[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7DASW3W7665Y0KT2KBNRTKV":3,"skill-same:01M7DASW3W7665Y0KT2KBNRTKV":47},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":28,"repositoryHtmlUrl":29,"repositoryHomepageUrl":6,"ownerHtmlUrl":30,"ownerBlogUrl":6,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M7DASW3W7665Y0KT2KBNRTKV","game-hacking-techniques",null,"Guide for game-hacking technique taxonomy and threat modeling relevant to game security. Use this skill when researching memory access, code injection, overlays, input simulation, engine-specific attack surfaces, or how modern anti-cheat systems constrain user-mode, kernel-mode, hypervisor, and DMA-based cheat implementations.","alicewe1\u002Falice_skill","alicewe1","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F102134151?v=4","01M7D9STCM8HAESM958JP0989K","Alice_we","01M7D9STDR22MTD9FTTWPAZNV8",64,"GPL-3.0","https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002F_modules\u002Fgame-hacking\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: game-hacking-techniques\ndescription: Guide for game-hacking technique taxonomy and threat modeling relevant to game security. Use this skill when researching memory access, code injection, overlays, input simulation, engine-specific attack surfaces, or how modern anti-cheat systems constrain user-mode, kernel-mode, hypervisor, and DMA-based cheat implementations.\nx-alice-class: game\n---\n\n# Game Hacking Techniques\n\n## Overview\n\nThis skill covers game-hacking techniques documented in the awesome-game-security collection, with emphasis on how cheats move from user mode to kernel mode, hypervisors, and DMA when defenders raise the bar. It is best used to understand the offensive side of the threat model that anti-cheat systems are designed to detect.\n\nTreat implementations, performance numbers, stealth rankings, and detection\nclaims as versioned threat-model examples rather than guarantees. Use\n[`research-rigor`](..\u002Fresearch-rigor\u002FSKILL.md) when converting them into a\nfactual claim or defensive decision.\n\n## README Coverage\n\n- `Cheat > Debugging`\n- `Cheat > Packet Sniffer&Filter`\n- `Cheat > Packet Capture&Parse`\n- `Cheat > SpeedHack`\n- `Cheat > Injection:Windows`\n- `Cheat > Injection:Linux`\n- `Cheat > Injection:Android`\n- `Cheat > Injection:IOS`\n- `Cheat > Injection:PlayStation`\n- `Cheat > DLL Hijack`\n- `Cheat > Hook`\n- `Cheat > Anti Signature Scanning`\n- `Cheat > RPM`\n- `Cheat > DMA`\n- `Cheat > W2S`\n- `Cheat > Overlay`\n- `Cheat > Render\u002FDraw`\n- `Cheat > UI Interface`\n- `Cheat > Vulnerable Driver`\n- `Cheat > Driver Communication`\n- `Cheat > EFI Driver`\n- `Cheat > QEMU\u002FKVM\u002FPVE\u002FVBOX`\n- `Cheat > Wine`\n- `Cheat > Anti Screenshot`\n- `Cheat > Spoof Stack`\n- `Cheat > Hide`\n- `Cheat > Anti Forensics`\n- `Cheat > Triggerbot & Aimbot`\n- `Cheat > WallHack`\n- `Cheat > HWID`\n- `Cheat > Bypass Page Protection`\n- `Cheat > SDK CodeGen`\n- `Cheat > Game Engine Explorer:*`\n- `Cheat > Explore UWP`\n- `Cheat > Explore AntiCheat System:*`\n- `Cheat > Game:*`\n- `Cheat > Launcher Abuser`\n- `Cheat > Linux Kernel Explorer`\n- `Cheat > Cheat Engine Plugins`\n- `Some Tricks > Windows Ring0`\n- `Some Tricks > Windows Ring3`\n- `Some Tricks > Linux`\n- `Some Tricks > Android`\n\n## Escalation Model\n\n### User-Mode\n- Read and write process memory\n- Inject DLLs or shellcode\n- Hook graphics or input APIs\n\n### Kernel-Mode\n- Use signed or vulnerable drivers for direct memory access\n- Bypass handle-based protections and inspect protected processes\n- Interact with callbacks, page tables, or kernel objects directly\n\n### Below the OS\n- Virtualize the system with a hypervisor\n- Read memory through PCIe DMA hardware\n- Move logic to external devices or secondary machines\n\n## Core Concepts\n\n### Memory Manipulation\n- Read Process Memory (RPM)\n- Write Process Memory (WPM)\n- Pattern scanning\n- Pointer chains\n- Structure reconstruction\n\n### Process Injection\n- DLL injection methods\n- Manual mapping\n- Shellcode injection\n- Thread hijacking\n- APC injection\n\n### Hooking Techniques\n- Inline hooking (detours)\n- IAT\u002FEAT hooking\n- VTable hooking\n- Hardware breakpoint hooks\n- Syscall hooking\n\n## Cheat Categories\n\n### Visual Cheats (ESP)\n```\n- World-to-Screen transformation\n- Player\u002Fentity rendering\n- Box ESP, skeleton ESP\n- Item highlighting\n- Radar\u002Fminimap hacks\n```\n\n### Aim Assistance\n```\n- Aimbot algorithms (memory-based and AI visual)\n- Triggerbot (auto-fire on crosshair detection)\n- No recoil\u002Fno spread\n- Bullet prediction and lead calculation\n- Silent aim (server-side angle manipulation)\n- AI visual aimbot (YOLO-based, no memory access required)\n```\n\n### AI Visual Cheats (Computer Vision Aimbot)\n```\nArchitecture overview:\nScreen-capture paradigm — uses frame capture, object detection, and input\ninjection. Some implementations can avoid process attachment, a cheat driver,\nand direct game-memory reads; that does not make the full pipeline artifact-free.\n\nTypical setup:\n┌─────────────────┐     screen capture      ┌──────────────────┐\n│  Gaming PC      │ ───────────────────────▶ │  AI Pipeline     │\n│  Game + OBS     │                          │  (same PC, or    │\n│                 │ ◀─────────────────────── │   second PC)     │\n└─────────────────┘     hardware input       │  YOLO model      │\n                        (KMBox \u002F Logitech)   │  TensorRT\u002FCUDA   │\n                                             └──────────────────┘\n\nDual-machine variant (maximum isolation):\n- Machine A (game): only runs game + OBS, sends frames via NDI\u002Fcapture card\n- Machine B (cheat): runs AI model, sends mouse commands via USB\u002Fnetwork\n  to hardware input device on Machine A\n- Game machine need not run the model or decision logic, though capture,\n  transport, and input-device artifacts can remain\n\nSingle-machine variant:\n- OBS + AI model run on the same PC\n- AI implemented as OBS filter plugin (looks like \"OBS is running\")\n- Mouse output via hardware device or driver-level injection\n\nPipeline stages:\n\n1. Frame Capture:\n   - OBS Game Capture (injects graphics hook DLL into game process)\n   - OBS Window Capture (no injection, uses DXGI Desktop Duplication)\n   - OBS plugin filter form (AI as OBS filter, minimal footprint)\n   - Direct framebuffer copy from GPU output layer (60+ FPS)\n   - Capture card (for dual-machine: HDMI\u002FDP input on cheat PC)\n\n2. AI Object Detection:\n   - Model: YOLOv5 \u002F YOLOv8 \u002F YOLOv10 \u002F YOLO11 (lightweight variants)\n   - Training: fine-tuned on game-specific screenshots\n     (enemy bodies, heads, torsos as labeled bounding boxes)\n   - Input: cropped region around crosshair (320x320 or 640x640)\n     to reduce inference cost\n   - Output: bounding boxes with class (head\u002Fbody\u002Fenemy) + confidence score\n   - Acceleration: TensorRT (NVIDIA), CUDA, DirectML, OpenVINO\n   - Set and measure the latency budget on the target capture path, model,\n     hardware, frame rate, and input transport\n\n3. Coordinate Transform and Aiming Logic:\n   - Convert pixel coordinates to mouse movement delta:\n     delta_x = (target_x - screen_center_x) * sensitivity\n     delta_y = (target_y - screen_center_y) * sensitivity\n   - Target selection: closest to crosshair, highest confidence,\n     head priority, or combined scoring\n   - FOV (Field of View) lock: only engage targets within\n     configurable pixel radius from crosshair center\n\n4. Attempts to mask automated trajectories:\n   - Gradual movement with an acceleration curve instead of an instant snap\n   - Synthetic jitter\n   - Bézier curve or cubic interpolation for path\n   - End-point correction (overshoot then settle)\n   - Configurable engagement probability\n   - Slight intentional offset (not pixel-perfect center-mass)\n   - Variable reaction delay\n   These transformations do not establish human equivalence; repeated\n   parametric behavior can itself become a feature.\n\n5. Mouse Movement Execution:\n   - Hardware input devices (see Input Simulation section below)\n   - Movement commands sent as physical HID reports\n   - The host receives protocol-conformant HID input rather than a user-mode\n     injection API call; device provenance and behavior may still be observable\n\nWhy OBS specifically:\n- Legitimate streaming software, used by millions of streamers\n- Blanket action against OBS-related processes would create substantial\n  collateral impact; process presence alone is not attribution\n- Game Capture provides fast, low-latency frame access\n- Plugin system can host filters inside OBS, but loaded plugins, behavior, and\n  surrounding telemetry may still be inspected\n- Supports D3D11, D3D12, Vulkan, OpenGL capture paths\n```\n\n### YOLO Model Training Pipeline (for Game AI Aimbot)\n```\nEnd-to-end workflow from raw game screenshots to deployed TensorRT model.\n\n1. Data Collection:\n   - Capture game screenshots during actual gameplay (OBS recording or replay)\n   - Capture diverse scenarios: different maps, lighting, character skins,\n     distances, poses, partial occlusion, smoke\u002Fflash effects\n   - Determine dataset size from coverage and learning curves; image count alone\n     does not guarantee robustness\n   - Include negative samples (empty scenes, friendlies, environment objects)\n\n2. Annotation \u002F Labeling:\n   - Tools: LabelImg (YOLO format), CVAT (collaborative), Roboflow (cloud),\n     Label Studio, makesense.ai (browser-based)\n   - YOLO format: one .txt per image, each line:\n     \u003Cclass_id> \u003Ccenter_x> \u003Ccenter_y> \u003Cwidth> \u003Cheight>\n     (all values normalized to 0-1 relative to image dimensions)\n   - Class definitions (typical):\n     0: enemy_body (full body bounding box)\n     1: enemy_head (head-only bounding box, for headshot targeting)\n     2: friendly (to avoid shooting teammates)\n   - Label head separately from body for head-priority targeting\n   - Quality control: consistent label boundaries, no missed instances\n\n3. Data Augmentation:\n   - Built-in Ultralytics augmentations (mosaic, mixup, copy-paste)\n   - Game-specific augmentations:\n     - Brightness\u002Fcontrast variation (simulate different map lighting)\n     - Random crop around crosshair area (match inference ROI)\n     - Motion blur (simulate fast movement)\n     - Noise injection (simulate compression artifacts)\n   - Avoid augmentations that distort aspect ratio\n     (characters would look unnatural, hurting accuracy)\n\n4. Training:\n   - Framework: Ultralytics YOLOv8\u002Fv10\u002Fv11\u002FYOLO11\n   - Base model: yolov8n.pt or yolov8s.pt (nano\u002Fsmall for speed)\n     or yolo11n.pt for latest architecture\n   - Training command:\n     yolo detect train data=game_dataset.yaml model=yolov8n.pt\n       epochs=100 imgsz=640 batch=16 device=0\n   - dataset.yaml structure:\n     path: \u002Fpath\u002Fto\u002Fdataset\n     train: images\u002Ftrain\n     val: images\u002Fval\n     names: {0: enemy_body, 1: enemy_head, 2: friendly}\n   - Key hyperparameters to tune include input size, learning rate, confidence\n     threshold, NMS IoU threshold, batch size, and augmentation policy\n   - Measure training and inference cost on the exact model, software stack,\n     precision, and target hardware\n\n5. Validation and Testing:\n   - Evaluate mAP@0.5 and mAP@0.5:0.95 on validation set\n   - Choose operating thresholds from precision\u002Frecall and downstream error\n     costs; no single mAP cutoff establishes reliable deployment\n   - Test inference speed on target hardware and evaluate held-out maps, skins,\n     patches, capture paths, and hard negatives\n\n6. Export to TensorRT (deployment):\n   - Step 1: Export to ONNX\n     yolo export model=best.pt format=onnx simplify=True opset=17\n   - Step 2: Convert ONNX to TensorRT engine\n     yolo export model=best.pt format=engine half=True device=0\n     (half=True enables FP16 precision)\n   - Or use trtexec directly:\n     trtexec --onnx=best.onnx --saveEngine=best.engine\n       --fp16 --workspace=4096\n   - Benchmark FP16 against FP32 on the exported model; latency, throughput, and\n     accuracy changes are hardware- and graph-specific\n   - INT8 can improve throughput but requires representative calibration data\n     and accuracy validation\n\n7. Runtime Integration:\n   - Load TensorRT engine in C++\u002FPython inference loop\n   - Input: preprocessed frame (resize, normalize, HWC→CHW, float32\u002F16)\n   - Decode the exporter\u002Fversion-specific output tensor; shapes and NMS\n     placement vary across model and runtime versions\n   - Apply NMS (Non-Maximum Suppression) to deduplicate detections\n   - Select target based on: closest to crosshair + highest confidence\n   - Convert pixel coordinates to mouse delta\n\nAlternative acceleration backends:\n- DirectML (AMD GPUs, Windows native)\n- OpenVINO (Intel GPUs\u002FCPUs)\n- ONNX Runtime with CUDA EP (cross-platform)\n- CoreML (macOS, less common for game cheats)\n```\n\n### Movement Cheats\n```\n- Speed hacks\n- Fly hacks\n- No clip\n- Teleportation\n- Bunny hop automation\n```\n\n### Miscellaneous\n```\n- Wallhacks\n- Skin changers\n- Unlock all\n- Economy manipulation\n```\n\n## Overlay & Rendering\n\n### Overlay Methods\n- **DirectX Hook**: D3D9\u002F11\u002F12 Present hook\n- **Vulkan Hook**: vkQueuePresentKHR hook\n- **OpenGL Hook**: wglSwapBuffers hook\n- **DWM Overlay**: Desktop Window Manager\n- **External Window**: Transparent overlay window\n- **Steam Overlay**: Hijacking Steam's overlay\n- **NVIDIA Overlay**: GeForce Experience hijack\n\n### Rendering Libraries\n- **Dear ImGui**: Immediate mode GUI\n- **GDI\u002FGDI+**: Windows graphics\n- **Direct2D**: Hardware-accelerated 2D\n\n## Memory Access Methods\n\n### User-Mode\n```\n- OpenProcess + ReadProcessMemory\n- NtReadVirtualMemory\n- Memory-mapped files\n- Shared memory sections\n```\n\n### Kernel-Mode\n```\n- Driver-based access\n- Physical memory access\n- MDL-based copying\n- KeStackAttachProcess\n```\n\n### Advanced Methods\n```\n- DMA (Direct Memory Access)\n- EFI runtime services\n- Hypervisor-based access\n- Hardware-based (FPGA)\n```\n\n## EFI\u002FUEFI Cheats\n\n### Boot-Time Loading\n```\n- EFI manual map: load unsigned driver payload during UEFI boot phase\n- ExitBootServices hook: intercept Windows boot to inject kernel code\n- Runtime DXE drivers: persist across OS boot via EFI runtime services\n- GetVariable\u002FSetVariable: communicate between EFI and OS runtime\n```\n\n### EFI-Based Memory Access\n```\n- Map physical memory via EFI runtime services\n- Bypass DSE entirely (code runs before Windows kernel loads)\n- Survive Secure Boot if firmware is compromised or test-signed\n- Combine with DMA for maximum stealth\n```\n\n### Detection Challenges\n```\n- Pre-OS or manually mapped payloads may avoid the normal Windows image-load\n  path, so a corresponding driver image callback need not occur\n- Artifacts depend on later payload stages; absence from MmUnloadedDrivers or\n  PiDDBCacheTable is not guaranteed\n- Secure Boot + TPM attestation is primary defense\n- Firmware integrity measurement (UEFI capsule verification)\n```\n\n## HWID Spoofing\n\n### Targets\n```\n- Disk serial: IOCTL_STORAGE_QUERY_PROPERTY, SMART data\n- NIC MAC address: NDIS OID_802_3_PERMANENT_ADDRESS\n- SMBIOS: motherboard serial, system UUID, BIOS vendor\n- GPU serial: registry-based or NVAPI\u002FADL queries\n- Monitor EDID: display serial number\n- Volume serial: NtQueryVolumeInformationFile\n- TPM EK: Endorsement Key fingerprint\n```\n\n### Techniques\n```\n- Disk filter driver: intercept IOCTL and replace serial in response\n- Registry value spoofing: modify cached hardware IDs\n- SMBIOS table patching: modify raw SMBIOS memory region\n- NIC driver hook: replace MAC in NDIS miniport response\n- Full HWID spoofer: coordinated spoofing across all identifiers\n```\n\n## Stack Spoofing\n\n### Return Address Spoofing\n```\n- Replace return address on stack before API call\n- Restore original after call returns\n- Evades stack-walk-based detection (RtlWalkFrameChain)\n- Techniques: JMP RBX gadget, synthetic frames, fiber-based\n```\n\n### Call Stack Reconstruction\n```\n- Build fake but plausible call stack frames\n- Match expected module return addresses (ntdll, kernel32)\n- Evade NtQueryInformationThread stack inspection\n- Tools: SpoofCallStack, Vulcan, CallStackSpoofer\n```\n\n### Detection & Evasion\n```\n- Anti-cheat walks thread stacks looking for non-module returns\n- Stack unwinding via .pdata \u002F UNWIND_INFO validation\n- Spoofed stacks must pass RtlVirtualUnwind consistency checks\n```\n\n## Driver Communication\n\n### Full Taxonomy (40+ methods in README)\n```\nIOCTL-based:\n- Standard DeviceIoControl with custom control codes\n- Buffered I\u002FO, Direct I\u002FO, METHOD_NEITHER\n\nData pointer swaps (abusing legitimate syscalls):\n- NtUserGetObjectInformation\n- NtConvertBetweenAuxiliaryCounterAndPerformanceCounter\n- NtUserRegisterRawInputDevices\n- NtGdiGetCOPPCompatibleOPMInformation\n- NtDxgkGetTrackedWorkloadStatistics\n- NtUserGetPointerInfoList\n- NtUserSetInformationThread\n- NtDCompositionSetChildRootVisual\n- Win32k syscall hooks\n\nShared memory:\n- Named shared sections (ZwCreateSection + ZwMapViewOfSection)\n- Physical memory mapping\n- Shared event objects for signaling\n\nCallback-based:\n- Registry callbacks (CmRegisterCallbackEx)\n- Minifilter communication ports (FltCreateCommunicationPort)\n- Object callbacks with embedded data\n\nUnconventional channels:\n- Named pipes from kernel\n- Window messages (NtUserPostMessage)\n- ETW provider channels\n- Socket from kernel (Winsock Kernel \u002F WSK)\n- File system filter callbacks\n- Debugging APIs (DbgPrint interception)\n```\n\n## World-to-Screen Calculation\n\n### Basic Formula\n```cpp\nVector2 WorldToScreen(Vector3 worldPos, Matrix viewMatrix) {\n    Vector4 clipCoords;\n    clipCoords.x = worldPos.x * viewMatrix[0] + worldPos.y * viewMatrix[4] + \n                   worldPos.z * viewMatrix[8] + viewMatrix[12];\n    clipCoords.y = worldPos.x * viewMatrix[1] + worldPos.y * viewMatrix[5] + \n                   worldPos.z * viewMatrix[9] + viewMatrix[13];\n    clipCoords.w = worldPos.x * viewMatrix[3] + worldPos.y * viewMatrix[7] + \n                   worldPos.z * viewMatrix[11] + viewMatrix[15];\n    \n    if (clipCoords.w \u003C 0.1f) return invalid;\n    \n    Vector2 NDC;\n    NDC.x = clipCoords.x \u002F clipCoords.w;\n    NDC.y = clipCoords.y \u002F clipCoords.w;\n    \n    Vector2 screen;\n    screen.x = (screenWidth \u002F 2) * (NDC.x + 1);\n    screen.y = (screenHeight \u002F 2) * (1 - NDC.y);\n    \n    return screen;\n}\n```\n\n## Engine-Specific Techniques\n\n### Unity (Mono)\n- Assembly-CSharp.dll analysis\n- Mono JIT hooking\n- Il2CppDumper for IL2CPP builds\n- Method address resolution\n\n### Unity (IL2CPP)\n- GameAssembly.dll analysis\n- Metadata recovery\n- Type reconstruction\n- Native hooking\n\n### Unreal Engine\n- GObjects\u002FGNames enumeration\n- UWorld traversal\n- SDK generation (Dumper-7)\n- Blueprint hooking\n\n### Source Engine\n- Entity list enumeration\n- NetVars parsing\n- ConVar manipulation\n- Signature scanning\n\n## Input Simulation\n\n### Software Methods\n- SendInput API\n- mouse_event\u002Fkeybd_event\n- DirectInput hooking\n- Raw input injection\n- Driver-based input (mouclass)\n\n### Kernel-Level\n- Mouse class service callback\n- Keyboard filter drivers\n- HID manipulation\n\n### Hardware Input Devices (for AI Visual Cheats)\n```\nHardware input devices produce genuine HID reports indistinguishable\nfrom ordinary input in the fields of an individual protocol-conformant report.\nDescriptors, timing, topology, firmware, and gameplay behavior can still\nprovide imperfect signals.\n\nKMBox series (KMBox Net, KMBox B Pro, KMBox B+):\n- Standalone hardware device connected via USB or network\n- Receives mouse\u002Fkeyboard commands over TCP\u002FUDP or serial\n- Generates real USB HID reports to the gaming PC\n- Gaming PC sees a standard USB mouse, not API-injected input\n- Network variant enables dual-machine setups\n- Supports relative movement, absolute positioning, button events\n- API: simple serial\u002Fnetwork protocol for move(dx, dy), click, etc.\n\nArduino \u002F Teensy \u002F STM32 microcontroller:\n- Custom firmware emulating USB HID device\n- Receives commands from cheat PC via serial\u002FUSB CDC\n- Generates USB HID mouse reports\n- Cheapest hardware option, fully customizable\n- Leonardo \u002F Pro Micro (ATmega32U4) most common for native USB HID\n\nLogitech driver exploitation:\n- Older versions of G HUB \u002F LGS (Logitech Gaming Software) expose\n  internal APIs for mouse movement\n- ghub_mouse_move() or lgs_mouse_move() via DLL injection into GHUB\n- Logitech devices have driver-level whitelist advantage\n- Specific driver versions required (newer versions patched)\n- No external hardware needed, but driver-version-dependent\n\nInterception driver (interception.sys):\n- Open-source keyboard\u002Fmouse filter driver\n- Intercepts and injects input at driver level\n- Commonly used with AI aimbots for zero-hardware-cost injection\n- Detectable by anti-cheat (driver signature known)\n\nHDMI\u002FDP KVM-style middleman:\n- Hardware device sitting between mouse and PC\n- Intercepts real mouse data, injects AI-calculated deltas\n- Can preserve much of the expected interface behavior, depending on its USB\n  descriptors, timing, topology, and electrical implementation\n- Potentially low software footprint but complex hardware setup\n\nIllustrative detection-surface ordering, not a universal ranking:\n1. Dedicated hardware — fewer software artifacts, but device and behavior\n   signals remain\n2. KVM middleman — limited host software footprint if protocol behavior matches\n3. Vendor-driver abuse — version and process\u002Fmodule artifacts may be available\n4. Known filter driver — driver identity and behavior may be available\n5. User-mode injection API — API\u002Fcall-path telemetry may be available\n```\n\n### KMBox Protocol Details\n```\nKMBox Net (network variant) — UDP-based protocol:\n\nPacket header (16 bytes, Little-Endian):\nOffset  Field      Size   Description\n0x00    MAC        4 B    Device UUID (unique per device, used for auth)\n0x04    RAND       4 B    Random value or parameter\n0x08    INDEXPTS   4 B    Incrementing sequence number (replay protection)\n0x0C    CMD        4 B    Command code\n\nKey command codes:\nThe values below are firmware\u002FAPI-version examples; verify them against the\nexact device implementation before analysis.\nCode          Command          Description\n0xAF3C2828    connect          Establish connection with device\n0xAEDE7345    mouse_move       Direct mouse movement (dx, dy)\n0xAEDE7346    mouse_automove   Human-like movement with interpolation\n0xA238455A    mouse_beizer     Bézier curve mouse movement\n0x9823AE8D    mouse_left       Left button press\u002Frelease\n0x238D8212    mouse_right      Right button press\u002Frelease\n0x97A3AE8D    mouse_middle     Middle button press\u002Frelease\n0xFFEEAD38    mouse_wheel      Scroll wheel\n0x123C2C2F    keyboard_all     Keyboard key event\n\nMouse API functions:\n- move(x, y):                 Direct relative movement, no interpolation\n- move_auto(x, y, ms):        Interpolated movement over ms milliseconds\n- move_beizer(x, y, ms,       Second-order Bézier curve with custom\n    x1, y1, x2, y2):          control points for trajectory shaping\n\nEncrypted variants (enc_*):   Same functions with packet-level encryption\n                               to resist network packet analysis\n\nPerformance:\n- Measure command rate, latency distribution, loss, buffering, and jitter on\n  the exact firmware, transport, host, and network; fixed figures do not\n  transfer across setups\n\nKMBox B \u002F B Pro (serial variant):\n- USB CDC serial communication (COM port)\n- Baud rate is firmware\u002Fconfiguration-specific (115200 is one common setting)\n- Simpler protocol: ASCII or binary command frames\n- Benchmark round-trip timing on the deployed serial stack\n\nPhysical keyboard\u002Fmouse monitoring:\n- monitor() function reads real user input from the device\n- Enables \"pass-through + inject\" mode:\n  real user input flows through normally,\n  AI-calculated deltas are added on top\n\nArduino \u002F Teensy HID protocol:\n- Custom serial command format (typically simple ASCII):\n  \"M,dx,dy\\n\"      — mouse move\n  \"C,button\\n\"      — click (1=left, 2=right, 3=middle)\n  \"K,keycode\\n\"     — keypress\n- USB HID report generated by ATmega32U4 (Leonardo)\n  or ARM-based Teensy (3.2, 4.0, 4.1)\n- HID report descriptor mimics standard mouse:\n  buttons (3 bits) + X delta (8-16 bits) + Y delta (8-16 bits)\n- No custom driver needed — OS uses generic HID driver\n\nLogitech driver API (exploitable versions):\n- G HUB versions prior to certain patches expose internal functions\n- Key DLLs: LGS (lcore.dll), G HUB (ghub_mouse.dll or internal APIs)\n- ghub_mouse_move(dx, dy) or equivalent internal symbol\n- Accessed via DLL injection into GHUB process\n  or LoadLibrary + GetProcAddress\n- Movement appears as Logitech device input in the HID stack\n- Patched in newer G HUB versions; specific version numbers\n  circulate in cheat communities\n```\n\n## Anti-Detection Techniques\n\n### Code Protection\n- Polymorphic code\n- Code virtualization\n- Anti-dump techniques\n- String encryption\n\n### Runtime Evasion\n- Stack spoofing\n- Return address manipulation\n- Thread context hiding\n- Module concealment\n\n## Development Workflow\n\n### External Cheat\n```\n1. Pattern scan for signatures\n2. Read game memory externally\n3. Process data in separate process\n4. Render overlay or use input simulation\n```\n\n### Internal Cheat\n```\n1. Inject into game process\n2. Hook rendering functions\n3. Access game objects directly\n4. Render through game's graphics context\n```\n\n## Learning Resources\n\n### Communities\n- UnknownCheats\n- GuidedHacking\n- Game Hacking Academy\n\n### Practice Targets\n- PWN Adventure (intentionally vulnerable)\n- CTF game challenges\n- Older\u002Funsupported games\n\n---\n\n## Data Source\n\n**Important**: This skill provides conceptual guidance and overview information. For detailed information use the following sources:\n\n### 1. Project Overview & Resource Index\n\nFetch the main README for the full curated list of repositories, tools, and descriptions:\n\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002FREADME.md\n```\n\nThe main README contains thousands of curated links organized by category. When users ask for specific tools, projects, or implementations, retrieve and reference the appropriate sections from this source.\n\n### 2. Repository Code Details (Archive)\n\nFor detailed repository information (file structure, source code, implementation details), the project maintains a local archive. If a repository has been archived, **always prefer fetching from the archive** over cloning or browsing GitHub directly.\n\n**Archive URL format:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Farchive\u002F{owner}\u002F{repo}.txt\n```\n\n**Examples:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Farchive\u002Fufrisk\u002Fpcileech.txt\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Farchive\u002F000-aki-000\u002FGameDebugMenu.txt\n```\n\n**How to use:**\n1. Identify the GitHub repository the user is asking about (owner and repo name from the URL).\n2. Construct the archive URL: replace `{owner}` with the GitHub username\u002Forg and `{repo}` with the repository name (no `.git` suffix).\n3. Fetch the archive file — it contains a full code snapshot with file trees and source code generated by `code2prompt`.\n4. If the fetch returns a 404, the repository has not been archived yet; fall back to the README or direct GitHub browsing.\n\n### 3. Repository Descriptions\n\nFor a concise English summary of what a repository does, the project maintains auto-generated description files.\n\n**Description URL format:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Fdescription\u002F{owner}\u002F{repo}\u002Fdescription_en.txt\n```\n\n**Examples:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Fdescription\u002F00christian00\u002FUnityDecompiled\u002Fdescription_en.txt\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Fdescription\u002Fufrisk\u002Fpcileech\u002Fdescription_en.txt\n```\n\n**How to use:**\n1. Identify the GitHub repository the user is asking about (owner and repo name from the URL).\n2. Construct the description URL: replace `{owner}` with the GitHub username\u002Forg and `{repo}` with the repository name.\n3. Fetch the description file — it contains a short, human-readable summary of the repository's purpose and contents.\n4. If the fetch returns a 404, the description has not been generated yet; fall back to the README entry or the archive.\n\n**Priority order when answering questions about a specific repository:**\n1. Description (quick summary) — fetch first for concise context\n2. Archive (full code snapshot) — fetch when deeper implementation details are needed\n3. README entry — fallback when neither description nor archive is available\n\n---\n\n## Compiled wiki\n\nPrefer the compiled domain overview at `wiki\u002Foverviews\u002Fgame-hacking.md` (see `wiki\u002Findex.md` and `wiki\u002FAGENTS.md`) before re-deriving synthesis from raw README\u002Farchive material.\n","# Game Hacking Techniques\n\n## Overview\n\nThis skill covers game-hacking techniques documented in the awesome-game-security collection, with emphasis on how cheats move from user mode to kernel mode, hypervisors, and DMA when defenders raise the bar. It is best used to understand the offensive side of the threat model that anti-cheat systems are designed to detect.\n\nTreat implementations, performance numbers, stealth rankings, and detection\nclaims as versioned threat-model examples rather than guarantees. Use\n[`research-rigor`](..\u002Fresearch-rigor\u002FSKILL.md) when converting them into a\nfactual claim or defensive decision.\n\n## README Coverage\n\n- `Cheat > Debugging`\n- `Cheat > Packet Sniffer&Filter`\n- `Cheat > Packet Capture&Parse`\n- `Cheat > SpeedHack`\n- `Cheat > Injection:Windows`\n- `Cheat > Injection:Linux`\n- `Cheat > Injection:Android`\n- `Cheat > Injection:IOS`\n- `Cheat > Injection:PlayStation`\n- `Cheat > DLL Hijack`\n- `Cheat > Hook`\n- `Cheat > Anti Signature Scanning`\n- `Cheat > RPM`\n- `Cheat > DMA`\n- `Cheat > W2S`\n- `Cheat > Overlay`\n- `Cheat > Render\u002FDraw`\n- `Cheat > UI Interface`\n- `Cheat > Vulnerable Driver`\n- `Cheat > Driver Communication`\n- `Cheat > EFI Driver`\n- `Cheat > QEMU\u002FKVM\u002FPVE\u002FVBOX`\n- `Cheat > Wine`\n- `Cheat > Anti Screenshot`\n- `Cheat > Spoof Stack`\n- `Cheat > Hide`\n- `Cheat > Anti Forensics`\n- `Cheat > Triggerbot & Aimbot`\n- `Cheat > WallHack`\n- `Cheat > HWID`\n- `Cheat > Bypass Page Protection`\n- `Cheat > SDK CodeGen`\n- `Cheat > Game Engine Explorer:*`\n- `Cheat > Explore UWP`\n- `Cheat > Explore AntiCheat System:*`\n- `Cheat > Game:*`\n- `Cheat > Launcher Abuser`\n- `Cheat > Linux Kernel Explorer`\n- `Cheat > Cheat Engine Plugins`\n- `Some Tricks > Windows Ring0`\n- `Some Tricks > Windows Ring3`\n- `Some Tricks > Linux`\n- `Some Tricks > Android`\n\n## Escalation Model\n\n### User-Mode\n- Read and write process memory\n- Inject DLLs or shellcode\n- Hook graphics or input APIs\n\n### Kernel-Mode\n- Use signed or vulnerable drivers for direct memory access\n- Bypass handle-based protections and inspect protected processes\n- Interact with callbacks, page tables, or kernel objects directly\n\n### Below the OS\n- Virtualize the system with a hypervisor\n- Read memory through PCIe DMA hardware\n- Move logic to external devices or secondary machines\n\n## Core Concepts\n\n### Memory Manipulation\n- Read Process Memory (RPM)\n- Write Process Memory (WPM)\n- Pattern scanning\n- Pointer chains\n- Structure reconstruction\n\n### Process Injection\n- DLL injection methods\n- Manual mapping\n- Shellcode injection\n- Thread hijacking\n- APC injection\n\n### Hooking Techniques\n- Inline hooking (detours)\n- IAT\u002FEAT hooking\n- VTable hooking\n- Hardware breakpoint hooks\n- Syscall hooking\n\n## Cheat Categories\n\n### Visual Cheats (ESP)\n```\n- World-to-Screen transformation\n- Player\u002Fentity rendering\n- Box ESP, skeleton ESP\n- Item highlighting\n- Radar\u002Fminimap hacks\n```\n\n### Aim Assistance\n```\n- Aimbot algorithms (memory-based and AI visual)\n- Triggerbot (auto-fire on crosshair detection)\n- No recoil\u002Fno spread\n- Bullet prediction and lead calculation\n- Silent aim (server-side angle manipulation)\n- AI visual aimbot (YOLO-based, no memory access required)\n```\n\n### AI Visual Cheats (Computer Vision Aimbot)\n```\nArchitecture overview:\nScreen-capture paradigm — uses frame capture, object detection, and input\ninjection. Some implementations can avoid process attachment, a cheat driver,\nand direct game-memory reads; that does not make the full pipeline artifact-free.\n\nTypical setup:\n┌─────────────────┐     screen capture      ┌──────────────────┐\n│  Gaming PC      │ ───────────────────────▶ │  AI Pipeline     │\n│  Game + OBS     │                          │  (same PC, or    │\n│                 │ ◀─────────────────────── │   second PC)     │\n└─────────────────┘     hardware input       │  YOLO model      │\n                        (KMBox \u002F Logitech)   │  TensorRT\u002FCUDA   │\n                                             └──────────────────┘\n\nDual-machine variant (maximum isolation):\n- Machine A (game): only runs game + OBS, sends frames via NDI\u002Fcapture card\n- Machine B (cheat): runs AI model, sends mouse commands via USB\u002Fnetwork\n  to hardware input device on Machine A\n- Game machine need not run the model or decision logic, though capture,\n  transport, and input-device artifacts can remain\n\nSingle-machine variant:\n- OBS + AI model run on the same PC\n- AI implemented as OBS filter plugin (looks like \"OBS is running\")\n- Mouse output via hardware device or driver-level injection\n\nPipeline stages:\n\n1. Frame Capture:\n   - OBS Game Capture (injects graphics hook DLL into game process)\n   - OBS Window Capture (no injection, uses DXGI Desktop Duplication)\n   - OBS plugin filter form (AI as OBS filter, minimal footprint)\n   - Direct framebuffer copy from GPU output layer (60+ FPS)\n   - Capture card (for dual-machine: HDMI\u002FDP input on cheat PC)\n\n2. AI Object Detection:\n   - Model: YOLOv5 \u002F YOLOv8 \u002F YOLOv10 \u002F YOLO11 (lightweight variants)\n   - Training: fine-tuned on game-specific screenshots\n     (enemy bodies, heads, torsos as labeled bounding boxes)\n   - Input: cropped region around crosshair (320x320 or 640x640)\n     to reduce inference cost\n   - Output: bounding boxes with class (head\u002Fbody\u002Fenemy) + confidence score\n   - Acceleration: TensorRT (NVIDIA), CUDA, DirectML, OpenVINO\n   - Set and measure the latency budget on the target capture path, model,\n     hardware, frame rate, and input transport\n\n3. Coordinate Transform and Aiming Logic:\n   - Convert pixel coordinates to mouse movement delta:\n     delta_x = (target_x - screen_center_x) * sensitivity\n     delta_y = (target_y - screen_center_y) * sensitivity\n   - Target selection: closest to crosshair, highest confidence,\n     head priority, or combined scoring\n   - FOV (Field of View) lock: only engage targets within\n     configurable pixel radius from crosshair center\n\n4. Attempts to mask automated trajectories:\n   - Gradual movement with an acceleration curve instead of an instant snap\n   - Synthetic jitter\n   - Bézier curve or cubic interpolation for path\n   - End-point correction (overshoot then settle)\n   - Configurable engagement probability\n   - Slight intentional offset (not pixel-perfect center-mass)\n   - Variable reaction delay\n   These transformations do not establish human equivalence; repeated\n   parametric behavior can itself become a feature.\n\n5. Mouse Movement Execution:\n   - Hardware input devices (see Input Simulation section below)\n   - Movement commands sent as physical HID reports\n   - The host receives protocol-conformant HID input rather than a user-mode\n     injection API call; device provenance and behavior may still be observable\n\nWhy OBS specifically:\n- Legitimate streaming software, used by millions of streamers\n- Blanket action against OBS-related processes would create substantial\n  collateral impact; process presence alone is not attribution\n- Game Capture provides fast, low-latency frame access\n- Plugin system can host filters inside OBS, but loaded plugins, behavior, and\n  surrounding telemetry may still be inspected\n- Supports D3D11, D3D12, Vulkan, OpenGL capture paths\n```\n\n### YOLO Model Training Pipeline (for Game AI Aimbot)\n```\nEnd-to-end workflow from raw game screenshots to deployed TensorRT model.\n\n1. Data Collection:\n   - Capture game screenshots during actual gameplay (OBS recording or replay)\n   - Capture diverse scenarios: different maps, lighting, character skins,\n     distances, poses, partial occlusion, smoke\u002Fflash effects\n   - Determine dataset size from coverage and learning curves; image count alone\n     does not guarantee robustness\n   - Include negative samples (empty scenes, friendlies, environment objects)\n\n2. Annotation \u002F Labeling:\n   - Tools: LabelImg (YOLO format), CVAT (collaborative), Roboflow (cloud),\n     Label Studio, makesense.ai (browser-based)\n   - YOLO format: one .txt per image, each line:\n     \u003Cclass_id> \u003Ccenter_x> \u003Ccenter_y> \u003Cwidth> \u003Cheight>\n     (all values normalized to 0-1 relative to image dimensions)\n   - Class definitions (typical):\n     0: enemy_body (full body bounding box)\n     1: enemy_head (head-only bounding box, for headshot targeting)\n     2: friendly (to avoid shooting teammates)\n   - Label head separately from body for head-priority targeting\n   - Quality control: consistent label boundaries, no missed instances\n\n3. Data Augmentation:\n   - Built-in Ultralytics augmentations (mosaic, mixup, copy-paste)\n   - Game-specific augmentations:\n     - Brightness\u002Fcontrast variation (simulate different map lighting)\n     - Random crop around crosshair area (match inference ROI)\n     - Motion blur (simulate fast movement)\n     - Noise injection (simulate compression artifacts)\n   - Avoid augmentations that distort aspect ratio\n     (characters would look unnatural, hurting accuracy)\n\n4. Training:\n   - Framework: Ultralytics YOLOv8\u002Fv10\u002Fv11\u002FYOLO11\n   - Base model: yolov8n.pt or yolov8s.pt (nano\u002Fsmall for speed)\n     or yolo11n.pt for latest architecture\n   - Training command:\n     yolo detect train data=game_dataset.yaml model=yolov8n.pt\n       epochs=100 imgsz=640 batch=16 device=0\n   - dataset.yaml structure:\n     path: \u002Fpath\u002Fto\u002Fdataset\n     train: images\u002Ftrain\n     val: images\u002Fval\n     names: {0: enemy_body, 1: enemy_head, 2: friendly}\n   - Key hyperparameters to tune include input size, learning rate, confidence\n     threshold, NMS IoU threshold, batch size, and augmentation policy\n   - Measure training and inference cost on the exact model, software stack,\n     precision, and target hardware\n\n5. Validation and Testing:\n   - Evaluate mAP@0.5 and mAP@0.5:0.95 on validation set\n   - Choose operating thresholds from precision\u002Frecall and downstream error\n     costs; no single mAP cutoff establishes reliable deployment\n   - Test inference speed on target hardware and evaluate held-out maps, skins,\n     patches, capture paths, and hard negatives\n\n6. Export to TensorRT (deployment):\n   - Step 1: Export to ONNX\n     yolo export model=best.pt format=onnx simplify=True opset=17\n   - Step 2: Convert ONNX to TensorRT engine\n     yolo export model=best.pt format=engine half=True device=0\n     (half=True enables FP16 precision)\n   - Or use trtexec directly:\n     trtexec --onnx=best.onnx --saveEngine=best.engine\n       --fp16 --workspace=4096\n   - Benchmark FP16 against FP32 on the exported model; latency, throughput, and\n     accuracy changes are hardware- and graph-specific\n   - INT8 can improve throughput but requires representative calibration data\n     and accuracy validation\n\n7. Runtime Integration:\n   - Load TensorRT engine in C++\u002FPython inference loop\n   - Input: preprocessed frame (resize, normalize, HWC→CHW, float32\u002F16)\n   - Decode the exporter\u002Fversion-specific output tensor; shapes and NMS\n     placement vary across model and runtime versions\n   - Apply NMS (Non-Maximum Suppression) to deduplicate detections\n   - Select target based on: closest to crosshair + highest confidence\n   - Convert pixel coordinates to mouse delta\n\nAlternative acceleration backends:\n- DirectML (AMD GPUs, Windows native)\n- OpenVINO (Intel GPUs\u002FCPUs)\n- ONNX Runtime with CUDA EP (cross-platform)\n- CoreML (macOS, less common for game cheats)\n```\n\n### Movement Cheats\n```\n- Speed hacks\n- Fly hacks\n- No clip\n- Teleportation\n- Bunny hop automation\n```\n\n### Miscellaneous\n```\n- Wallhacks\n- Skin changers\n- Unlock all\n- Economy manipulation\n```\n\n## Overlay & Rendering\n\n### Overlay Methods\n- **DirectX Hook**: D3D9\u002F11\u002F12 Present hook\n- **Vulkan Hook**: vkQueuePresentKHR hook\n- **OpenGL Hook**: wglSwapBuffers hook\n- **DWM Overlay**: Desktop Window Manager\n- **External Window**: Transparent overlay window\n- **Steam Overlay**: Hijacking Steam's overlay\n- **NVIDIA Overlay**: GeForce Experience hijack\n\n### Rendering Libraries\n- **Dear ImGui**: Immediate mode GUI\n- **GDI\u002FGDI+**: Windows graphics\n- **Direct2D**: Hardware-accelerated 2D\n\n## Memory Access Methods\n\n### User-Mode\n```\n- OpenProcess + ReadProcessMemory\n- NtReadVirtualMemory\n- Memory-mapped files\n- Shared memory sections\n```\n\n### Kernel-Mode\n```\n- Driver-based access\n- Physical memory access\n- MDL-based copying\n- KeStackAttachProcess\n```\n\n### Advanced Methods\n```\n- DMA (Direct Memory Access)\n- EFI runtime services\n- Hypervisor-based access\n- Hardware-based (FPGA)\n```\n\n## EFI\u002FUEFI Cheats\n\n### Boot-Time Loading\n```\n- EFI manual map: load unsigned driver payload during UEFI boot phase\n- ExitBootServices hook: intercept Windows boot to inject kernel code\n- Runtime DXE drivers: persist across OS boot via EFI runtime services\n- GetVariable\u002FSetVariable: communicate between EFI and OS runtime\n```\n\n### EFI-Based Memory Access\n```\n- Map physical memory via EFI runtime services\n- Bypass DSE entirely (code runs before Windows kernel loads)\n- Survive Secure Boot if firmware is compromised or test-signed\n- Combine with DMA for maximum stealth\n```\n\n### Detection Challenges\n```\n- Pre-OS or manually mapped payloads may avoid the normal Windows image-load\n  path, so a corresponding driver image callback need not occur\n- Artifacts depend on later payload stages; absence from MmUnloadedDrivers or\n  PiDDBCacheTable is not guaranteed\n- Secure Boot + TPM attestation is primary defense\n- Firmware integrity measurement (UEFI capsule verification)\n```\n\n## HWID Spoofing\n\n### Targets\n```\n- Disk serial: IOCTL_STORAGE_QUERY_PROPERTY, SMART data\n- NIC MAC address: NDIS OID_802_3_PERMANENT_ADDRESS\n- SMBIOS: motherboard serial, system UUID, BIOS vendor\n- GPU serial: registry-based or NVAPI\u002FADL queries\n- Monitor EDID: display serial number\n- Volume serial: NtQueryVolumeInformationFile\n- TPM EK: Endorsement Key fingerprint\n```\n\n### Techniques\n```\n- Disk filter driver: intercept IOCTL and replace serial in response\n- Registry value spoofing: modify cached hardware IDs\n- SMBIOS table patching: modify raw SMBIOS memory region\n- NIC driver hook: replace MAC in NDIS miniport response\n- Full HWID spoofer: coordinated spoofing across all identifiers\n```\n\n## Stack Spoofing\n\n### Return Address Spoofing\n```\n- Replace return address on stack before API call\n- Restore original after call returns\n- Evades stack-walk-based detection (RtlWalkFrameChain)\n- Techniques: JMP RBX gadget, synthetic frames, fiber-based\n```\n\n### Call Stack Reconstruction\n```\n- Build fake but plausible call stack frames\n- Match expected module return addresses (ntdll, kernel32)\n- Evade NtQueryInformationThread stack inspection\n- Tools: SpoofCallStack, Vulcan, CallStackSpoofer\n```\n\n### Detection & Evasion\n```\n- Anti-cheat walks thread stacks looking for non-module returns\n- Stack unwinding via .pdata \u002F UNWIND_INFO validation\n- Spoofed stacks must pass RtlVirtualUnwind consistency checks\n```\n\n## Driver Communication\n\n### Full Taxonomy (40+ methods in README)\n```\nIOCTL-based:\n- Standard DeviceIoControl with custom control codes\n- Buffered I\u002FO, Direct I\u002FO, METHOD_NEITHER\n\nData pointer swaps (abusing legitimate syscalls):\n- NtUserGetObjectInformation\n- NtConvertBetweenAuxiliaryCounterAndPerformanceCounter\n- NtUserRegisterRawInputDevices\n- NtGdiGetCOPPCompatibleOPMInformation\n- NtDxgkGetTrackedWorkloadStatistics\n- NtUserGetPointerInfoList\n- NtUserSetInformationThread\n- NtDCompositionSetChildRootVisual\n- Win32k syscall hooks\n\nShared memory:\n- Named shared sections (ZwCreateSection + ZwMapViewOfSection)\n- Physical memory mapping\n- Shared event objects for signaling\n\nCallback-based:\n- Registry callbacks (CmRegisterCallbackEx)\n- Minifilter communication ports (FltCreateCommunicationPort)\n- Object callbacks with embedded data\n\nUnconventional channels:\n- Named pipes from kernel\n- Window messages (NtUserPostMessage)\n- ETW provider channels\n- Socket from kernel (Winsock Kernel \u002F WSK)\n- File system filter callbacks\n- Debugging APIs (DbgPrint interception)\n```\n\n## World-to-Screen Calculation\n\n### Basic Formula\n```cpp\nVector2 WorldToScreen(Vector3 worldPos, Matrix viewMatrix) {\n    Vector4 clipCoords;\n    clipCoords.x = worldPos.x * viewMatrix[0] + worldPos.y * viewMatrix[4] + \n                   worldPos.z * viewMatrix[8] + viewMatrix[12];\n    clipCoords.y = worldPos.x * viewMatrix[1] + worldPos.y * viewMatrix[5] + \n                   worldPos.z * viewMatrix[9] + viewMatrix[13];\n    clipCoords.w = worldPos.x * viewMatrix[3] + worldPos.y * viewMatrix[7] + \n                   worldPos.z * viewMatrix[11] + viewMatrix[15];\n    \n    if (clipCoords.w \u003C 0.1f) return invalid;\n    \n    Vector2 NDC;\n    NDC.x = clipCoords.x \u002F clipCoords.w;\n    NDC.y = clipCoords.y \u002F clipCoords.w;\n    \n    Vector2 screen;\n    screen.x = (screenWidth \u002F 2) * (NDC.x + 1);\n    screen.y = (screenHeight \u002F 2) * (1 - NDC.y);\n    \n    return screen;\n}\n```\n\n## Engine-Specific Techniques\n\n### Unity (Mono)\n- Assembly-CSharp.dll analysis\n- Mono JIT hooking\n- Il2CppDumper for IL2CPP builds\n- Method address resolution\n\n### Unity (IL2CPP)\n- GameAssembly.dll analysis\n- Metadata recovery\n- Type reconstruction\n- Native hooking\n\n### Unreal Engine\n- GObjects\u002FGNames enumeration\n- UWorld traversal\n- SDK generation (Dumper-7)\n- Blueprint hooking\n\n### Source Engine\n- Entity list enumeration\n- NetVars parsing\n- ConVar manipulation\n- Signature scanning\n\n## Input Simulation\n\n### Software Methods\n- SendInput API\n- mouse_event\u002Fkeybd_event\n- DirectInput hooking\n- Raw input injection\n- Driver-based input (mouclass)\n\n### Kernel-Level\n- Mouse class service callback\n- Keyboard filter drivers\n- HID manipulation\n\n### Hardware Input Devices (for AI Visual Cheats)\n```\nHardware input devices produce genuine HID reports indistinguishable\nfrom ordinary input in the fields of an individual protocol-conformant report.\nDescriptors, timing, topology, firmware, and gameplay behavior can still\nprovide imperfect signals.\n\nKMBox series (KMBox Net, KMBox B Pro, KMBox B+):\n- Standalone hardware device connected via USB or network\n- Receives mouse\u002Fkeyboard commands over TCP\u002FUDP or serial\n- Generates real USB HID reports to the gaming PC\n- Gaming PC sees a standard USB mouse, not API-injected input\n- Network variant enables dual-machine setups\n- Supports relative movement, absolute positioning, button events\n- API: simple serial\u002Fnetwork protocol for move(dx, dy), click, etc.\n\nArduino \u002F Teensy \u002F STM32 microcontroller:\n- Custom firmware emulating USB HID device\n- Receives commands from cheat PC via serial\u002FUSB CDC\n- Generates USB HID mouse reports\n- Cheapest hardware option, fully customizable\n- Leonardo \u002F Pro Micro (ATmega32U4) most common for native USB HID\n\nLogitech driver exploitation:\n- Older versions of G HUB \u002F LGS (Logitech Gaming Software) expose\n  internal APIs for mouse movement\n- ghub_mouse_move() or lgs_mouse_move() via DLL injection into GHUB\n- Logitech devices have driver-level whitelist advantage\n- Specific driver versions required (newer versions patched)\n- No external hardware needed, but driver-version-dependent\n\nInterception driver (interception.sys):\n- Open-source keyboard\u002Fmouse filter driver\n- Intercepts and injects input at driver level\n- Commonly used with AI aimbots for zero-hardware-cost injection\n- Detectable by anti-cheat (driver signature known)\n\nHDMI\u002FDP KVM-style middleman:\n- Hardware device sitting between mouse and PC\n- Intercepts real mouse data, injects AI-calculated deltas\n- Can preserve much of the expected interface behavior, depending on its USB\n  descriptors, timing, topology, and electrical implementation\n- Potentially low software footprint but complex hardware setup\n\nIllustrative detection-surface ordering, not a universal ranking:\n1. Dedicated hardware — fewer software artifacts, but device and behavior\n   signals remain\n2. KVM middleman — limited host software footprint if protocol behavior matches\n3. Vendor-driver abuse — version and process\u002Fmodule artifacts may be available\n4. Known filter driver — driver identity and behavior may be available\n5. User-mode injection API — API\u002Fcall-path telemetry may be available\n```\n\n### KMBox Protocol Details\n```\nKMBox Net (network variant) — UDP-based protocol:\n\nPacket header (16 bytes, Little-Endian):\nOffset  Field      Size   Description\n0x00    MAC        4 B    Device UUID (unique per device, used for auth)\n0x04    RAND       4 B    Random value or parameter\n0x08    INDEXPTS   4 B    Incrementing sequence number (replay protection)\n0x0C    CMD        4 B    Command code\n\nKey command codes:\nThe values below are firmware\u002FAPI-version examples; verify them against the\nexact device implementation before analysis.\nCode          Command          Description\n0xAF3C2828    connect          Establish connection with device\n0xAEDE7345    mouse_move       Direct mouse movement (dx, dy)\n0xAEDE7346    mouse_automove   Human-like movement with interpolation\n0xA238455A    mouse_beizer     Bézier curve mouse movement\n0x9823AE8D    mouse_left       Left button press\u002Frelease\n0x238D8212    mouse_right      Right button press\u002Frelease\n0x97A3AE8D    mouse_middle     Middle button press\u002Frelease\n0xFFEEAD38    mouse_wheel      Scroll wheel\n0x123C2C2F    keyboard_all     Keyboard key event\n\nMouse API functions:\n- move(x, y):                 Direct relative movement, no interpolation\n- move_auto(x, y, ms):        Interpolated movement over ms milliseconds\n- move_beizer(x, y, ms,       Second-order Bézier curve with custom\n    x1, y1, x2, y2):          control points for trajectory shaping\n\nEncrypted variants (enc_*):   Same functions with packet-level encryption\n                               to resist network packet analysis\n\nPerformance:\n- Measure command rate, latency distribution, loss, buffering, and jitter on\n  the exact firmware, transport, host, and network; fixed figures do not\n  transfer across setups\n\nKMBox B \u002F B Pro (serial variant):\n- USB CDC serial communication (COM port)\n- Baud rate is firmware\u002Fconfiguration-specific (115200 is one common setting)\n- Simpler protocol: ASCII or binary command frames\n- Benchmark round-trip timing on the deployed serial stack\n\nPhysical keyboard\u002Fmouse monitoring:\n- monitor() function reads real user input from the device\n- Enables \"pass-through + inject\" mode:\n  real user input flows through normally,\n  AI-calculated deltas are added on top\n\nArduino \u002F Teensy HID protocol:\n- Custom serial command format (typically simple ASCII):\n  \"M,dx,dy\\n\"      — mouse move\n  \"C,button\\n\"      — click (1=left, 2=right, 3=middle)\n  \"K,keycode\\n\"     — keypress\n- USB HID report generated by ATmega32U4 (Leonardo)\n  or ARM-based Teensy (3.2, 4.0, 4.1)\n- HID report descriptor mimics standard mouse:\n  buttons (3 bits) + X delta (8-16 bits) + Y delta (8-16 bits)\n- No custom driver needed — OS uses generic HID driver\n\nLogitech driver API (exploitable versions):\n- G HUB versions prior to certain patches expose internal functions\n- Key DLLs: LGS (lcore.dll), G HUB (ghub_mouse.dll or internal APIs)\n- ghub_mouse_move(dx, dy) or equivalent internal symbol\n- Accessed via DLL injection into GHUB process\n  or LoadLibrary + GetProcAddress\n- Movement appears as Logitech device input in the HID stack\n- Patched in newer G HUB versions; specific version numbers\n  circulate in cheat communities\n```\n\n## Anti-Detection Techniques\n\n### Code Protection\n- Polymorphic code\n- Code virtualization\n- Anti-dump techniques\n- String encryption\n\n### Runtime Evasion\n- Stack spoofing\n- Return address manipulation\n- Thread context hiding\n- Module concealment\n\n## Development Workflow\n\n### External Cheat\n```\n1. Pattern scan for signatures\n2. Read game memory externally\n3. Process data in separate process\n4. Render overlay or use input simulation\n```\n\n### Internal Cheat\n```\n1. Inject into game process\n2. Hook rendering functions\n3. Access game objects directly\n4. Render through game's graphics context\n```\n\n## Learning Resources\n\n### Communities\n- UnknownCheats\n- GuidedHacking\n- Game Hacking Academy\n\n### Practice Targets\n- PWN Adventure (intentionally vulnerable)\n- CTF game challenges\n- Older\u002Funsupported games\n\n---\n\n## Data Source\n\n**Important**: This skill provides conceptual guidance and overview information. For detailed information use the following sources:\n\n### 1. Project Overview & Resource Index\n\nFetch the main README for the full curated list of repositories, tools, and descriptions:\n\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002FREADME.md\n```\n\nThe main README contains thousands of curated links organized by category. When users ask for specific tools, projects, or implementations, retrieve and reference the appropriate sections from this source.\n\n### 2. Repository Code Details (Archive)\n\nFor detailed repository information (file structure, source code, implementation details), the project maintains a local archive. If a repository has been archived, **always prefer fetching from the archive** over cloning or browsing GitHub directly.\n\n**Archive URL format:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Farchive\u002F{owner}\u002F{repo}.txt\n```\n\n**Examples:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Farchive\u002Fufrisk\u002Fpcileech.txt\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Farchive\u002F000-aki-000\u002FGameDebugMenu.txt\n```\n\n**How to use:**\n1. Identify the GitHub repository the user is asking about (owner and repo name from the URL).\n2. Construct the archive URL: replace `{owner}` with the GitHub username\u002Forg and `{repo}` with the repository name (no `.git` suffix).\n3. Fetch the archive file — it contains a full code snapshot with file trees and source code generated by `code2prompt`.\n4. If the fetch returns a 404, the repository has not been archived yet; fall back to the README or direct GitHub browsing.\n\n### 3. Repository Descriptions\n\nFor a concise English summary of what a repository does, the project maintains auto-generated description files.\n\n**Description URL format:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Fdescription\u002F{owner}\u002F{repo}\u002Fdescription_en.txt\n```\n\n**Examples:**\n```\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Fdescription\u002F00christian00\u002FUnityDecompiled\u002Fdescription_en.txt\nhttps:\u002F\u002Fraw.githubusercontent.com\u002Fgmh5225\u002Fawesome-game-security\u002Frefs\u002Fheads\u002Fmain\u002Fdescription\u002Fufrisk\u002Fpcileech\u002Fdescription_en.txt\n```\n\n**How to use:**\n1. Identify the GitHub repository the user is asking about (owner and repo name from the URL).\n2. Construct the description URL: replace `{owner}` with the GitHub username\u002Forg and `{repo}` with the repository name.\n3. Fetch the description file — it contains a short, human-readable summary of the repository's purpose and contents.\n4. If the fetch returns a 404, the description has not been generated yet; fall back to the README entry or the archive.\n\n**Priority order when answering questions about a specific repository:**\n1. Description (quick summary) — fetch first for concise context\n2. Archive (full code snapshot) — fetch when deeper implementation details are needed\n3. README entry — fallback when neither description nor archive is available\n\n---\n\n## Compiled wiki\n\nPrefer the compiled domain overview at `wiki\u002Foverviews\u002Fgame-hacking.md` (see `wiki\u002Findex.md` and `wiki\u002FAGENTS.md`) before re-deriving synthesis from raw README\u002Farchive material.\n",{"name":5,"description":7,"x-alice-class":24},"game","main","19a54ac988e9929160334271f5c1924ddd9b227e","_modules\u002Fgame-hacking",19,"https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill","https:\u002F\u002Fgithub.com\u002Falicewe1","写一些自己用得上的工具，能自动化就绝不手动 · 自动化 \u002F 工程效率 \u002F Agent 技能生态 · 保持好奇，持续构建","2026-09-30 00:41:20",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add alicewe1\u002Falice_skill --skill game-hacking-techniques",{"tool":39,"command":40},"bunx","bunx skills add alicewe1\u002Falice_skill --skill game-hacking-techniques",{"tool":42,"command":43},"pnpm","pnpm dlx skills add alicewe1\u002Falice_skill --skill game-hacking-techniques","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002F_modules\u002Fgame-hacking\u002FSKILL.md",true,[],[48,55,62,69,76,83],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":52,"categories":53,"keywords":54,"publishedAt":19},"01M7DBMD9P6R27SWQCW5HYVWG0","alice","Alice 六类技能总路由。用户输入 Alice 或要求查看 Alice 技能菜单时，严格输出固定菜单；输入攻\u002F防模式、六类入口或直接描述任务时，按类别→领域→模块渐进路由。","https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002Faliceskill\u002FSKILL.md",[],[],{"skillNo":56,"name":57,"nameZh":6,"description":58,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M7DBM8X5M7YXPYBVFMWF8HRC","alice-reverse","逆向分析渐进路由（二进制 \u002F 移动端 \u002F 协议 \u002F 取证 \u002F 保护机制 \u002F 运行时观测）。触发词：二进制\u002F移动端\u002F反编译\u002FFrida\u002FIDA\u002FGhidra\u002FPE\u002FELF\u002FAPK\u002F协议\u002F取证；先选领域索引，再按任务匹配度与评分选模块，禁止一次读取全部领域。","https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002Falice-reverse\u002FSKILL.md",[],[],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":19},"01M7DBM630E2YKQP29194TT7HC","alice-pentest","web安全渐进路由（Web \u002F API \u002F 资产发现 \u002F 漏洞验证 \u002F 云与网络安全）。触发词：Web\u002FAPI\u002F资产发现\u002F端口\u002F漏洞验证\u002F云与网络安全；先选领域索引，再按任务匹配度与评分选模块，禁止一次读取全部领域。","https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002Falice-pentest\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M7DBM20M2VDYV5D3MYDGBTQX","alice-game","游戏攻防渐进路由（游戏客户端 \u002F 内存 \u002F 图形 \u002F 反作弊 \u002F 安全研究）。触发词：游戏客户端\u002F内存\u002F图形\u002F偏移\u002F反作弊\u002F模拟器\u002F覆盖层；先选领域索引，再按任务匹配度与评分选模块，禁止一次读取全部领域。","https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002Falice-game\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M7DBKXVEKB8931MZG23VACFN","alice-crack","卡密授权渐进路由（卡密 \u002F 激活 \u002F 授权验证 \u002F 许可证管理 \u002F 本地授权分析）。触发词：卡密\u002F激活\u002F授权验证\u002Flicense\u002F机器码\u002F试用状态；先选领域索引，再按任务匹配度与评分选模块，禁止一次读取全部领域。","https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002Falice-crack\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M7DBKTMT90MGV2WP8K73TMQD","alice-assist","技能指令渐进路由（技能路由 \u002F 评分 \u002F 吸收 \u002F 迁移 \u002F 自动化 \u002F 文档与平台集成）。触发词：技能创建\u002F吸收\u002F评分\u002F迁移\u002F自动化\u002F文档\u002F平台指令；先选领域索引，再按任务匹配度与评分选模块，禁止一次读取全部领域。","https:\u002F\u002Fgithub.com\u002Falicewe1\u002Falice_skill\u002Fblob\u002Fmain\u002Falice-assist\u002FSKILL.md",[],[]]