Agent tool calls should expose inspectable arguments and results, not only summaries

What version of the Codex App are you using (From “About Codex” dialog)?

26.901.6511.0

What subscription do you have?

pro

What platform is your computer?

windows x64

What issue are you seeing?

Codex should provide a user-accessible way to inspect the actual tool call arguments sent by the model and the tool results returned back to the model.

Recent UI changes increasingly reduce tool activity to summaries such as command counts, generic tool names, or high-level status messages.

For an agent, however, tool calls are not merely implementation details. They are a core part of the model's observable behavior.

A user debugging or evaluating an agent needs to be able to answer:

What tool did the model choose?
What exact arguments did the model generate?
What did the tool actually return?
What information was returned to the model before its next decision?

If these details are available internally to the Codex harness but omitted from the user-facing interface, we lose an important layer of agent observability.

What steps can reproduce the bug?

try it

What is the expected behavior?

It should be dismissible by the user rather than being automatically closed in the background

Additional information

No response

Start by locating the user-facing tool activity interface and the boundary where tool-call details are passed to the Codex App. Define how users can inspect the exact arguments and returned results, and clarify the expected dismissible behavior before implementation; the issue names no files or tests.