Four Must‑Watch GitHub Projects: Bitlocker Exploit, ESP32 Dashboard, AI‑Powered 3D Generation, and A‑Share Data Toolkit

YellowKey: Windows 11 Bitlocker Bypass Vulnerability

YellowKey exploits an undocumented component in the Windows Recovery Environment (WinRE) to bypass Bitlocker encryption without the key, requiring only physical access.

Vulnerability Overview

The attack leverages a privileged component loaded during Bitlocker recovery. The steps are:

Format a USB drive as NTFS/FAT32/exFAT.

Place a specific file in System Volume Information\FsTx on the drive.

Insert the drive, reboot, and enter WinRE (Shift + Restart).

Hold the Ctrl key during loading.

The system presents a SYSTEM‑privileged shell.

The process can also be performed by removing the target disk, deploying the files on another machine, and reinstalling the disk, requiring no external device.

Scope and Mitigation

All Windows 11 editions are affected.

Windows Server 2022/2025 are affected.

Windows 10 is not affected.

Await Microsoft security updates.

Enable TPM + PIN two‑factor authentication for Bitlocker.

Maintain strict physical‑access controls.

GitHub: https://github.com/Nightmare-Eclipse/YellowKey<br/>Stars: 1029

Clawdmeter: ESP32‑S3 Desktop Dashboard for Claude Code Usage

Clawdmeter is an ESP32‑S3‑based desktop device with a 2.16‑inch 480×480 AMOLED touchscreen that visualizes Claude Code usage data via BLE.

Hardware Specs

ESP32‑S3R8 processor

2.16" CO5300 AMOLED display (480×480, QSPI)

CST9220 capacitive touch controller

AXP2101 power‑management IC (supports Li‑ion battery)

QMI8658 six‑axis inertial sensor

Three physical buttons: left long‑press sends Space (Claude voice mode), right press sends Shift+Tab (mode switch), middle power button toggles the screen.

Software Architecture

A Python daemon runs on the host PC and performs:

Read ~/.claude/.credentials.json to obtain the OAuth token.

Call Anthropic’s lightweight API endpoint every 60 seconds to fetch usage data.

Push the data to the ESP32 via BLE GATT.

Render the dashboard on the device using the LVGL library.

The on‑screen animation changes intensity based on usage frequency, using assets from the claudepix project.

Quick Start

# Clone the project
git clone https://github.com/HermannBjorgvin/Clawdmeter.git
cd Clawdmeter/firmware
# Flash firmware with PlatformIO
pio run -t upload --upload-port /dev/ttyACM0
# Pair via Bluetooth
bluetoothctl scan le
bluetoothctl pair XX:XX:XX:XX:XX:XX
bluetoothctl trust XX:XX:XX:XX:XX:XX
# Install and start the daemon
./install.sh
systemctl --user start claude-usage-daemon

Target audience: heavy Claude Code users, developers who want to quantify AI‑assisted coding efficiency, and desktop‑hardware enthusiasts.

GitHub: https://github.com/HermannBjorgvin/Clawdmeter<br/>Stars: 693

Pixal3D: Tencent ARC’s Single‑Image 3D Asset Generator

Pixal3D, accepted at SIGGRAPH 2026, generates high‑quality 3D assets from a single image using a pixel‑alignment approach that explicitly back‑projects pixel features into 3D space, achieving near‑reconstruction‑level geometry and PBR texture fidelity.

Technical Idea

Unlike traditional attention‑based methods that loosely inject image features, Pixal3D establishes a direct correspondence between pixels and 3D points, improving detail preservation.

Branches

main : Built on Microsoft TRELLIS.2 backbone, offering superior performance.

paper : Original research version based on Direct3D‑S2, reproducing the results reported in the paper.

Quick Start

# Install base environment per TRELLIS.2 instructions
# Install additional dependencies
pip install -r requirements.txt
# Install utility library
pip install https://github.com/LDYang694/Storages/releases/download/20260430/utils3d-0.0.2-py3-none-any.whl
# Run inference
python inference.py --image assets/test_image/0.png --output ./output.glb

An online demo is available on Hugging Face Spaces.

Suitable for game developers, content creators, 3D modelers, and researchers needing rapid 3D asset generation.

GitHub: https://github.com/TencentARC/Pixal3D<br/>Stars: 535

a‑stock‑data: AI‑Assisted A‑Share Research Toolkit

a‑stock‑data consolidates fragmented A‑share data sources into a skill package that AI coding assistants (Claude Code, Codex, OpenClaw) can invoke directly.

Six‑Layer Architecture (21 Endpoints)

Market Layer : mootdx + Tencent Finance – K‑lines, order‑book, real‑time quotes, PE/PB/market cap.

Research Report Layer : Eastmoney + akshare + iwencai – report lists, PDF download, consensus EPS, NL search.

Signal Layer : Tonghuashun + Baidu Stock – strong stocks, thematic attribution, north‑bound capital, concept plates, fund flows,龙虎榜, unlock calendars.

News Layer : CFI news, individual stock news, global news.

Fundamental Data : 37‑field quarterly reports, 9‑category F10 company info.

Announcement Layer : 巨潮 cninfo full‑text announcements.

Quick Start

# Create skill directory
mkdir -p ~/.claude/skills/a-stock-data
# Download SKILL.md
curl -o ~/.claude/skills/a-stock-data/SKILL.md \
    https://raw.githubusercontent.com/simonlin1212/a-stock-data/main/SKILL.md
# Install dependencies
pip install mootdx akshare requests pandas stockstats

After installation, users can ask Claude Code, "Help me evaluate 688017," and the AI will automatically invoke the appropriate skill.

Built‑in Research Flows

Single‑Stock Valuation : Real‑time price → consensus → forward PE/PEG/years to digest (≈30 s).

Batch Comparison : Horizontal valuation ranking of multiple stocks (≈1 min).

Thematic Report : Multi‑keyword NL search + PDF cross‑reference (≈2 min).

New‑Issue Research : Institutional coverage → valuation → concept → capital →龙虎榜 → unlock calendar (≈1 min).

Only the iwencai semantic search requires an API key; the other seven data sources are free.

GitHub: https://github.com/simonlin1212/a-stock-data<br/>Stars: 496