Should terminal create support an agent-aware mode?

[zfaustk]

I have a command override configured for Codex through agentCmdOverrides.codex. The override works when Orca launches Codex through its built-in agent entry.

For a fresh agent in an existing worktree, Orca's CLI currently recommends:

orca terminal create --worktree active --command "codex"

That path treats --command literally. It does not apply agentCmdOverrides.codex, agentDefaultArgs.codex, or agentDefaultEnv.codex, so an agent following Orca's own CLI guidance can silently launch the stock command instead of the user's configured Codex setup.

Typing codex directly in a shell should remain literal; Orca should not rewrite arbitrary shell input. The missing piece seems narrower: terminal create has no agent-aware counterpart to worktree create --agent codex for an existing worktree.

Is this intentional, or would an agent-aware terminal-create mode fit the CLI?

This is related to #2284, #9136, and #12930, but does not require custom agents or multiple profiles. It affects an existing built-in agent configuration.

[zfaustk]

After tracing the current launch paths, one narrow option would be:

orca terminal create --worktree active --agent codex

• Make --agent and --command mutually exclusive.
• Resolve the command, default arguments, and environment through the same agent settings used by worktree create --agent.
• Preserve the explicit agent identity and session-startup behavior.
• Keep --command and commands typed directly into a shell literal.

This avoids guessing from the first token of an arbitrary command and stays separate from the broader custom-profile work in #2284, #9136, and #12930.

If this direction fits Orca, would you be open to a focused PR from me? I can cover the CLI schema and validation, plus local and remote terminal-create tests.