feat(ai-eval): add E2E failure evaluation skill and Slack formatter

[amisskii]

Summary

• Adds Claude Code skill (ai-eval/skills/evaluate-e2e-results/SKILL.md) for automated classification of Playwright E2E test failures
• Adds Slack Block Kit formatter (scripts/slack_formatter.py) as optional post-processing step

Evaluation skill

• Parses json-results.json, error-context.md, and Playwright trace archives
• 3-layer analysis: error-context.md → json-results.json errors → trace analysis
• Classifies failures as regression | test_bug | flaky | infra | unknown with confidence levels
• Evidence-grounded root causes — escalates to trace analysis when Layer 1 shows WHAT failed but not WHY
• Skipped test analysis — distinguishes by-design skips from cascade skips caused by failures
• Handles missing json-results.json gracefully (falls back to JUnit XML)
• Designed for --print mode with read-only tools in CI pipelines

Slack formatter

• Transforms evaluation JSON into Slack Block Kit message
• Shows classification emoji, root cause, recommended action per failure
• Cascade skip summary when failures cause additional test skips
• Links to artifacts and CI run

Test plan

• Run skill locally against sample test artifacts with claude --print --append-system-prompt-file
• Verify JSON output is valid and parseable
• Verify Slack formatter produces valid Block Kit JSON
• Test with all-pass, single-failure, and multi-failure test runs

Related: containers/podman-desktop-internal#1030

🤖 Generated with Claude Code