The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a read action that allows the agent to retrieve message history from Discord channels. This external, untrusted content could contain malicious instructions intended to manipulate the agent's behavior.
Ingestion points: The read action documented in SKILL.md allows the agent to ingest external Discord channel content.
Boundary markers: The skill instructions do not specify any boundary markers or delimiters to differentiate between system instructions and data retrieved from Discord.
Capability inventory: The skill utilizes the message tool to perform various actions including send, react, read, edit, delete, poll, pin, thread-create, search, and set-presence. Notably, the send action supports local file system access via the media parameter.
Sanitization: There are no explicit instructions or mechanisms mentioned for sanitizing or validating the content read from Discord before it is processed by the agent.
[DATA_EXFILTRATION]: The examples for the send action demonstrate support for a media parameter that accepts file:// URIs (e.g., file:///tmp/example.png). This provides a mechanism for the agent to access the local filesystem and transmit the resulting data to a remote Discord channel.

discord