feat(e2e): real-world test-site matrix + harness + CI

[aram-devdocs]

Summary

• Adds e2e-sites/ with six framework fixtures (vanilla HTML/CSS, static Tailwind, React + Vite, Vue 3 + Vite, Angular 17, Next.js 14 static export). Each renders the same intentional design-system violations.
• Adds crates/plumb-e2e/ — a dev-only harness that builds each fixture, serves its dist/ over loopback HTTP via tiny_http, runs plumb lint three times, asserts byte-equality, and diffs the target-rule violation breakdown against expected.json.
• Wires three workflows: e2e-sites.yml (18-leg matrix on PR), pages.yml extension (publishes fixtures alongside the docs at https://plumb.aramhammoudeh.com/test-sites/<framework>/), and dogfood.yml extension (re-lints the deployed copies daily).
• Includes a small fix(cli) commit that threads auto_fetch_chromium through plumb watch — the workspace failed to compile after feat(cdp): chromium auto-fetch via chromiumoxide fetcher (#78) #220 landed.

What the harness asserts

For every site, the harness:

1. Optionally runs just build inside the fixture.
2. Spawns a 127.0.0.1:0 HTTP server pointed at dist/.
3. Runs plumb lint http://127.0.0.1:<port>/ --config e2e-sites/plumb.toml --format json three times.
4. Asserts the three outputs are byte-identical.
5. Parses violations and asserts each target_rule count matches expected.json::by_rule_id, plus that the total target-rule count matches total_target_violations.

Non-target rule violations are reported but not asserted on — the matrix narrows the surface to the design-system invariants under test, not every Chromium-rendering quirk.

Local results

PASS  html-css        target_violations=9   non_target=2  by_rule_id={"color/palette-conformance": 1, "spacing/grid-conformance": 4, "spacing/scale-conformance": 4}
PASS  tailwind-html   target_violations=9   non_target=3  by_rule_id={"color/palette-conformance": 1, "spacing/grid-conformance": 4, "spacing/scale-conformance": 4}
PASS  react-vite      target_violations=9   non_target=3  by_rule_id={"color/palette-conformance": 1, "spacing/grid-conformance": 4, "spacing/scale-conformance": 4}
PASS  vue             target_violations=9   non_target=3  by_rule_id={"color/palette-conformance": 1, "spacing/grid-conformance": 4, "spacing/scale-conformance": 4}
PASS  angular         target_violations=9   non_target=3  by_rule_id={"color/palette-conformance": 1, "spacing/grid-conformance": 4, "spacing/scale-conformance": 4}
PASS  nextjs          target_violations=17  non_target=8  by_rule_id={"color/palette-conformance": 1, "spacing/grid-conformance": 8, "spacing/scale-conformance": 8}
OK    6 site(s) passed

just validate is green; just determinism-check is green.

The Next.js fixture's higher count is documented in e2e-sites/nextjs/README.md — Next injects a hidden <next-route-announcer> that uses the visually-hidden margin: -1px trick, contributing 4 grid + 4 scale violations on top of the intentional p-[13px] hero.

Test plan

• CI matrix e2e-sites runs on this PR — every leg should be green.
• Pages preview deploys https://plumb.aramhammoudeh.com/test-sites/<framework>/ for all six fixtures.
• Daily dogfood job lints the deployed copies and asserts the same target-rule counts.
• just test-e2e passes locally with a Chromium binary available.
• just validate stays green.

🤖 Generated with Claude Code

[github-actions]

I now have everything I need. Here is the review.

---

PR #223 — plumb-e2e harness + e2e-sites matrix + dogfood diff

Crates touched

plumb-cli (minor), plumb-e2e (new, publish = false), e2e-sites/ (fixtures), .github/workflows/ (CI), justfile, Cargo.{toml,lock}

---

1. Determinism invariants

No violations. plumb-e2e is a test harness that drives the plumb binary as a subprocess; it has no output path that could sneak nondeterminism into plumb-core. IndexMap is used everywhere a map appears in a public API (runner.rs, expected.rs). No SystemTime::now / Instant::now in library code. The Duration constants in server.rs are compile-time values, not wall-clock reads.

---

2. Workspace layering

Clean. plumb-e2e has no internal plumb-crate dependencies — it treats plumb as a subprocess binary, which is the correct shape for an integration harness. #![forbid(unsafe_code)] is present in both lib.rs and main.rs. The crate is in members but correctly absent from default-members, so cargo build and cargo test --workspace skip it.

---

3. Error handling

No unwrap/expect in library code. Test code uses expect freely; clippy.toml has allow-expect-in-tests = true, so these are exempt. HarnessError is a proper thiserror-derived enum. anyhow is used only in main.rs and in the publish = false harness, which is the equivalent of plumb-cli::main — acceptable.

---

4. Findings

Warning — server.rs:177: dead-code arms in path-traversal guard

.any(|seg| seg == ".." || seg.starts_with("..\\") || seg.starts_with("../"))

After split(['/', '\\']), no segment can ever contain / or \, so starts_with("..\\") and starts_with("../") are permanently unreachable. The load-bearing check is seg == ".." (correct). Dead branches in a security-sensitive function are confusing; simplify to just seg == "..".

Warning — crates/plumb-e2e/Cargo.toml:37: tiny_http pinned directly, not via workspace

tiny_http = "0.12" is not declared in [workspace.dependencies]. Since this is the only consumer and publish = false, cargo-deny still audits the resolved version — no correctness risk. It is inconsistent with workspace convention, where new deps are expected to be routed through the workspace manifest. Minor.

---

5. Test coverage

expected.rs, runner.rs, server.rs, and workspace.rs all have #[cfg(test)] blocks covering the happy path and failure paths. server.rs has a live TCP integration test and a path-traversal rejection test. Coverage is appropriate for the scope.

---

6. Documentation

All public items are documented. Fallible public functions have # Errors sections. CHANGELOG entry is present under ## [Unreleased] and contains no AI-tell phrases.

---

Verdict: APPROVE

[aram-devdocs]

Next.js leg is advisory in 48dc7ab — Chromium times out on the heavier hydration; fix in follow-up.