[Behavior Mgmt] Gemini CLI should prefer using automated linting and formatting tools vs. manually making changes

[gundermanc]

What would you like to be added?

Gemini CLI should prefer utilizing automated refactorings, when possible, to accomplish its goal.

This means, for example, that it should utilize eslint --fix prior to trying to fix linter errors itself and prettier instead of formatting files manually, if applicable.

Why is this needed?

• Completion time -- automated compiler-based tools are much faster and more effective at performing automated refactors and reformatting than an agent because they are deterministic, parallelizable, and require significantly less wall clock time due to the lack of an LLM call.

• Limited Context Window -- As described in Gemini CLI often performs badly when the context window gets large #9791, asking GCLI to fix a linter error in a moderately large project commonly causes it to run into context rot, confusion, and/or looping. I experienced this first hand with Disallow redundant typecasts. #15030. It was significantly faster doing it "manually" with eslint . --fix.

Additional context

I think this is something fixable via prompt changes. I'd scope it to include:

• Creation and initialization of projects -- should ideally be guided to use ecosystem relevant tools.
• Linting, reformatting -- should utilize --fix and --write or the equivalent.

[gundermanc]

Scope of work:

• System prompt changes.
• Behavioral evals for a few specific cases.