Who picks which tool to use

Which tool gets used for a given step is decided entirely by the model itself, never by you naming one explicitly in your message. You describe the outcome you want; tCode works out the mechanism.

Asking "what does this function do" leads naturally to read_file or grep depending on what is already known; asking "run the linter" leads to bash. This mirrors how you would actually talk to a colleague, telling them what you need checked, not which specific command to type to check it.

This division of labor is part of what keeps prompts natural rather than requiring you to learn a tool-specific vocabulary before you can use tCode effectively.

It also means the same underlying request can lead to a different tool choice depending on what is already known in the conversation. "Where is this configured" might trigger a grep the first time it comes up, but skip straight to read_file on a follow-up once the relevant file has already been identified earlier in the same turn.

You are never required to name a tool explicitly to get it used, and doing so does not change anything about how the request is handled; tCode reads intent from what you are actually asking for, not from tool names mentioned in the prompt.