The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the agent-browser package globally using npm from a third-party mirror and executes the mdpdf package via bunx from external registries.
[REMOTE_CODE_EXECUTION]: The script crawl_xueqiu_home_timeline_api.py utilizes an automated browser to evaluate and execute arbitrary JavaScript code via the fetch API to retrieve timeline data from Xueqiu.
[COMMAND_EXECUTION]: The skill executes multiple shell scripts to manage system processes (using pkill and pgrep), install software via brew and npm, and configure browser debugging ports.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted posts from social media and passes them to AI subagents for summarization and analysis.
Ingestion points: Xueqiu timeline data retrieved in crawl_xueqiu_home_timeline_api.py and saved to local Markdown files.
Boundary markers: Absent; the instructions rely on simple line prefix checks (e.g., // or >) which are insufficient to prevent adversarial instructions embedded in the content.
Capability inventory: The agent has the ability to execute system commands, write to the filesystem, and spawn multiple subagents.
Sanitization: Only basic HTML tag removal is performed, leaving the content unvetted for malicious prompt segments that could influence subagent behavior.

crawl-xueqiu-my-timeline