test-and-verify-walnut

# /test-and-verify-walnut Every test must answer: **"If I reverted my code changes, would this test fail?"** NO → delete it. **FIRST**: Read `.claude/skills/walnut-console-ops/SKILL.md` for UI layout and interaction patterns. ``` Main Agent (context) → Agent 1 (design) → Agent 2 (code) ∥ Agent 3 (Playwright) → Agent 4 (quality gate) ``` --- ## Phase 0: Main Agent — Context 1. `git diff --stat HEAD~1` + `git log --oneline -5` 2. Read plan files: `.plan`, `~/.claude/plans/`, `.tasks/*/TASK.md` 3. Classify change type: | Change Type | Primary Tests | |---|---| | Frontend-only | Playwright UI | | Backend-only | API + unit + server logs | | Full-stack | Console E2E + Playwright + unit | | Bug fix | Reproduce → fix → re-verify | 4. Bundle context → pass to all agents. --- ## Phase 1: Agent 1 — Test Designer (read-only) Design tests in 2 categories. Do NOT implement. **Category A — UI (Playwright)**: 2-4 scenarios. For each: pre-conditions, steps (real clicks, NO `page.goto()` SPA nav), assertions (must include downstream verification — see below), screenshot points. **Category B — Code (vitest)**: 2-4 tests, as E2E as possible. For each: name, tier, setup, exercise, assert (HTTP response + WS event + persisted data). **Self-check each test**: "Would this pass with code reverted?" YES = delete. **E2E means full consequence**: Every test must verify the **downstream effect** of the action, not just the immediate UI change. If the action triggers a backend round-trip, t...