The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on an external binary, lark-cli, to fetch document content and upload generated media to Feishu/Lark. It executes these commands via Python's subprocess.run using argument lists, which prevents shell injection.
[EXTERNAL_DOWNLOADS]: The skill connects to the OpenRouter API (openrouter.ai) to perform text-based theme analysis and image generation using Gemini models. It also interacts with WeChat's material library API when the upload feature is enabled.
[PROMPT_INJECTION]: As the skill processes untrusted article content from Markdown files or Feishu documents to generate AI prompts, it is technically exposed to indirect prompt injection. However, the implementation requires an explicit --confirm-generate flag from the user after they have reviewed the final generated prompt, which serves as a robust human-in-the-loop mitigation.
[DATA_EXFILTRATION]: The skill sends article excerpts to OpenRouter for processing and uploads generated images to WeChat and Feishu. These data flows are transparent and are the primary intended functions of the skill.
[CREDENTIALS_UNSAFE]: The skill manages sensitive API keys (OpenRouter and WeChat) via environment variables and local .env files in the user's home directory (~/.hermes/.env), following standard security practices for secret management.

article-to-wechat-cover