-private.ps1, bin/secret-audit.ps1, .githooks/*, bin/export-brain-template.ps1):

• bin/brain-doctor.ps1 is green.
• tests/remote-privacy-unit.ps1 passes (hermetic; no live network needed).
• If the change affects live GitHub visibility detection, tests/remote-privacy-live.ps1 was run manually against a real private/public repo pair and the result is in the PR description.
• The change was tested in a throwaway clone with origin removed, not in a clone of the public template or any launch/production repo.

Directory Path: ./ Next Sequential File: CONTRIBUTING.md Related Workflows: CONTRIBUTING.md, README.md, bin/remote-privacy.ps1, tests/remote-privacy-unit.ps1 AI Execution Prompts: protocol/CONTRACT.md