System prompt conflicting with my desired workflow

What version of Kimi Code CLI is running?

Kimi Code v0.12.0

Which open platform/subscription were you using?

the api key one

Which model were you using?

k2.7-coding

What platform is your computer?

debian

What issue are you seeing?

I have strict guidelines (mentioned in multiple places) about not modifying existing tests unless that's the task (ie. when implementing features or changes, create new tests, don't modify existing ones).

The agent included a "modify existing tests" step in a runfile, and when I quizzed it, it said:

● Yes, you're right. The kimi-code harness system prompt does contain a testing guideline
in the "General Guidelines for Coding" section, roughly:

│ "If user mentioned any failed tests, you should make sure they pass after the
│ changes."

It would be good if there were no guidelines like this in the system prompt which got in the way of the agent behaving the way I want it to.

What steps can reproduce the bug?

Try to stop the agent from ever modifying existing tests when implementing new features and bug fixes

What is the expected behavior?

Follow my instructions

Additional information

No response

No file or test is named. Start by locating the Kimi Code harness system prompt and its “General Guidelines for Coding” section, then reproduce the conflict with the stated test-modification instruction. Done means the agent follows the user's instruction without the system prompt overriding it.