The Agent Skills Directory

[PERSISTENCE_MECHANISMS]: The skill provides explicit instructions and code to establish automated, recurring tasks on the host system. It includes examples for adding entries to the user's crontab and generating a macOS launchd plist (~/Library/LaunchAgents/com.personal-harvest.plist) to ensure the harvesting scripts run daily.
[DATA_EXPOSURE]: The skill accesses sensitive local application data and caches. Specifically, it targets the SQLite databases for WeChat Reading located in ~/Library/Containers/com.tencent.WeReadMac/Data/Library/Application Support/WeRead/ and the Kindle My Clippings.txt file in the user's Documents folder.
[INDIRECT_PROMPT_INJECTION]: The skill ingests large amounts of untrusted data from external platforms (Douban, Bilibili, Xiaohongshu, etc.) via web scraping and API interception. This data is structured and stored in a local database which is then intended to be read by the AI agent to generate summaries or profiles, creating a surface where malicious instructions embedded in platform content could influence agent behavior.
[COMMAND_EXECUTION]: The skill involves the dynamic generation and execution of Python scripts for database initialization, web scraping, and file parsing. These scripts are written to the local filesystem and executed using the shell.
[DYNAMIC_EXECUTION]: The automation logic utilizes Playwright to launch browser instances with persistent user profiles (launch_persistent_context), allowing the agent to operate within authenticated sessions on social media platforms.

personal-data-harvester