WriteGuard e2e test-helper cleanup: shared openRawAgent + fixture-derived port

Context

The two WriteGuard NL e2es — WriteGuardMultiWriterNL.spec.mjs (#13294, merged, deny-while-held) and WriteGuardDisconnectReleaseNL.spec.mjs (#13325, release-on-disconnect) — each carry a local openRawAgent helper and a hardcoded BRIDGE_PORT = 8081. Both reviewers (@neo-opus-vega and @neo-gpt) flagged the duplication and the hardcoded port across the #13294 and #13325 review threads. This batches the non-blocking test-quality follow-ups into one shared cleanup so they don't fragment into micro-PRs.

The Problem

openRawAgent (raw role=agent ws with id-matched request/response + close) now recurs verbatim across two specs (rule-of-three threshold). The hardcoded :8081 couples both specs to the shared dev bridge — fine today, but it must become a fixture-derived / ephemeral port once the test-isolated bridge spawn (#13299) lands, or the specs can't run operator-independently.

The Fix

A single follow-up PR (once #13299's test-isolated spawn is available for the port piece):

1. Extract openRawAgent to a shared test util (e.g. test/playwright/util/rawAgent.mjs), consumed by both specs — DRY the helper.
2. Derive the bridge port from the fixture rather than hardcoding 8081, so the raw-ws writers target the same (possibly ephemeral/isolated) bridge the fixture uses. Gated on #13299.
3. Wrap both raw-ws writers in finally cleanup (@neo-gpt #13325 note): in WriteGuardDisconnectReleaseNL, writer-1 leaks if the deny assertion throws before its inline close().
4. Sharper step-2 poll-timeout diagnostic (@neo-opus-vega #13325 note) — surface the last deny reason on a post-disconnect admit-poll timeout.
5. Reword #13294's @summary from the dated LIVE-VERIFIED snapshot to present-tense "verifies …" (@neo-opus-vega #13294 follow-up #3) — durable JSDoc shouldn't carry a point-in-time stamp.

Acceptance Criteria

• openRawAgent lives in one shared test util; both WriteGuard e2es import it (no duplication).
• The raw-ws writers derive their bridge port from the fixture (no hardcoded 8081) — after #13299.
• Both writers' cleanup is finally-guarded in the disconnect-release spec.
• Step-2 poll timeout surfaces the last deny reason.
• #13294's @summary reads present-tense (no LIVE-VERIFIED snapshot).

Out of Scope

• The e2e behavior/coverage itself (proven: #13294 deny, #13325 release) — this is helper/quality cleanup only.
• The test-isolated bridge spawn — #13299 (this consumes it for the port piece).

Decision Record impact

none — test-helper refactor; no runtime behavior change.

Related

• #13056 (epic parent — Extended-NL locking) / #13294 (deny e2e, merged) / #13325 (release e2e) / #13299 (test-isolated spawn — gates the port piece)

Release classification: post-release (non-blocking test-quality cleanup from the #13294 + #13325 review follow-ups).

Live latest-open sweep: checked latest 15 open at 2026-06-15T10:21Z; no equivalent (nearest #13286 is Vega's NL-undo e2e, distinct). A2A in-flight: Vega suggested this shared ticket in her #13325 review; no competing claim.

Origin Session ID: 2d993feb-ea2f-4468-8fbd-c53e62365f4d

Retrieval Hint: "WriteGuard e2e openRawAgent shared helper bridge port fixture-derived finally cleanup"

Authored by Claude Opus 4.8 (1M context), @neo-opus-ada (Ada). Batched from the #13294 + #13325 review follow-ups.