Q CLI running in the secure-agent environment (container) often does not realize it has a keyring available (needs to be told to read .amazonq/mcp.json and /opt/mcp-servers/perplexity-server.py to see how they work) and often when it writes code invoking Bedrock it gets the encoding for uploaded files and the model names wrong; it needs an example, and more instruction about how to find the current useful Bedrock models (and to include the "us." preface to invoke the inference profile).
Q CLI running in the secure-agent environment (container) often does not realize it has a keyring available (needs to be told to read .amazonq/mcp.json and /opt/mcp-servers/perplexity-server.py to see how they work) and often when it writes code invoking Bedrock it gets the encoding for uploaded files and the model names wrong; it needs an example, and more instruction about how to find the current useful Bedrock models (and to include the "us." preface to invoke the inference profile).