The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a voice-to-text chat interface that accepts arbitrary user queries via the Alexa ChatIntent slot ({query}). This input is directly incorporated into the LLM message history, creating a surface for indirect prompt injection attacks.
Ingestion points: SKILL.md (the chat_handler function in Python processes the slots["query"].value variable).
Boundary markers: Absent; user voice input is interpolated into the message history without delimiters or instructions to the model to ignore potential commands hidden within the input.
Capability inventory: The Lambda function has network access to the Anthropic API, read/write access to DynamoDB (persistence), and access to Amazon Polly (voice synthesis).
Sanitization: Absent; while the response is truncated to prevent timeouts, no validation, escaping, or filtering is applied to the incoming user query before it is sent to the LLM backend.
[EXTERNAL_DOWNLOADS]: The documentation instructs users to install developer tools and libraries from well-known official registries.
Evidence: npm install -g ask-cli, pip install awscli, and the dependencies defined in the Requirements.Txt section (ask-sdk-core, ask-sdk-dynamodb-persistence-adapter, anthropic, and boto3). These are legitimate libraries from recognized organizations (Amazon and Anthropic).
[COMMAND_EXECUTION]: The skill provides instructions for executing shell commands to configure the development environment and deploy cloud resources.
Evidence: Shell blocks containing ask configure, aws configure, ask deploy, and aws lambda create-function. These commands are standard for Alexa skill development and AWS resource management.

amazon-alexa