Do not fail successful browser probes when browser close exceeds cleanup timeout

[chubes4]

Problem

wordpress.editor-open and wordpress.editor-validate-blocks can complete navigation, readiness, and artifact capture successfully, then report the whole recipe step as failed because browser cleanup exceeds the fixed 1000ms close timeout. A seven-surface fixture run recorded four runtime_execution_failed findings solely from operation=browser-close-timeout.

Evidence

• Homeboy run: b280bf88-2257-4915-9d80-f4c52ff6a538
• WP Codebox: 0.18.4
• Browser: HeadlessChrome 151.0.7922.34
• Fixture: 37-art-gallery-exhibition
• Affected steps: editor-open and editor-validate-blocks
• Captured summaries show navigation/readiness succeeded and artifacts were written before close cleanup timed out.

Reproduction:

node tools/run-fixture-matrix.mjs --static-site-importer /path/to/static-site-importer --blocks-engine /path/to/blocks-engine --local --target-fixture 37-art-gallery-exhibition --surface-coverage 6

Acceptance

• Browser close cleanup has a realistic bounded timeout or terminates the browser process deterministically.
• A cleanup timeout retains diagnostics without converting an otherwise successful probe and its valid artifacts into a workload failure.
• Repeated multi-surface editor-open and editor-validate-blocks runs complete without close-timeout findings.

AI Assistance

OpenAI GPT-5.6 Sol via OpenCode ran the seven-surface matrix, inspected retained editor artifacts, and drafted this report. Chris Huber reviewed the evidence and remains responsible for the issue.

[chubes4]

Fixture-matrix evidence on released WP Codebox 0.18.4 confirms this affects completed wordpress.editor-open and wordpress.editor-validate-blocks operations.

Transformer/content was immutable across runs. Observed:

• Round7 first run: validation completed, teardown failed once with operation=browser-close-timeout cause=Browser close exceeded 1000ms.
• Identical retry: succeeded, 346/346 blocks valid.
• Round9: both editor-open and editor-validation completed their browser work but failed teardown with the same 1000ms close timeout; visual compare still completed at 3.53%.

Artifacts: /tmp/fixture-89-combined-v0184-round7, /tmp/fixture-89-combined-v0184-round7-retry, /tmp/fixture-89-combined-v0184-round9.

This confirms cleanup timing is nondeterministic and currently invalidates successful command evidence.

AI assistance: OpenAI GPT-5.6 Sol using OpenCode isolated the teardown failure from transformer/editor validity and drafted this comment. Chris reviewed the evidence.