test(idea-posterior): budget browser-probe timeouts for slow CI Chrome starts

[fkguo]

Problem

skills/idea-posterior/tests/test_reader_surface_interactions.py::test_interactive_visible_probabilities_share_static_precision fails on GitHub Actions (job python-contracts-test, step "Python tests (interactive argument-graph reader surface)") since ~2026-08-07 with subprocess.TimeoutExpired ... timed out after 5 seconds while launching google-chrome --headless=new ... --virtual-time-budget=3000 --dump-dom, preceded by a D-Bus connect error on stderr. PR #49 run 31140161012 hit it twice (fresh run + rerun, different runner instances, identical failure); the same test on the same code passes locally. Main's last green CI was 2026-08-03, so the regression tracks a runner-image/Chrome update, not a code change.

Diagnosis

The uncaught exception saying "after 5 seconds" can only originate from the post-SIGTERM cleanup communicate(timeout=5): the main communicate(timeout=15)'s own TimeoutExpired is always caught by the surrounding except. So on the updated runner Chrome first stalled past the 15 s main timeout (consistent with a startup D-Bus connect wait; the libdbus default method-call timeout is 25 s), was SIGTERMed, and then also failed to exit within 5 s. Raising only the 5 s cleanup value would therefore not make CI green — Chrome would still be killed at 15 s mid-startup, --dump-dom would never emit, and the test would fail on assert stdout, stderr instead.

Fix

Raise both waits to 30 s (main communicate 15 → 30, post-SIGTERM cleanup 5 → 30). This does not weaken what the test asserts: the in-page work is bounded by Chrome's --virtual-time-budget=3000 (virtual ms), communicate() returns as soon as the process exits, and every DOM-content assertion (probability precision, control inventory, reader-surface contract) is untouched. The wall-clock values are purely a guard against a wedged browser.

Repo-wide, this file is the only place that launches a browser subprocess with fixed timeouts (grep -rn "communicate(timeout=" — the two review-swarm hits are parameterized launcher code, not browser tests).

Verification

• python3 -m pytest skills/idea-posterior/tests/test_reader_surface_interactions.py -q → 2 passed locally
• node scripts/check-idea-posterior-reader-surface-anti-drift.mjs → passes (the lock pins interaction/contract anchors, not timeout values, and forbids pytest.skip here — so "skip on CI" was never a legal alternative)
• git diff --check clean
• Independent read-only review of the diff returned SHIP with two non-blocking notes, dispositioned as follows:
  1. 30 s is calibrated to a single ~25 s D-Bus stall and cannot be validated locally (macOS has no D-Bus); this PR's own CI run is the acceptance check. If the runner ever stalls on two buses (~50 s), the right next step is an environment-level fix (e.g. point DBUS_SESSION_BUS_ADDRESS at an unreachable address so connects fail fast), not a larger number.
  2. The cleanup path has no SIGKILL escalation (pre-existing shape, unchanged here). Left out to keep the reviewed diff minimal; on an ephemeral CI VM the residual process-group leak is inert. skills/review-swarm/scripts/bin/run_multi_task.py:834 already has the stronger pattern if a follow-up wants it.

Note for PR #49

PR #49's python-contracts-test failure is this same environment issue, not a defect in that branch. It should clear once this lands on main and #49's CI runs against an updated merge base (branch update / rebase / new push — a plain re-run of the old run reuses the old merge commit).

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.