Allow per-session model and effort overrides from `claude agents` dispatch input

Feature request

Please support per-session model and effort overrides when dispatching a new background session from the claude agents UI.

Today, claude agents can be launched with defaults such as:

claude agents --model opus --effort high

And a background task can be started from the shell with:

claude --bg --model sonnet --effort medium "..."

However, once inside the claude agents UI, the dispatch input does not appear to provide a direct way to choose --model or --effort for only the next session.

Why this matters

Different background tasks need different cost/performance tradeoffs:

• Simple refactors or typo fixes may only need sonnet with low or medium effort.
• Architecture reviews may need opus with high or max effort.
• Debugging, review, and implementation tasks often benefit from different model/effort settings.

Current workarounds

The current alternatives are workable but indirect:

1. Restart claude agents with different default --model / --effort values.
2. Create multiple custom agents with fixed model and effort frontmatter.
3. Attach to the session after dispatch and manually run /model or /effort.

Proposed behavior

Allow model and effort to be specified directly in the dispatch input for a newly created background session.

Possible syntax examples:

@repo --model opus --effort high investigate the root cause of this flaky test

/model opus /effort high review the current PR for architecture risks

Any syntax that fits existing Claude Code UI conventions would be fine.

Expected result

A user should be able to dispatch a new background session from claude agents and override the model and effort for that specific session without leaving agent view, restarting claude agents, or relying on preconfigured custom agents.

This would make agent view much more flexible for users running multiple concurrent tasks with different complexity and cost requirements.