Zero-Rust-2-ZIG

Rust → Zig port of ZeroClaw with side-by-side eval and benchmark harness.

The Rust workspace under rust/ is the reference implementation. The Zig workspace under zig/ is the new implementation. evals/ holds language-agnostic golden fixtures; benches/ holds the cross-language benchmark runner.

Layout

rust/         extracted ZeroClaw Rust workspace (reference, READ-MOSTLY)
zig/          new Zig workspace
eval-tools/   standalone Rust binaries that produce golden fixture outputs
evals/        fixtures + Python driver for byte-equal parity tests
benches/      cross-language benchmark runner + normalized results
scripts/      one-shot setup helpers
docs/         porting-notes.md, decisions.md (running logs)

Quick start

# 1. Toolchains
. "$HOME/.cargo/env"
rustc --version    # 1.87.0
zig version        # 0.14.1

# 2. Rust baseline
cd rust && cargo bench --bench agent_benchmarks
cd .. && benches/runner/run_rust.sh > benches/results/baseline-rust-$(date +%F).json

# 3. Zig pilot (after pilot lands)
cd zig && zig build test
zig build bench -Doptimize=ReleaseFast > ../benches/results/raw-zig-$(date +%F).json

# 4. Eval parity
python3 evals/driver/run_evals.py --rust eval-tools/target/release --zig zig/zig-out/bin

# 5. Comparison report
python3 benches/runner/compare.py \
  benches/results/baseline-rust-$(date +%F).json \
  benches/results/raw-zig-$(date +%F).json \
  > benches/results/reports/$(date +%F)-comparison.md

See the plan for context, scope, and acceptance gates.