# Nw Optimize Tests

> Consolidates a test scope to the fewest tests that preserve coverage and behavior, deleting duplication, parametrize inflation, language-guarantee tests, AST-shape tests, and stale migration nets. An independent read-only reviewer with veto always validates the result.

- Skill: `nwave-ai/nw-optimize-tests-2` (Agent Skill)
- Install (CLI): `npx skillmds@latest add nwave-ai/nw-optimize-tests-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/nwave-ai/nw-optimize-tests-2/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: nWave-ai (https://skillmd.com/u/nwave-ai)
- Updated: 2026-09-21
- Page: https://skillmd.com/skills/nwave-ai/nw-optimize-tests-2

---


# NW-OPTIMIZE-TESTS

Act when invoked. Do not wait for approval, do not present a plan table, and do
not produce a report artifact.

Dispatch `nw-test-optimizer` over `{scope}` (methodology: `nw-test-optimization`).

Invariants:

- Run the exact project-declared command vector before and after, in the same
  environment and over the same scope. Compare terminal process timing only.
- Production files in the diff: zero.
- Coverage and observed behavior preserved; any drop is justified in the reply.
- Consolidate by deletion, never by adding a test to cover a removed one.
- Tests must prove observable behavior, never source prose or AST shape.

After the mutation, always dispatch `nw-test-optimizer-reviewer` over the
resulting diff. It reads only, and it holds veto: production drift, coverage or
behavior loss, and any deletion not mapped to a stated pattern are refusals the
optimizer must repair before the work is done. Its verdict is prose in the
reply, not an artifact or a schema.

Out of scope: authoring new tests, production refactoring, test infrastructure.