The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from customer messages via email, WhatsApp, and WeChat connectors into the tickets base as described in references/support-schema.md. This ingestion surface is mitigated by a mandatory human-in-the-loop review process and a programmatic quality gate (support-qa) defined in app/app/js/support-model.js. The boundary markers consist of explicit status transitions and automated scans for unapproved commitment language. While the skill possesses capabilities to propose sends, refunds, and escalations, these are secured through required human approval and the enforcement of the 'SHIP' verdict before execution markers are recorded by the trusted script.

kelly-support