The Agent Skills Directory

[DATA_EXFILTRATION]: The skill manages the AGENTBAY_API_KEY by reading from and writing to a configuration file located at ~/.config/agentbay/api_key in both scripts/crawl.py and scripts/login.py. This ensures credentials persist across sessions but involves access to a sensitive path.
[EXTERNAL_DOWNLOADS]: During PDF report generation, scripts/reporter/pdf_export.py attempts to download CJK font files from the cdn.jsdelivr.net content delivery network if they are not found locally or on the system.
[COMMAND_EXECUTION]: The skill's core functionality relies on the agent executing various Python scripts (crawl.py, login.py, report.py, write_processed.py) and utilizing the AgentBay SDK to perform remote browsing tasks via an automated session.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data retrieved from external platforms such as XHS, Weibo, Douyin, Zhihu, Baidu, and Bing. This data is fed into the agent for sentiment analysis, creating an attack surface where malicious instructions embedded in social media posts could influence the agent's behavior.
Ingestion points: External content (titles, posts, comments) crawled in scripts/crawl.py and processed in scripts/crawler/crawler.py.
Boundary markers: The agent is provided with analysis instructions in scripts/sentiment/sentiment_instruction.md, but these lack explicit delimiters or instructions to ignore commands potentially embedded within the data results.
Capability inventory: The skill environment allows for terminal command execution, file system writes, and network interactions through the provided scripts and SDK adapter.
Sanitization: Content is not sanitized for malicious prompts before being processed by the agent; scripts/sentiment/write_processed.py only ensures valid JSON formatting.

agentbay-monitor-skills