The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents functions such as aiChat() and aiModel() which are designed to ingest user-supplied prompts and configuration parameters. While these functions are standard for AI integration, they represent an attack surface where untrusted data could influence agent behavior if not properly delimited or sanitized by the implementing developer.
Ingestion points: params and prompt arguments in aiModel(), aiChat(), and aiService() functions.
Boundary markers: The documentation does not explicitly mandate specific delimiters for user input, though it provides a structured API for parameter passing.
Capability inventory: The documented functions facilitate network operations to various AI provider endpoints (OpenAI, Anthropic, AWS Bedrock, etc.).
Sanitization: The skill focuses on configuration and does not describe specific sanitization logic for the content of the prompts themselves.
[CREDENTIALS_UNSAFE]: The documentation discusses API key management and AWS credential resolution. It identifies standard environment variables (e.g., OPENAI_API_KEY, AWS_ACCESS_KEY_ID) and provides placeholders (e.g., sk-..., abc123) for illustrative purposes. The skill explicitly recommends using environment variables over hardcoding secrets in source code, which is a recognized security best practice.

bx-ai-models