The Agent Skills Directory

[DATA_EXFILTRATION]: The skill extracts personal information from WeChat Moments (social media posts, likes, and comments) and transmits it to an external third-party API service (https://api.qingyuntop.top/v1) for AI processing. This involves the exfiltration of potentially sensitive personal data to a service outside the user's direct control.
[COMMAND_EXECUTION]: The skill uses pyautogui to perform automated interactions with the host's graphical user interface, including mouse movements, clicks, and screen capturing. While intended for automation, this capability allows the skill to manipulate application windows on the desktop.
[PROMPT_INJECTION]: The wechat_radar_briefing.py script is vulnerable to indirect prompt injection. It interpolates raw, untrusted text from WeChat posts directly into prompts sent to a Large Language Model (LLM).
Ingestion points: The load_moments function reads extracted content from JSON files previously generated from WeChat screenshots.
Boundary markers: No explicit sanitization or delimiters are used to isolate the untrusted WeChat content from the LLM instructions.
Capability inventory: The skill possesses the ability to perform network operations (via httpx) and file system writes (generating reports).
Sanitization: The script performs keyword matching and basic regex filtering but does not prevent the LLM from following instructions embedded within the processed social media content.

wechat-ai-radar