ci: track flaky and slow tests over time

[markijbema]

Retries currently turn transient CLI test failures into green results, while short-lived JUnit artifacts make recurring instability and slow tests difficult to identify.

This adds an Allure Report 2 test-health job that consumes the JUnit XML already produced by Bun and Gradle. Linux, Windows, and JetBrains remain separate so platform-specific outcomes are not combined. Each workflow run publishes three standalone HTML previews containing retries, recent status history, duration trends, and slow-test views.

Successful main-branch runs also deploy a stable GitHub Pages dashboard at https://kilo-org.github.io/kilocode/. Main-branch history is carried forward through GitHub artifacts so flaky, regressed, fixed, and duration trends accumulate without an external analytics service or stored credential. Reporting and deployment remain non-blocking observability aids.

The CLI test runner retains every failed file attempt separately from the final retry-aware JUnit result. Allure receives the final outcome first and prior attempts as retries, so a recovered test remains green while still being marked as a status-changing retry. Allure 2.42.1 is downloaded from its official release and verified against the release SHA-256 before processing results.

[kilo-code-bot]

Code Review Summary

Status: No Issues Found | Recommendation: Merge

Files Reviewed (2 files)

• .github/workflows/test.yml - Added test-health and deploy-test-health jobs with Allure Report 2 integration (260 additions). All changes wrapped in kilocode_change markers.
• packages/opencode/script/test-runner.ts - Refactored to track per-attempt JUnit XML files and produce separate unit and test-health merged outputs.

---

_{Reviewed by deepseek-v4-pro-20260423 · 299,112 tokens}

_{Review guidance: REVIEW.md from base branch main}