Recover gracefully when default agent setup or launch fails

[baseman70]

Summary

When setting up or switching coding agents via omarchy-default-agent --install <agent> (or selecting an uninstalled agent in the Omarchy menu), the script writes the selection to ~/.config/omarchy/defaults/agent and executes omarchy-agent --inline inside a floating terminal for the initial onboarding session.

If the agent's authentication or startup fails (e.g. an expired token, missing credentials, deprecated OAuth endpoints, or a user aborting setup with Ctrl+C), the broken agent previously remained permanently set as the default, and the terminal would abruptly close with no graceful recourse.

Solution

1. State Preservation: Capture previous_agent from ~/.config/omarchy/defaults/agent before updating the selection.
2. Failure Recovery: Trap INT during inline execution and check the exit code of omarchy-agent --inline. If it fails:
   • Revert defaults/agent to previous_agent (or remove the file if no default was previously configured).
   • Print Could not launch <Agent> to stderr.
   • Prompt the user with gum confirm "Would you like to select a different agent?".
   • If confirmed, summon the agent picker (exec omarchy-menu summon setup.default.agent).
   • If declined, exit with status 1 with the clean reverted state preserved.
3. Unit Tests:
   • Added assertions to test/shell.d/default-agent-test.sh covering:
     • Unset state reversion and menu reopening on confirmation.
     • Previous state restoration (pi) on confirmation.
     • User cancellation / SIGINT (Ctrl+C / exit code 130) state recovery.
     • Non-zero exit code (1) when alternate selection is declined.
     • Terminal clear-screen sequence (\033[2J\033[3J\033[H) verification before launch attempt.
     • Native package-managed agent failure recovery (OpenClaw).
   • Verified that all 50 unit tests in test/shell.d/default-agent-test.sh and 108 tests in ./test/cli pass.