The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill determines which subsequent skill to load based on user-provided input. While this creates a routing surface, the skill is designed for triage and specifically directs the agent to refer to service-specific implementation skills rather than executing complex commands itself.
Ingestion points: User requests are matched against a routing table to determine the target skill.
Boundary markers: The skill provides clear instructions on scope and limitations, though explicit data delimiters for the user input are not defined in this routing layer.
Capability inventory: The skill utilizes tool calls (aws___retrieve_skill) and file system access (skills/<skill>/SKILL.md) to transition to implementation skills.
Sanitization: The routing logic uses semantic matching against a predefined table of allowed service names, which limits the potential for arbitrary skill loading.
[Security Guidance and Best Practices]: The instructions actively encourage security controls such as encryption in transit (MACsec, IPsec, TLS), logging for incident response, and the use of AWS Secrets Manager instead of hardcoded credentials. These are standard security recommendations for AWS environments.
[Trusted Resource References]: The skill references official AWS documentation and the AWS Well-Architected Framework, which are recognized sources for authoritative cloud architecture guidance.

aws-networking