The Agent Skills Directory

[PERSISTENCE]: The skill implements a 'subscription' feature designed to maintain daily execution. On macOS, it creates a LaunchAgent configuration file at ~/Library/LaunchAgents/com.qoder.gzh-ai-feed.plist. On Linux and other systems, it modifies the user's crontab to schedule the script at 09:00 every day. While advertised as a feature, these mechanisms modify system-level schedulers to maintain long-term access.\n- [COMMAND_EXECUTION]: The script assets/daily_report.py utilizes subprocess.run with shell=True to manage crontab entries. It also invokes launchctl to load and unload background tasks. Executing shell commands with string interpolation presents a security risk if the execution environment or input variables are manipulated.\n- [INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to data poisoning due to a lack of sanitization. 1. Ingestion points: Data is fetched from https://redfox.hk in assets/daily_report.py. 2. Boundary markers: No markers or warnings are used to delimit external content. 3. Capability inventory: The script writes local files and triggers the default web browser via subprocess.run. 4. Sanitization: External article titles, authors, and cover URLs are directly interpolated into the HTML report in the generate_category_cards function, allowing for potential XSS if the API returns malicious strings.\n- [EXTERNAL_DOWNLOADS]: The skill depends on the requests library and performs network operations to https://redfox.hk to retrieve AI news content. The usage of this external service is consistent with the skill's stated purpose.

gzh-ai-feed