The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses curl and jq to interact with Discord's REST API. Commands are executed locally to manage messages, channels, and server metadata.
[DATA_EXFILTRATION]: The skill sends data to discord.com, which is a well-known and legitimate service. This includes sending message content, creating webhooks, and modifying server configurations as part of its primary purpose.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from an external source.
Ingestion points: Untrusted data enters the agent context through the 'Get Channel Messages' and 'Get Specific Message' functions in SKILL.md which fetch external content via API.
Boundary markers: The instructions lack delimiters or explicit warnings to the agent to ignore instructions embedded within the fetched message content.
Capability inventory: The skill provides significant capabilities in SKILL.md including sending/deleting messages, creating webhooks, and creating channels which could be abused if the agent obeys instructions found in fetched messages.
Sanitization: There is no evidence of sanitization or filtering applied to the content retrieved from Discord before it is processed by the agent.

discord