The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to process untrusted data from various external channels such as email and web forms.
Ingestion points: Untrusted data enters the agent context through Gmail, Microsoft 365, HubSpot, and form notifications.
Boundary markers: The skill instructions explicitly treat incoming content as data about the prospect rather than instructions, and it references a shared protocol for handling untrusted content.
Capability inventory: The agent has the ability to write to HubSpot and post to internal team channels like Slack or RingEx Chat.
Sanitization: The skill contains logic to detect and hold inquiries that address the assistant directly or request sensitive information like credentials or payments.
Automated Data Routing: Inquiries categorized as 'hot' are automatically routed to pre-approved internal channels.
This feature enables rapid team response by automatically sharing lead details and inquiry text to a shared workspace after an initial one-time approval during setup.
Human-in-the-Loop Safeguards: A core safety feature of the skill is that no drafted replies are sent to external prospects without manual human approval.
This provides a critical check against automated social engineering or unauthorized messaging.
Spam and Reputation Protection: The skill implements a 'Volume Guard' to identify and flag suspicious surges in inbound traffic, such as form loops or bot activity.
This prevents the agent from inadvertently processing or replying to bulk spam, which helps protect the sender's email reputation.

speed-to-lead