The Agent Skills Directory

[Data Separation Guardrails]: The skill explicitly defines emails, transcripts, and external documents as untrusted content. It instructs the agent to treat them strictly as data/evidence and never as instructions, reducing the risk of indirect prompt injection.
[Least Privilege Architecture]: Write actions or modifications to the CRM are explicitly restricted; the skill is defined as read-only and hands off any backfill tasks to a separate specific tool (update-opportunity) rather than executing changes autonomously.
[Safe Fallback Design]: The skill handles missing tool connectors gracefully by relying entirely on user-provided or uploaded files rather than attempting to download arbitrary packages or execute external code.

win-loss-review