The Agent Skills Directory

[COMMAND_EXECUTION]: Multiple scripts execute external commands and utilities. douyin_download.py runs playwright install chromium to set up the browser environment. xiaoyuzhou_opencli.py executes the opencli command for podcast interaction. wechat_mp_local.py calls the macOS open utility to handle login pages.
[EXTERNAL_DOWNLOADS]: The douyin_download.py script automatically triggers the download and installation of the Chromium browser binary via Playwright. While Playwright is a well-known tool from Microsoft, this involves the automated retrieval of executable binaries from a remote source.
[REMOTE_CODE_EXECUTION]: xiaoyuzhou_stepfun.py executes a Python script located at a path determined by the YICHEN_STEP_ASR_SCRIPT environment variable or a relative path to a sibling skill. This allows for the runtime execution of external code logic.
[CREDENTIALS_UNSAFE]: The skill accesses sensitive authentication information from the environment and local file system. xiaohongshu_fetch.py reads XHS_COOKIE from environment variables, and wechat_mp_local.py retrieves authentication keys from the local application support directory for the WeChat exporter.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted content from various social media platforms. 1. Ingestion points: Untrusted data enters the agent context via URLs processed by x_known_url.py, xiaohongshu_fetch.py, and douyin_download.py. 2. Boundary markers: The skill does not implement delimiters or 'ignore embedded instructions' warnings for the external content it processes. 3. Capability inventory: The skill has the ability to write to the file system, execute shell commands, and perform network operations. 4. Sanitization: Only basic HTML unescaping and tag removal are performed, which does not prevent instructional content from influencing the agent.

yichen-content-archive