Consolidates

• feat: Promote v18 improvements to unleashed-g #22 — feat: Promote v18 improvements to unleashed-g
• Next Gemini version (g-18 -> production) #27 — Next Gemini version (g-18 → production)

Context

The Gemini variant (unleashed-g-*.py) currently lacks v18 features (mirror, friction, resize, model pause). Promoting these features is non-trivial because Gemini's TUI differs from Claude Code's.

Depends on

• question: Is the Gemini variant needed? Unresolved blockers #19 — Is the Gemini variant needed? If the answer is "no" or "not right now," close this issue too. The Gemini variant has unresolved blockers (different TUI, different permission patterns, approval echo bug).

If we proceed

1. Port mirror + friction logger (generic, should work)
2. Adapt permission patterns for Gemini's TUI
3. Test approval flow (the echo/double-approval bug from g-18)
4. Add transcript save (trivial — same as c-20)
5. Reduce approval timing to 0.1s

If we sunset

1. Keep existing unleashed-g-18.py as-is for occasional use
2. Stop investing in Gemini-specific features
3. Close question: Is the Gemini variant needed? Unresolved blockers #19, feat: Promote v18 improvements to unleashed-g #22, Next Gemini version (g-18 -> production) #27