fix(cdp): accept Chromium major-version range; e2e tests fail loud

[aram-devdocs]

Summary

• Replaces the hard PINNED_CHROMIUM_MAJOR == 131 equality check with a MIN_SUPPORTED_CHROMIUM_MAJOR..=MAX_SUPPORTED_CHROMIUM_MAJOR range (131..=150), unblocking plumb lint <real-url> on every reasonably current Chrome/Chromium release. Closes fix(cdp): hard Chromium 131 pin blocks all real-URL lints on modern systems #117.
• Makes the e2e-chromium test path fail loudly when no usable Chromium is on the host. The previous code silently treated ChromiumNotFound / UnsupportedChromium as a pass; it now propagates the error unless the user explicitly opts into the skip via PLUMB_E2E_CHROMIUM_SKIP=1 (with a tracing::warn! line). Closes test(cdp): e2e-chromium tests silently pass when Chromium is missing or wrong version #118.
• Updates docs (install-chromium.md, cli.md, runbook spec, AGENTS/CLAUDE files) to describe the new range.

Test plan

• cargo fmt --all -- --check
• cargo clippy --workspace --all-targets --all-features -- -D warnings
• cargo test --workspace
• cargo test -p plumb-cdp --test driver_contract --features e2e-chromium (Chromium 139 on host, in range, real e2e pass)
• just determinism-check — byte-identical across 3 runs
• cargo deny check
• ./target/debug/plumb lint https://example.com --format json — emits real a11y/touch-target violation against live example.com page (acceptance criterion for Phase 1 / [RUNBOOK] Phase 1 — Real CDP driver + foundational rules #10)

🤖 Generated with Claude Code

[github-actions]

Here is the full review.

---

Review: fix(cdp): accept Chromium major-version range + follow-up commits

Commits on branch: 4f59489, 75f0b0c, b0fb3c1.

---

Determinism — PASS

The range predicate (MIN_SUPPORTED_CHROMIUM_MAJOR..=MAX_SUPPORTED_CHROMIUM_MAJOR).contains(&found) and the updated #[error(...)] format string are both pure functions of their inputs. No wall-clock, no HashMap, no env-var leaking into snapshot output. Determinism invariants hold.

---

Workspace layering — PASS

tracing-subscriber is added only to [dev-dependencies] in crates/plumb-cdp/Cargo.toml and its usage is entirely inside #[cfg(feature = "e2e-chromium")]-gated test code. No new internal dependency edges, no unsafe added outside plumb-cdp. Layer discipline intact.

---

Error handling — PASS

No unwrap/expect introduced in library code. The new CdpError::UnsupportedChromium variant is properly derived via thiserror. The chromium_install_hint / validate_chromium_product_major path is clean.

---

Test coverage — CONDITIONAL PASS

The rewritten unit tests (detects_unsupported_chromium_major, accepts_supported_chromium_majors) correctly test below-min, above-max, and an in-range value. One fragility:

crates/plumb-cdp/src/lib.rs:1153

let in_range = "HeadlessChrome/140.0.0.0";

140 is a magic literal, not derived from the constants. If MAX_SUPPORTED_CHROMIUM_MAJOR is ever reduced below 140, this test either silently passes with the wrong semantics or fails with an opaque assertion message. Should be:

let mid = (MIN_SUPPORTED_CHROMIUM_MAJOR + MAX_SUPPORTED_CHROMIUM_MAJOR) / 2;
let in_range = format!("HeadlessChrome/{mid}.0.0.0");

---

Documentation — PASS (with caveat)

All public items (MIN_SUPPORTED_CHROMIUM_MAJOR, MAX_SUPPORTED_CHROMIUM_MAJOR, CdpError::UnsupportedChromium fields) have docs. install-chromium.md, cli.md, CHANGELOG.md, and both AGENTS.md files are updated consistently. The --executable-path comment at lib.rs:432 documents a non-obvious invariant (the override doesn't bypass the version check) — appropriate use of a comment.

Unverifiable: docs/local/prd.md is gitignored. The plumb-cdp/AGENTS.md invariant says the range "must match the PRD (§9, §16)." This cannot be confirmed from the diff. The author should assert in the PR description that §9 and §16 have been updated.

---

Missing ADR — BLOCKER

The original crates/plumb-cdp/AGENTS.md (before this PR) states explicitly:

The pinned Chromium major version (PINNED_CHROMIUM_MAJOR) must match the PRD (§9, §16) and the docs. Changing it needs an ADR.

This PR does not merely bump the pin — it replaces the exact-pin mechanism with a range, which is a larger architectural decision: accepting a 20-major window trades away strict rendering reproducibility for host compatibility. That trade-off is exactly what the ADR requirement exists to capture: the context, alternatives considered, and the re-validation commitment when the upper bound moves.

docs/adr/ currently contains only 0001-bootstrap-conventions.md. No new ADR (e.g., 0002-chromium-version-range.md) appears in the diff. The updated AGENTS.md says future range shifts need an ADR — but that clause doesn't retroactively cover the current shift.

---

Punch list

#	File:Line	Severity	Issue
1	docs/adr/ (missing file)	BLOCKER	No ADR for pin→range architectural change; plumb-cdp/AGENTS.md rule explicitly requires one.
2	crates/plumb-cdp/src/lib.rs:1153	WARNING	Hardcoded 140 literal — derive from constants to avoid silent drift.
3	docs/local/prd.md (not in diff)	WARNING	PRD §9/§16 alignment unverifiable; assert in PR description that it was updated.

---

Verdict: REQUEST_CHANGES