Keep prompted Hermes sessions interactive

Hermes oneshot deliberately exits after answering, which closes the agent terminal. Seed the TUI chat session instead, keep inherited flags after the subcommand for older Hermes parsers, and bind the query as one argument so dash-prefixed prompts remain data.

Co-Authored-By: Codex XHigh <noreply@openai.com>

test/shell.d/default-agent-test.sh

@@ -462,6 +462,7 @@ assert_launch claude claude --permission-mode auto -- "Review this project"
 assert_launch codex codex --approve-for-me -- "Review this project"
 assert_launch crush crush run "Review this project"
 assert_launch grok grok --permission-mode bypassPermissions -- "Review this project"
+assert_launch hermes hermes chat --yolo --tui "--query=Review this project"
 assert_launch agy agy --dangerously-skip-permissions --prompt-interactive "Review this project"
 assert_launch copilot copilot --allow-all --interactive "Review this project"
 pass "agent launcher adapts initial prompts for every supported agent"
@@ -474,6 +475,7 @@ assert_bypass claude claude --permission-mode auto
 assert_bypass codex codex --approve-for-me
 assert_bypass crush crush --yolo
 assert_bypass grok grok --permission-mode bypassPermissions
+assert_bypass hermes hermes --yolo
 assert_bypass agy agy --dangerously-skip-permissions
 assert_bypass copilot copilot --allow-all
 pass "agent launcher skips permission prompts for every supported agent"