`! resurrect` is slower on Windows than on WSL

[max-nothacker]

Observation

! resurrect works on Windows across all three CLIs (Claude / Codex / Antigravity), but is noticeably slower than on WSL.

Timing (measured — 10 calls each, full path including the clipboard write)

Path	~per call
WSL (bash -> clip.exe via interop)	~105 ms
Windows Git Bash (Claude, direct)	~135-155 ms
Windows cmd/PowerShell -> resurrect.cmd -> bash.exe -> script (Codex/Antigravity)	~160 ms

Likely cause (to confirm further)

Every call spawns clip.exe (a Windows process) to set the clipboard — the dominant cost everywhere (~100 ms even on WSL, which spends almost all of it waiting on the interop spawn: user+sys ~= 0.1 s of 1.05 s for 10 calls).
Windows is slower on top of that because:

1. MSYS / Git Bash has no native fork() — bash startup + the clip.exe spawn are emulated, costing more CPU than WSL's native Linux fork (Git Bash user+sys ~= 1.1 s vs WSL ~= 0.1 s for 10 calls).
2. Codex/Antigravity add an extra process hop: cmd/PowerShell -> resurrect.cmd -> bash.exe -> script.

It is process-spawn overhead, not the script logic.

Intended approach

1. Measure & confirm (done — see table above).
2. Decide whether to optimize. Candidate fixes (all low value):
   • Drop the extra hop for Codex/Antigravity by making resurrect.cmd a standalone batch (removes the bash.exe spawn) — but duplicates logic (SSoT cost).
   • Background the clipboard write (| clip.exe &) — risky (race: script may exit before clip.exe finishes reading stdin).
   • Accept it: ~150 ms is fine for an interactive helper.

Priority

Low — not important. Filed for tracking.