AgentSkillsCN

Load Test Designer

以贴近实际的负载模型与性能指标，精心设计负载测试方案。

SKILL.md
--- frontmatter
name: Load Test Designer
description: Design load tests with realistic workload models and performance criteria
category: testing
version: 1.0.0
triggers:
  - release-preparation
  - performance-baseline
  - capacity-validation
globs: "**/tests/load/**,**/k6/**,**/benchmarks/**"

Load Test Designer Skill

Design load tests with realistic workload models and performance criteria.

Trigger Conditions

  • •Pre-release performance gate
  • •Performance baseline request
  • •User invokes with "design load test" or "performance test"

Input Contract

  • •Required: System under test and target endpoints
  • •Required: Performance targets (latency SLOs, throughput)
  • •Optional: Production traffic profile, expected growth

Output Contract

  • •Load test configuration with workload model
  • •Performance criteria (pass/fail thresholds)
  • •Results analysis with regression detection
  • •Capacity recommendations

Tool Permissions

  • •Read: Production access logs, performance baselines, infra configs
  • •Write: Load test configs, results reports
  • •Execute: Load test tools (k6, locust, artillery)

Execution Steps

  1. •Build workload model from production traffic patterns
  2. •Define performance criteria (p50, p95, p99 latency targets)
  3. •Configure load test scenarios (ramp-up, sustained, spike)
  4. •Execute tests in staging environment
  5. •Collect and analyze results
  6. •Compare against baseline; flag regressions >5% p99
  7. •Generate report with capacity recommendations

Success Criteria

  • •Workload model reflects production patterns
  • •All SLO targets met during sustained load
  • •No regression >5% on p99 latency
  • •Saturation point identified

Escalation Rules

  • •Escalate if p99 regression >20% on critical path
  • •Escalate if saturation point is below expected peak traffic
  • •Escalate if soak test reveals memory leak or connection exhaustion

Example Invocations

Input: "Design a load test for the /api/v1/search endpoint targeting 1000 RPS"

Output: Load test config (k6): ramp 0→1000 RPS over 5min, sustain 1000 RPS for 15min, spike to 2000 RPS for 2min. Pass criteria: p50 <100ms, p95 <300ms, p99 <800ms, error rate <0.1%. Saturation test: ramp until p99 >1s. Soak: 500 RPS for 8 hours monitoring heap and FD count.