TLDR

Stabilize the useResumeCommand() callback dependencies so slash-command argument completion no longer re-runs on every render. This fixes the interactive CLI crash where typing /model could get stuck on "Loading suggestions..." and repeatedly throw Maximum update depth exceeded.

Screenshots / Video Demo

N/A — no permanent UI change. This PR fixes an interactive re-render loop in slash-command completion.

Dive Deeper

User-visible behavior

In the interactive CLI, typing a slash command that enters argument completion mode could trigger an infinite render loop. A concrete reproduction was:

1. Type /model
2. Type a trailing space
3. Observe the suggestions area stay on Loading suggestions...
4. The app repeatedly logs Maximum update depth exceeded

This was easiest to hit when the command's completion() returned null or an empty array, because the completion flow kept re-running while also writing completion state back into React.

Root cause

The loop was not caused by modelCommand itself. The actual instability came from useResumeCommand().

useResumeCommand() memoized handleResume() with the entire historyManager object in its dependency list. useHistory() returns a fresh object each render, even though its methods are individually stable. That meant:

• handleResume() got recreated every render
• slashCommandActions in AppContainer changed every render because it depends on handleResume
• commandContext in slashCommandProcessor changed every render because it depends on actions
• useSlashCompletion() treats commandContext as an effect dependency, so argument completion re-fired every render
• the completion effect kept setting loading/suggestion state, producing a render-feedback loop

So /model merely exposed the bug because it enters slash argument completion and its completion() path was invoked over and over.

Fix

This PR removes the unstable object dependency from useResumeCommand():

• derive hasHistoryManager as a boolean guard
• destructure clearItems and loadHistory from historyManager
• depend on those stable callbacks instead of the whole historyManager object
• use optional calls when resetting and loading history

This keeps handleResume() referentially stable across normal renders, which in turn keeps slashCommandActions, commandContext, and slash completion stable.

Reviewer Test Plan

1. Start the interactive CLI.
2. Type /model and then a space.
3. Confirm the app does not enter a render loop.
4. Confirm the UI does not spam Maximum update depth exceeded.
5. Confirm the suggestions area exits loading normally when no suggestions are returned.
6. Sanity check /resume still works and restores prior session history.

Good extra cases:

• /model --
• /model --fast
• any other slash command with argument completion that can return no results

Testing Matrix

Linked issues / bugs

Related to the interactive slash-command completion render loop triggered by /model when command argument completion returned no suggestions.