The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a crisis detection mechanism that processes untrusted user messages and history to trigger system-level interventions. Ingestion points: User message content and conversation history are passed to the assessCrisisLevel function in SKILL.md. Boundary markers: While the skill employs a safetySystemPrompt to guide AI behavior, the code snippets do not implement explicit delimiters to isolate user-provided text within the assessment logic. Capability inventory: Detection of high-risk indicators triggers notifyEmergencyContact (a network-based notification) and logCrisisEvent (a database write operation). Sanitization: The validateResponseSafety function provides basic output filtering for specific unsafe phrases using regular expressions.
[DATA_EXFILTRATION]: The skill includes an emergency contact system that performs network operations to send SMS or push notifications to pre-configured contacts. This functionality is essential for the skill's crisis management purpose and includes explicit instructions to exclude conversation content from the alerts to preserve user privacy.

crisis-response-protocol