The Agent Skills Directory

[SAFE]: The skill design prioritizes security by implementing a mandatory human-in-the-loop policy, where the agent is prohibited from sending messages without explicit user confirmation (SKILL.md).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted buyer messages via the message_history tool, creating an attack surface for indirect prompt injection (SKILL.md). This risk is mitigated by a multi-step verification process that includes scanning drafts for risk keywords and requiring manual approval before calling the message_send tool. Ingestion points: Buyer conversation history retrieved in SKILL.md. Boundary markers: No explicit delimiter markers are used for buyer input. Capability inventory: Access to sending marketplace messages via message_send. Sanitization: Automated keyword scanning through the risk-guard reference and manual user review before transmission (SKILL.md).

goofish-reply-buyer