The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests prospect replies via the list_pending_replies tool, which constitutes external data that could potentially contain malicious instructions.
Ingestion points: Prospect message content returned by the list_pending_replies tool.
Boundary markers: None explicitly defined in the prompt instructions to isolate external text.
Capability inventory: The decide_reply tool can trigger external communication (sending messages, booking links, invites) and state changes (permanent opt-outs).
Sanitization: No programmatic sanitization is mentioned; however, the skill mandates a human-in-the-loop review process.
[SAFE]: The skill follows security best practices by explicitly instructing the agent never to perform sensitive or permanent actions (send, invite, booking_link, opt_out) without the user's explicit confirmation, effectively mitigating risks associated with untrusted input.

handle-replies