The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes untrusted data from Discord, which could contain instructions intended to manipulate the agent.
Ingestion points: Script functions in discord-admin.sh and discord-ctrl.sh fetch message history, audit logs, and search results from Discord channels and guilds.
Boundary markers: There are no explicit delimiters or instructions to treat data from the Discord API as untrusted content.
Capability inventory: The skill provides broad administrative powers, such as kicking/banning members and modifying server permissions, which could be abused if the agent is compromised via injection.
Sanitization: Content retrieved from Discord is not sanitized or validated before being presented to the agent.
[COMMAND_EXECUTION]: The skill relies on local shell scripts (discord-admin.sh and discord-ctrl.sh) to perform its primary functions. These scripts execute curl to interact with the Discord API and jq for data processing.
[METADATA_POISONING]: The SKILL.md documentation claims several advanced capabilities that are missing from the actual shell script implementations.
Discrepancies: Commands like msg-send-file, member-avatar-set, and several granular automod-* operations are documented but not implemented in discord-admin.sh or discord-ctrl.sh.

discord-server-ctrl