The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a ReAct-based agent framework that processes natural language queries to trigger tools and AWS service calls, creating a surface for indirect prompt injection. Ingestion points: User queries are ingested through Agent() and MultiAgentOrchestrator.run() calls. Boundary markers: The provided code examples lack delimiters or specific instructions to isolate user input from system directives. Capability inventory: The agents are capable of performing database searches and invoking AWS CloudWatch metrics via boto3. Sanitization: No input validation or output sanitization logic is demonstrated in the implementation examples.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions to install third-party libraries strands-agents, strands-agents-tools, and @strands-agents/sdk. It also references official sample code from the awslabs/amazon-bedrock-agentcore-samples GitHub repository, which is a well-known source for AWS development resources.

aws-strands