The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a handler that captures user voice input from Alexa and directly appends it to the message history sent to the Claude API. This creates a vulnerability surface where adversarial user input could attempt to influence the LLM's behavior.
Ingestion points: User input is captured via the {query} slot in the ChatIntent defined in SKILL.md.
Boundary markers: Absent; the chat_handler in SKILL.md appends the raw query string directly to the messages list.
Capability inventory: The skill uses the anthropic SDK for message generation and boto3 for AWS services (DynamoDB, Polly, Secrets Manager). It does not include capabilities for arbitrary shell execution or unsafe file system writes.
Sanitization: Absent; the provided Python code does not include validation or escaping logic for the user-supplied query string.
[EXTERNAL_DOWNLOADS]: The skill instructions include the installation of official command-line tools and libraries required for the Alexa and AWS environment.
Evidence: Fetches the official ask-cli via npm and the awscli via pip. It also references standard libraries such as ask-sdk-core, anthropic, and boto3 in a requirements file. These are well-known resources from trusted providers.

amazon-alexa-v2