Feature request: Add /compact command (manual context compaction)

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

Codex 260205.1301

What subscription do you have?

Pro

What issue are you seeing?

The Codex CLI has a /compact command that allows users to manually
compact the conversation context at any point. The Codex App only
has autocompact, which triggers automatically when the context window
fills up.

This is a meaningful gap. Manual /compact is useful not just for
managing token limits, but as a proactive measure to reduce model
hallucinations. By compacting early and often, the model works with
a cleaner, more focused context — which in practice leads to more
accurate and grounded responses, especially in long coding sessions.

In the CLI I regularly run /compact after completing a subtask or
before switching focus within a session. This keeps the model
"on track" and noticeably reduces drift and confabulation. The
autocompact in the App only kicks in when context is nearly full,
which is too late — by that point the model has already been
reasoning over a bloated, noisy context for many turns.

Expected: A /compact command (or UI button) in the Codex App that
lets users manually trigger context compaction at will, matching
the CLI behavior.

Current: Only autocompact exists, with no manual override.

What steps can reproduce the bug?

This is a feature request, not a bug.

Steps showing the missing functionality:

1. Open the Codex App
2. Start a long coding session (10+ turns)
3. Try to manually compact context — no command or button available
4. Context continues to grow until autocompact eventually triggers

In contrast, Codex CLI workflow:

1. Open Codex CLI
2. Work on a subtask for several turns
3. Type /compact to summarize and reset context
4. Continue with clean, focused context
5. Repeat as needed — model stays grounded throughout

Use case: Frequent manual compaction helps reduce hallucinations
by keeping the working context small and relevant, rather than
waiting for automatic compaction at near-full context window.

What is the expected behavior?

_No response_

Additional information

_No response_