The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an interface for ingesting untrusted external data from the public telephony network (voice transcripts and SMS messages) into the agent context.
Ingestion points: Webhook event payloads (e.g., payload['data']['transcript'] and recentHistory) originating from external callers.
Boundary markers: The provided implementation examples do not include explicit prompt delimiters or instructions to ignore embedded commands in transcripts.
Capability inventory: The skill grants the agent the ability to initiate outbound phone calls, send SMS messages, and perform account management tasks such as buying or releasing numbers.
Sanitization: The documentation does not specify sanitization or filtering protocols for external content before interpolation into LLM prompts.
[DATA_EXFILTRATION]: The skill instructions handle sensitive API credentials (sk_live_...). It mitigates risk by providing explicit security rules that forbid the agent from transmitting the API key to any destination other than the official api.agentphone.to domain.
[COMMAND_EXECUTION]: The skill provides numerous curl command templates to facilitate API interaction. These are standard for developer-oriented skills and are used to interact with the service infrastructure owned by the vendor.

agentphone