Skip to content
Back to skills

Ring 1 Bootkit

ASecurity

Use when reversing or detecting a Ring-1.io-class bootkit cheat: bootmgfw implant, Hyper-V SLAT/MTF hooks, cloned game page tables, EPT-hidden inject, or the Aftermath Labs ring-1.io corpus and 2026 writeup.

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 17, 2026
toolsgogit

Works with

  • cli

Security analysis

A100/100

Pro scans all 3 files and shows the line behind each finding

Scanned October 17, 2026

npx -y skills add NetVar1337/agent-skill-canon --skill ring-1-bootkit --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Ring 1 Bootkit?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Ring 1 Bootkit
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/netvar1337-ring-1-bootkit/badge)](https://www.skillsdirectory.com/skills/netvar1337-ring-1-bootkit)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: ring-1-bootkit
description: "Use when reversing or detecting a Ring-1.io-class bootkit cheat: bootmgfw implant, Hyper-V SLAT/MTF hooks, cloned game page tables, EPT-hidden inject, or the Aftermath Labs ring-1.io corpus and 2026 writeup."
version: 1.0.0
license: MIT
metadata:
  package: local-operator-skills
  category: game
  author: Admin
  source: https://github.com/aftermathlabs/ring-1.io
  upstream: https://aftermathlabs.net/blog/04/02/2026/
  triggers:
    - ring-1
    - ring-1.io
    - bootloader implant
    - aftermath ring
    - bootmgfw cheat
---

# Ring-1.io bootkit analysis

Aftermath Labs (xeroxz / noahware / Eggsy / AVX) dumped and deobfuscated the Ring-1.io commercial cheat stack. PRIMARY here is *analysis and detection of that architecture*, not a rebuild.

Corpus: `https://github.com/aftermathlabs/ring-1.io` (also `backengineering/ring-1.io`)  
Writeup: `https://aftermathlabs.net/blog/04/02/2026/`  
Deobf engine they used: in-house BLARE2 (`https://back.engineering/blare/`) — not public. Use `virtualization-deobfuscation` + `llvm-lift-deobfuscation` as the stand-in.

## Artifact map

| File | Role |
|---|---|
| `bootmgfw.bin` | Tampered Windows boot manager |
| `bootloader-implant-obfuscated.bin` | On-disk implant |
| `bootloader-implant-deobfuscated.bin` + `.i64` | Decompressed/decrypted implant + IDA DB |
| `loader-deobfuscated.bin` | Usermode/chain loader |
| `*-cheat.bin` | Injected game modules (Apex, BattleBit, COD6, EFT, Grayzone, R6) |

Hash every blob before opening. Work offline.

## Workflow

1. **Confirm the chain, do not start in the game DLL.** Order on the writeup: Loader → bootloader implant → map into Hyper-V → SLAT + MTF → GPA redirect → VMEXIT hooks → implant comms → clone target CR3 → insert PTEs → hypercall to load tables → hide tables + contents via EPT → usermode hooks.
   - Done when a one-page chain diagram names the file that implements each hop.

2. **Bootloader implant.** Open `bootloader-implant-deobfuscated.i64` (IDA 9.4). Recover: how `bootmgfw.ImgArchStartBootApplication` / `BlImgStartBootApplication` is hooked, where winload/hvloader hooks land, how the Hyper-V image is found. This is Voyager’s hookchain (`hypervisor-memory-introspection`).
   - Done when hook sites have names + RVAs in the IDB.

3. **Hyper-V payload.** Recover SLAT setup, EPT-violation handler, MTF single-step, VMEXIT table. GPA redirect = the implant’s physical R/W. Compare to Voyager (append-to-hv image) vs hyper-reV (standalone attachment, timestamps restored).
   - Done when EPT-violation and MTF paths are commented and the comms ABI (CPUID/VMCALL/shared page) is written down.

4. **Process injection.** Clone the game’s page tables, insert malicious PTEs, ask the HV to switch/load them, hide the new tables and the payload pages with EPT execute/read split. Game module `*-cheat.bin` is the payload, not the persistence.
   - Done when one cheat.bin is mapped, its imports/entry named, and the PTE-clone step is cited to a function in the implant.

5. **Detections (write these as a lab checklist).** Aftermath section “Possible Detections”:
   - `bootmgfw.efi` hash / Authenticode vs known-good
   - Measured-boot / TPM PCR for the replaced boot manager (`tpm-attestation-research`)
   - Hyper-V image size / allocation-after-hvix64 shift (Voyager-class)
   - EPT split on game or ntos pages → `ept-hook-detection`
   - Extra CR3 / cloned PML4 not backing a real process
   - Unexpected VMEXIT rate on the game’s `.text`
   - Done when each detector has a command or probe and a pass/fail on the corpus (static) or a lab VM.

6. **Do not treat vmp2 handler tables as the deobf plan.** Aftermath retired that. Themida (2026-05-09) and Tencent VM (2026-07-31) posts: incremental lift, const/memory model, DCE, branch fold, lower 1:1. Route the `*-cheat.bin` protectors through `virtualization-deobfuscation`.
   - Done when the protector on each cheat.bin is labeled and the recovery product is chosen.

## Pair with

- `hypervisor-memory-introspection` — Voyager / hyper-reV mechanics
- `ept-hook-detection` — guest-side SLAT-hook probes
- `secure-boot-uefi-research` / `analyzing-uefi-bootkit-persistence`
- `virtualization-deobfuscation` / `llvm-lift-deobfuscation`
- `anti-cheat-bypass` / `eac-kernel-driver-re` — what Ring-1 is hiding from

## Verification

- [ ] All corpus hashes recorded
- [ ] Boot hook RVAs named in the IDB
- [ ] SLAT/MTF/comms ABI written
- [ ] One game module’s entry + inject primitive identified
- [ ] Detection checklist has an observable per row
- [ ] No live multiplayer client was instrumented

Files in this skill

  • SKILL.md4.5 KB
  • agents/openai.yaml43 B
  • references/aftermath-blog.md2.4 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.