The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill programmatically retrieves sensitive authentication tokens from local configuration files. Specifically, zhuoran_selfie.py reads FEISHU_APP_ID and FEISHU_APP_SECRET from the ~/.openclaw/.env file, and generate_portrait_base.py accesses environment variables stored in ~/.openclaw/config/main.env.
[COMMAND_EXECUTION]: The skill contains logic for executing high-risk workflows and unsanitized system calls. The file generate_undress_video.py is dedicated to generating sexually suggestive video content ('undress video') using prompts such as 'beautiful young woman slowly removing her wet camisole top'. Furthermore, scripts/clawra-selfie.ts uses child_process.exec to run commands with parameters like channel and message that are not sanitized, which could lead to command injection.
[PROMPT_INJECTION]: The file templates/soul-injection.md contains instructions designed to override the AI agent's default operational guidelines. It defines a persona and specific keyword triggers that force the agent to prioritize the selfie generation skill over standard safety or behavioral constraints.
[EXTERNAL_DOWNLOADS]: The skill performs multiple network operations to upload local media and download generated content from third-party services, including api.wavespeed.ai and fal.run, which are outside the scope of recognized trusted providers.

zhuoran-selfie