Skip to content

Security: tpikachu/BrainCue

SECURITY.md

Security Policy

BrainCue is a local-first desktop app whose security model is small and strict: the API key lives only in the main process (encrypted at rest via the OS keychain), the renderer is untrusted and reaches main only through a typed, zod-validated IPC bridge, and app windows are excluded from screen capture in Privacy Mode. Anything that weakens one of those invariants is a security bug, even if nothing "crashes".

Reporting a vulnerability

Please do not open a public issue for security problems.

Report privately via GitHub security advisories. Include what you found, how to reproduce it, and what an attacker gains.

You can expect an acknowledgement within 72 hours and a status update at least weekly until resolution. Confirmed reports are credited in the release notes unless you ask otherwise.

Scope

In scope — anything violating the invariants in docs/07-API-KEY-SECURITY.md and CONTRIBUTING.md:

  • API key reaching the renderer, logs, IPC payloads, or disk in plaintext
  • IPC handlers reachable with unvalidated/forged payloads, or renderer code gaining main-process capabilities
  • App content becoming visible to screen capture while Privacy Mode is on (including via native OS surfaces: dialogs, tooltips, popups)
  • Unapproved memory content reaching prompts, or local data exfiltration
  • Supply-chain issues in our build/release pipeline

Out of scope: vulnerabilities requiring an already-compromised machine, social engineering, and issues in third-party services (report those upstream).

Fixing security bugs

If you want to submit the fix as well: report privately first, then open the PR once we've agreed on disclosure. Security fixes follow the normal contribution gate (see CONTRIBUTING.md) plus a maintainer security review.

There aren't any published security advisories