Test Perf K6

k6 Performance Testing (workflow)

Operating mode (token-efficient)

• Treat this skill as a router + governor.
• Do not load multiple procedures. Select exactly one procedure below and follow it end-to-end.
• Optimize for reproducible performance signals (thresholds + stable environments).

Routing (pick one procedure)

Shared non-negotiables (apply to all procedures)

1. Versioned scenarios

   • Store k6 scripts in-repo and review like code.
   • Keep scenarios small and composable.

2. Explicit thresholds

   • Every CI-gated scenario must have thresholds (p95/p99 latency, error rate).
   • Thresholds must be justified and documented.

3. Environment control

   • Prefer stable staging environments for consistent results.
   • Do not run load/stress tests against production unless explicitly approved.

4. Artifact contract (for CI + trend analysis)

   • Standardize under: artifacts/k6/
   • Export at least a JSON summary (and optionally JUnit-like output if you convert).

5. No secrets in repo

   • Inject tokens via CI secrets / env vars.
   • Never commit API tokens in scripts.

Minimal inputs you should capture before changing code

• Target endpoint(s) and base URL(s)
• Auth strategy for load tests (token acquisition and rotation strategy)
• Target SLOs / acceptable latency and error rates
• Load model (users, arrival rate, duration)
• Whether tests are PR-gating vs scheduled (nightly)

Verification

• If you changed skills:

  • Prefer host-repo tooling if present:
    • node .ai/scripts/lint-skills.mjs --strict
  • Always run the local validator:
    • node .ai/skills/testing/test-perf-k6/scripts/validate-skill.mjs

• If you changed perf scripts:

  • k6 version (or docker run --rm grafana/k6 version)
  • k6 run <script.mjs>

Boundaries

• Do not edit .codex/skills/ or .claude/skills/ directly (generated).
• Do not set thresholds so loose that regressions pass unnoticed.
• Do not run high-load tests without coordinating capacity and rate limits.