Startup performance

KronuZSH compiles the zsh files it loads and generates the fzf and zoxide initialization once during install. The optimization keeps startup synchronous: the first command gets completion, syntax highlighting, autosuggestions, and the Git prompt, exactly like every command after it.

Benchmark

Measured on an Intel Mac with zsh 5.9.2, inside a Git repository, using 16 iterations of zsh-bench:

zsh-bench --iters 16 --login no --git yes

Configuration	First prompt	First command	Command	Input	Capabilities
Text sources, generated integrations per shell	188.8 ms	209.2 ms	15.3 ms	9.0 ms	all four
Compiled sources and install-time generated integrations	147.3-150.3 ms	167.6-170.4 ms	14.9-15.2 ms	8.6-9.2 ms	all four

Across two final 16-iteration runs, the startup path moved 38.5-41.5 ms earlier: 20.4-22.0% less time to the first prompt and 18.5-19.9% less time to the first command. The steady-state command and input differences are run-to-run noise.

Why there is no deferred plugin loading

zsh-defer was tested against the same configuration after compilation. Deferring fast-syntax-highlighting changed first-prompt latency from 158.0 ms to 153.9 ms, a 4.2 ms shift, but the first command had no syntax highlighting. Deferring completion, autosuggestions, keybindings, aliases, functions, or environment changes creates the same race in a more damaging place.

The shell does not claim work is finished while it is still moving pieces behind the prompt. Compilation removes parsing work. Install-time generation removes process launches. Both preserve the first command's behavior.

This matches the failure mode described in zsh-bench's deferred-initialization analysis: first-prompt timing alone rewards an incomplete shell. Keep the capability fields next to the latency numbers.