This project is under active development. Only the latest commit on main is supported.
| Branch | Supported |
|---|---|
main |
✅ |
| Any tagged release | ✅ |
| Older branches | ❌ |
Do NOT open a public GitHub issue for security vulnerabilities.
If you discover a security vulnerability in antimalware-core, please report it privately:
- Open a GitHub Security Advisory on this repository.
- Include a clear description of the vulnerability, the affected component, and steps to reproduce.
- Allow up to 72 hours for an initial response.
The security model of this project is formally documented in two files:
- AGENTS.md — Interaction rules, dependency policy, and hard limits.
- BRAIN.md — Architectural Decision Records (ADR log) and known threat mitigations.
The Named Pipe server at \\.\pipe\antimalware_core_ipc enforces a layered validation chain before processing any message:
- T1 — Anti-squatting:
FILE_FLAG_FIRST_PIPE_INSTANCEensures only the service can hold the pipe. - T2 — DACL: Restrictive ACL allows only
SYSTEMandINTERACTIVEprincipals. - T3 — Remote block:
PIPE_REJECT_REMOTE_CLIENTSprevents SMB-based access. - T4 — Size limit: 8 KB maximum payload, validated before memory allocation.
- T5 — Process identity: Client's executable path is verified against a hardcoded whitelist via
GetNamedPipeClientProcessId→QueryFullProcessImageNameW.
- Zero unaudited external crates. Every dependency is listed in
BRAIN.mdwith maintainer, license, and surface-of-attack evaluation. - Preference hierarchy: Win32 native API → Microsoft crates (
windows-rs) → audited community crates. cargo auditmust pass before any dependency change is merged.
Every unsafe block must have a // SAFETY: comment that explicitly justifies why the invariants upheld by the compiler are maintained manually. This is enforced during code review.
Files in the quarantine vault (C:\ProgramData\AntimalwareCore\Quarantine\) use a protected DACL (inheritance disabled) with:
ALLOW NT AUTHORITY\SYSTEM: GENERIC_ALLALLOW BUILTIN\Administrators: GENERIC_ALL- All other principals: implicit deny
A DENY Everyone ACE is intentionally avoided. Everyone (S-1-1-0) includes SYSTEM; placing a DENY ACE for Everyone after the ALLOW SYSTEM ACE would still deny SYSTEM, because Windows evaluates ACEs sequentially and an ACCESS_DENIED_ACE for a matching SID immediately terminates access checks.
The following are not security vulnerabilities in the context of this project:
- Behaviors that require the attacker to already have
SYSTEMorAdministratorsprivileges. - Denial of service via resource exhaustion on the scanning host (the scanner is designed to be cancelled).
- Issues in third-party dependencies (report to the respective upstream project; we will update our dependency).