AGENTS.md

AGENTS.md

Before Changes

• Performance/optimization/resource work: if cargo run --bin xtask -- mock-deploy lacks claimed-gain metric, update it; capture baseline before edits; claim gains only with before/after same target/config.

Verify

• MUST pass: cargo fmt --all -- --check; cargo clippy --all-targets -- -D warnings; cargo test.
• Fmt fail: run cargo fmt --all, recheck. Windows lock: cargo test --target-dir target/codex-test, report.
• CSS change (static/index.html, static/input.css): npm run build:css; git diff --exit-code static/tailwind.min.css.
• JS change (static/app.js): node --check static/app.js.
• Never commit failing checks.

Runtime Proof

• Default proof: cargo run --bin xtask -- mock-deploy.
• Any implementation change MUST add/update cargo run --bin xtask -- mock-deploy proof.
• Proof MUST verify outcome and stress changed runtime path; performance/resource claims MUST measure.
• Unit tests never replace runtime proof.
• Any change MUST use full build/deploy proof; never use --skip-build --skip-deploy.
• Cloudflare/ACME/HTTPS off unless .deployenv enables them.

Architecture

• UI embedded via include_str! / include_bytes!; no runtime UI files.
• DNS/API hot-reload uses shared Arc<RwLock<DomainStore>> / Arc<RwLock<RewriteMap>>.
• Never hold parking_lot guards across .await; clone/copy first.
• Async DB uses tokio::task::spawn_blocking.

DNS

• Normalize via normalize_domain().
• DB wildcards keep *.; runtime wildcards store suffix.
• build() borrows records; records outlive send_response.
• Prefer send_response for ResponseInfo; fallback uses ResponseInfo::from(Header).