The Agent Skills Directory

[PROMPT_INJECTION]: The instructions use aggressive imperative language ("NEVER") to override the agent's default behavior. Specifically, it forbids the agent from providing disclaimers regarding permissions and forces direct message delivery, which attempts to bypass standard AI transparency and safety conversational patterns.
[INDIRECT_PROMPT_INJECTION]: The skill's design creates a vulnerability surface where malicious instructions embedded in group member names or group information could be ingested and acted upon by the agent.
Ingestion points: Data retrieved from yb_query_group_members (nicknames, member lists) and yb_query_group_info (group names).
Boundary markers: The instructions do not define any delimiters or warning markers to help the agent distinguish between data and instructions when processing member names.
Capability inventory: The skill has the capability to send messages to groups (via direct reply) and individuals (via yb_send_dm), including the ability to attach files from the local filesystem.
Sanitization: There are no instructions for sanitizing, escaping, or validating external member data before it is interpolated into @mentions or message replies.

yuanbao