The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/fetch_wechat_article.py is configured to automatically clone a third-party repository from https://github.com/jo-inc/camofox-browser.git into the local ~/.hermes/camofox-browser directory if it is not found during execution.\n- [REMOTE_CODE_EXECUTION]: After cloning the external repository, the skill runs npm install and npm start. This allows code from an unverified remote source to execute with local user privileges, presenting a supply chain attack surface.\n- [COMMAND_EXECUTION]: Multiple scripts, including scripts/run.py and scripts/publish_wechat_article_to_feishu.py, use subprocess.run to execute system commands such as git, npm, and lark-cli, incorporating user-provided article URLs in the process.\n- [DATA_EXFILTRATION]: The skill utilizes lark-cli to transmit scraped article data to external Feishu servers. It also reads and writes to the local filesystem (home directory) for logging, configuration, and repository management.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by processing arbitrary data from external WeChat URLs.\n
Ingestion points: fetch_article in scripts/fetch_wechat_article.py (reads data from external article URLs).\n
Boundary markers: Absent; article content is converted to Markdown/JSON without delimiter isolation.\n
Capability inventory: subprocess.run (to execute lark-cli and other tools) and network access via urllib.request.\n
Sanitization: Performs basic regex cleaning of WeChat-specific UI noise but does not validate article content for embedded instructions.

wechat-article-camofox