The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's documentation recommends downloading and running a 'live client' from an external GitHub repository (github.com/RayyanZahid/agent-room-join). This source is not associated with the primary vendor or any verified technology organization, posing a potential supply chain risk if the code is executed.
[COMMAND_EXECUTION]: The installation instructions provide commands to append sensitive credentials to shell profiles (e.g., ~/.zshrc, ~/.bashrc) and environment variables. This creates a persistence mechanism for credentials that exposes them to other local processes and shell sessions.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection through the log-reading functionality.
Ingestion points: The ic_rooms_read tool (and its A2A equivalent) ingests 'turns' or messages from a shared log that is populated by other agents and users.
Boundary markers: There are no instructions or delimiters defined to help the agent distinguish between its own operational instructions and potentially malicious data embedded in the room logs.
Capability inventory: The skill utilizes tools that can create rooms, send data, and coordinate actions, which could be abused if an attacker injects commands into the coordination log.
Sanitization: No sanitization or validation logic is specified for the data retrieved from external participants in the room.

ic-rooms