aws-secrets-manager

To prevent secret leakage in AI agent context, use runtime dynamic references with the {{resolve:secretsmanager:...}} syntax. This ensures credentials, API keys, and passwords are resolved at execution time by a wrapper script rather than exposed in prompts or logs.

The asm-exec wrapper substitutes {{resolve:secretsmanager:...}} dynamic references in command arguments and environment variables at runtime. It resolves the secret values during execution without exposing the plaintext to the LLM context.

Yes, using dynamic references for AWS Secrets Manager requires proper IAM permissions to be configured. Additionally, you need either the AWS Secrets Manager Agent available on localhost:2773 or a SigV4-signed MCP endpoint to handle the backend resolution.

Yes, you can use SigV4-signed MCP endpoints as a flexible backend for secret resolution instead of the AWS Secrets Manager Agent (SMA). Both backends support resolving dynamic references securely at runtime without exposing plaintext values.

The runtime resolution approach requires that the agent strictly uses dynamic references and that proper backend infrastructure like SMA or SigV4-signed MCP endpoints is configured. It relies on asm-exec integration and will not protect secrets if plaintext values are passed directly into the context.

You should use dynamic references whenever your AI agent handles credentials, API keys, tokens, or passwords. Passing secrets directly risks leaking plaintext values into logs, prompts, or downstream tool calls, whereas runtime resolution keeps them out of the LLM context entirely.