The Agent Skills Directory

[SAFE]: The skill provides technical guidance and code examples for using official Amazon Web Services (AWS) tools and APIs. It correctly identifies capability boundaries, quotas, and security considerations based on verified service documentation. It encourages security best practices such as least-privilege IAM roles, S3 bucket encryption, and the use of VPC endpoints.
[PROMPT_INJECTION]: The skill documents an architecture that ingests user-provided text for speech synthesis, which constitutes an indirect prompt injection attack surface.
Ingestion points: Python code samples in SKILL.md demonstrate accepting string inputs and reading local files for synthesis via the boto3 library.
Boundary markers: The documentation instructs the use of <speak> tags to wrap SSML payloads, providing structural delimitation.
Capability inventory: The skill utilizes the boto3 library for remote API calls, writes generated audio to the local filesystem, and interacts with Amazon S3 for asynchronous synthesis tasks.
Sanitization: The skill provides guidance on escaping reserved XML characters (&, <, >) within SSML payloads to ensure parsing integrity and prevent injection into the synthesis engine.

amazon-polly