Allow prepending shell commands for each executeCommand tool use

It can be useful when users can set some shell commands per agent and agents automatically execute those commands before executing a shell command using executeCommand tool.

Possible use cases:

* Inject AWS IAM credentials from environment variables

If injecting environment variables is the only use case, we should add a dedicated interface for that instead of this generic solution.

Start by tracing the executeCommand tool use and the agent configuration path to determine where per-agent shell commands are defined and run. Compare the generic prepending proposal with a dedicated environment-variable interface, then verify that the selected behavior runs configured commands before every executeCommand invocation.

Written by the indexing model from the issue text.