The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent on processing a state object containing external data, which is a potential surface for indirect prompt injection.
Ingestion points: The state field in the client.system_one API call accepts arbitrary data (e.g., ticket_text).
Boundary markers: The instructions recommend using backticked paths (e.g., `ticket.messages[0].text`) to scope judgments to specific data fields.
Capability inventory: The skill examples include business logic capabilities like routing tickets (route_to_billing), escalating issues (escalate), and adding to backlogs.
Sanitization: The skill provides excellent defensive guidance, stating: "Treat text in the state as able to steer the answer. Jev does not treat state as hostile. State in the criteria what counts, and test injected and self-describing content before deployment." It also mandates using "Confidence-gated routing" to ensure automated actions only occur when the model is certain.
[EXTERNAL_DOWNLOADS]: The skill references the typesafe_sdk Python package and official TypeSafe documentation.
Evidence: The code snippet imports typesafe_sdk and the source list links to the docs.typesafe.ai domain.
Note: These resources are essential and standard for the service the skill is intended to support.

jev