Memoriant Performance Test Skill

A Claude Code plugin that generates complete load test plans from service profiles. Describe your service's traffic shape and SLOs — get back steady-state, burst, and soak test configurations with stage durations, SLO-aligned checks, and metrics to watch. Interprets results from previous runs and appends tamper-evident evidence records.

No servers. No Docker. Just install and use.

Install

/install NathanMaine/memoriant-perf-test-skill

Cross-Platform Support

Claude Code (Primary)

/install NathanMaine/memoriant-perf-test-skill

OpenAI Codex CLI

git clone https://github.com/NathanMaine/memoriant-perf-test-skill.git ~/.codex/skills/perf-test
codex --enable skills

Gemini CLI

gemini extensions install https://github.com/NathanMaine/memoriant-perf-test-skill.git --consent

Skills

Skill	Command	What It Does
Performance Test	/perf-test	Generate steady, burst, and soak test plans from a service profile. Interpret metrics results. Log JSONL evidence.

Agent

Agent	Best Model	Specialty
Performance Test Planner	Sonnet 4.6	Load plan generation, SLO validation, metrics interpretation, evidence logging

Quick Start

# Generate a load test plan
/perf-test

# Or trigger directly
"Generate a load test plan for our checkout API — 50 RPS baseline, 200 peak, burst factor 3. SLOs: p95 < 300ms, error rate < 1%."

# Interpret results from a previous run
"Here are my soak test results: p95=480ms, p99=950ms, error_rate=0.008. SLO is p95<300, p99<600."

What You Get

For each service profile, the skill generates three test scenarios:

Scenario	Purpose	Duration
Steady-state	Validate normal load meets SLOs	~13 minutes
Burst	Validate graceful degradation under spikes	~9 minutes
Soak	Detect memory leaks and gradual degradation	~70 minutes

Each scenario includes:

• Stage-by-stage RPS targets and durations
• SLO-aligned pass/fail checks
• Metrics to watch during the test
• Clear pass criteria

Service Profile Format

service: checkout-api
summary: Handles checkout flows for web and mobile clients.
traffic:
  baseline_rps: 50
  peak_rps: 200
  burst_factor: 3
slo:
  latency_ms:
    p95: 300
    p99: 600
  error_rate: 0.01
endpoints:
  - path: /cart/submit
    method: POST
    critical: true

Evidence Logging

Every plan generation or metrics interpretation appends a record to a local JSONL evidence log:

{"ts":"2025-03-25T14:30:00Z","service":"checkout-api","scenarios":["steady","burst","soak"],"interpretation":false,"outcome":"plan-generated","plan_hash":"sha256:..."}

The log is append-only — suitable for compliance and sprint definition-of-done verification.

Source

Built from NathanMaine/agent-perf-test-generator.

License

MIT