The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core function is to retrieve and process content from various social media platforms (Weibo, Xiaohongshu, Bilibili, Zhihu, etc.). This content is untrusted and is subsequently presented to the AI agent for summarization. The skill does not currently employ explicit boundary markers or specific instructions to the agent to ignore potentially malicious prompts embedded within these search results, which is a common surface for indirect injection attacks.
[DYNAMIC_EXECUTION]: The upstream_bridge.py module facilitates communication with another AI agent skill. It searches for a script named last30days.py within standard installation directories or follows a path provided in the LAST30DAYS_UPSTREAM environment variable. The skill then executes this script using subprocess.run. Although the module performs verification steps to ensure the target script is valid, this represents code execution from paths determined at runtime.
[COMMAND_EXECUTION]: The skill utilizes a SessionStart hook defined in hooks/hooks.json to run a bash script (check-config.sh) whenever a session begins. This script accesses sensitive configuration files, such as .env and cookie stores, to verify their presence and security permissions. While the script is implemented with security in mind (avoiding eval and using a whitelist of known keys), it is an automated command execution triggered by the skill's presence in the environment.

last30days-cn