The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, such as website chat transcripts and prospect email replies, creating a potential attack surface for indirect prompt injection.
Ingestion points: The agents ingest data from website chat transcripts (agents/sales-conversational-agent-strategist.md), prospect email replies (agents/sales-outbound-strategist.md), and various CRM records.
Boundary markers: The instructions explicitly address these risks by requiring a "test suite" for conversational agents that includes prompt-injection attempts and by mandating human review ("Draft" mode) for actions based on automated reply classification.
Capability inventory: The skill coordinates actions including writing to CRM fields, booking calendars, and sending emails via third-party tools (Drift, Outreach, Salesloft).
Sanitization: The instructions emphasize that LLM-generated classifications are evidence to be verified by a human rather than instructions to be executed autonomously.
[SAFE]: The skill demonstrates best practices in sales methodology and technical evaluation. It includes specific sections on transparency, ethics (e.g., EU AI Act compliance), and data privacy (e.g., demo environment integrity). All external references point to well-known services, government regulations, or established open-source methodology repositories.

sales-enablement