We currently support the latest main branch. Security fixes land there first
and are backported to the most recent tagged release when feasible.
If you believe you've found a security issue — particularly in the areas of:
- Secret handling (API keys, tokens)
- Permission misuse (notification listener, accessibility service, SMS, camera)
- Arbitrary code execution via SKILL.md or tool payloads
- Network calls exposing user data
- Local storage that should be encrypted but isn't
Please do NOT open a public GitHub issue.
Instead, report privately via GitHub's Security Advisory flow or contact the maintainer directly.
We'll acknowledge within 72 hours and aim to ship a fix within 14 days for critical issues (definition: RCE, data exfiltration, arbitrary permission escalation).
OpenDash is designed to be on-device first. The threat model assumes:
- Untrusted input via voice / STT — must not produce unsafe tool calls
- Tool boundaries — each tool validates its own arguments; the LLM is never trusted to pre-validate
- Skills are descriptive, not executable — SKILL.md is instructions to the LLM only, it cannot run arbitrary code
- Network calls are minimized; weather / search / news use public APIs with no authentication
- Secrets stored in
SecurePreferences(EncryptedSharedPreferences), never in logs
- Android Accessibility Service exposes screen content to the agent. Users must grant this explicitly, but compromise of the agent would compromise everything visible on screen.
- Notification listener similarly exposes notification content.
- SMS send tool can send on user's behalf; always confirm with the user before calling.