AgentSkillsCN

skyline-nightlytests

在查看夜间测试结果或调查测试失败时,可运用这一技能。

SKILL.md
--- frontmatter
name: skyline-nightlytests
description: Use this skill when looking at nightly test results or investigating test failures.

Skyline Nightly Test Analysis

Full documentation: ai/docs/mcp/nightly-tests.md

MCP Tools (Quick Reference)

ToolPurpose
get_daily_test_summary(report_date)Daily report for all 6 folders → ai/.tmp/
save_test_failure_history(test_name, start_date, container_path)Stack trace pattern analysis → ai/.tmp/
save_run_log(run_id)Full test log for grep/search → ai/.tmp/
get_run_failures(run_id)Stack traces for a specific run
get_run_leaks(run_id)Leak details for a specific run

Test Folders

FolderDurationBranch
Nightly x64540 minmaster (default)
Performance Tests720 minmaster
Release Branch540 minrelease
Release Branch Performance Tests720 minrelease
Integration540 minassigned
Integration with Perf Tests720 minassigned

Typical Workflow

  1. Daily review: get_daily_test_summary("2025-12-15") → read report
  2. Pattern analysis: If test failed on multiple machines, use save_test_failure_history to check if same root cause
  3. Deep dive: Use save_run_log when stack traces aren't sufficient

Anomaly Detection

  • Short duration (< expected) = crash or premature termination
  • Full duration + low test count = hang
  • Stddev-based: 3σ/4σ from trained mean flags anomalies

For server-side queries, examples, and detailed workflows, see ai/docs/mcp/nightly-tests.md.