Sts2 Vision

杀戮尖塔2视觉识别DPS监控系统 - 通过屏幕捕获和OCR识别实时监控战斗数据

This package appears to implement the advertised local OCR/DPS monitor but has several inconsistencies you should address before installing: - Dependency & OS checks: SKILL.md lists mss, opencv-python, pytesseract, numpy, but the code also imports win32gui (pywin32), pynput, and optionally easyocr. Those are platform/permission-sensitive (win32gui is Windows-only; pynput may require accessibility permissions on macOS). Confirm your OS and install the correct packages. - Privacy & permissions: several monitor variants start a global mouse listener and log click coordinates and timestamps; the tool also writes screenshots and JSON reports to disk. If you run it, expect local captures of on-screen content and mouse activity—run only on a machine where that is acceptable. - No network observed: review the omitted files before trusting — current visible files do not perform network exfiltration, but verify the remaining files for any network calls. - Code issues: there is a bug/typo in main.py (default config key " rois" has a leading space) and other rough edges; test in a controlled environment first. - Best practices: run in an isolated VM or sandbox, inspect and run the code locally (not as an elevated/system service), review/modify the mouse-listening code if you do not want global click logging, and ensure saved screenshots/report files are stored where you expect or disabled via config. If you need stronger assurance, ask the maintainer for an updated SKILL.md listing all runtime dependencies, OS target, and an explicit description of what is logged and saved.

Type: OpenClaw Skill Name: sts2-vision Version: 1.0.0 The skill bundle is a vision-based DPS and combat monitoring system for the game Slay The Spire 2. It utilizes standard libraries such as mss for screen capture, OpenCV for image processing, and pytesseract/easyocr for OCR to track player and enemy health values. The inclusion of a mouse listener (pynput) in online_monitor.py and monitor_v2.py is used to attribute damage based on screen click regions, which aligns with the stated goal of distinguishing between players in a multiplayer setting. No indicators of data exfiltration, malicious persistence, or unauthorized remote execution were found; all reports are saved locally as JSON files.