The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a platform for processing untrusted end-user data from webhooks, including caller IDs, DTMF digits, and message content, which creates a potential surface for indirect prompt injection. The instructions include explicit security guidance for the agent to validate this data and avoid direct interpolation into prompts or shell commands.
Ingestion points: Webhook payloads and callback events (call.incoming, call.menu, call.answered) described in SKILL.md.
Boundary markers: The skill does not provide specific delimiters for user data in its examples but contains strong textual instructions to treat payloads as untrusted.
Capability inventory: The skill utilizes the Sinch Functions SDK, providing capabilities to place calls, send SMS/WhatsApp messages, and access local storage or databases via the FunctionContext.
Sanitization: The instructions explicitly mandate validation and sanitization of user-controlled fields before use.
[EXTERNAL_DOWNLOADS]: The skill documents the installation of the official Sinch CLI via @sinch/cli and refers to the @sinch/functions-runtime package. These are standard, legitimate resources provided by the vendor for the platform's operation.
[COMMAND_EXECUTION]: The skill provides instructions for using the sinch command-line tool for authentication (sinch auth login), project initialization, development, and deployment. These commands are standard for the serverless development workflow described.

sinch-functions