The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides templates for executing shell commands via the Bash tool that interpolate user-controlled variables such as $SCRIPT, $AVATAR_ID, and $VOICE_ID directly into a curl payload. This structure poses a risk of command injection if the agent or user provides input that contains shell-special characters, which could be used to terminate the intended command and execute unauthorized shell operations.
[DATA_EXFILTRATION]: The skill initiates network requests to https://floe-production.up.railway.app for polling render status. This involves sending an API key ($FLOE_API_KEY) and video identifiers to an external, third-party service that is not part of the primary platform or standard whitelisted domains.
[DATA_EXFILTRATION]: Instructions within the skill involve reading from a local configuration file located at ~/.gsai/openclaw-routing.json. Accessing configuration files within the user's home directory to determine routing for human delegation represents a risk of exposing local setup details or metadata.

c-heygen