The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill designs a system to process user voice input (via the query slot) which is subsequently handled by an LLM backend. This creates a surface for indirect prompt injection where a user might attempt to override agent instructions via spoken commands.
Ingestion points: Handled in references/detailed-guide.md through the ChatIntent logic.
Boundary markers: The provided templates do not include specific delimiters for the LLM prompt, though the system-level separation between the Alexa NLU and the Lambda logic provides some structural isolation.
Capability inventory: The backend code has permissions to interact with DynamoDB, Polly, and external APIs (Anthropic).
Sanitization: The skill demonstrates safe practices by using html.escape on user input before inclusion in SSML responses to prevent control character injection.
[EXTERNAL_DOWNLOADS]: The documentation includes standard setup procedures requiring the installation of official command-line interfaces from well-known service providers.
Evidence: Instructions in references/detailed-guide.md to install ask-cli via npm and awscli via pip. These are legitimate tools for the Amazon Alexa and AWS ecosystems.

amazon-alexa