--- name: md-cursor-e2e description: > Use when user invokes /md-cursor-e2e, asks to write new Playwright E2E tests, debug failing specs, or add test coverage for a Multideal feature. Also when tests are flaky and need cursor-agent to diagnose and fix. --- # Multideal cursor-agent E2E cursor-agent does all test work — writes, runs, debugs, fixes. Claude scopes task, reads output, judges correctness, re-instructs if needed. **REQUIRED:** Read `md-verify` for auth patterns, selectors, hard-won lessons. --- ## cursor-agent invocation **NEVER** `run_in_background: true` — Claude must read output to know when done. Run foreground via `ctx_execute` (no timeout ceiling): ``` mcp__plugin_context-mode_context-mode__ctx_execute( language: "shell", code: `/home/user/Projects/multideal/.claude/skills/md-cursor-e2e/ca.sh --workspace /home/user/Projects/multideal/apps/web --trust "PROMPT" --task-slug ""` ) ``` - `--task-slug` is the test/feature identifier (e.g., "checkout-flow", "vendor-onboard") — appended to log filename for easy monitoring - cursor-agent has full tool access: reads, writes files, runs shell commands --- ## Your role (Claude) 1. Scope task → write cursor-agent prompt 2. Run cursor-agent → read output 3. Verify: correct spec? Tests pass? 4. Tests failed or spec wrong → re-instruct cursor-agent with specific fix 5. Repeat until tests pass and behavior correct Never write test code. Never run playwright. --- ## Phase 0 — Scope (Claude) Before dispatching: 1. Read existing specs in `tests/e2e/` — check if area already covered 2. Decide localhost vs live: **default localhost** (no budget). Live only for deployed-specific behavior (live payments, CSP headers, Firebase OTP) 3. Identify auth role: customer / vendor / admin --- ## Phase 1 — Write + Run (cursor-agent) Dispatch cursor-agent to write spec AND run it: ``` You are writing and running Playwright E2E tests for the Multideal project — a Hebrew-first PWA marketplace. Your job: write the spec, then run it, then report results. ## Workspace layout (relative to apps/web/) - tests/e2e/ — all spec files - tests/e2e/fixtures/ — auth-helper.ts, seed-live-upsert.ts - tests/e2e/_utils/ — index.ts (shared helpers) - tests/e2e/_live-guard.mjs — import at top of any spec hardcoding dev.multi.deal URLs - playwright.config.ts — localhost (default, no budget cost) - playwright.live.config.ts — dev.multi.deal (set ALLOW_LIVE_E2E=1) ## Auth — NEVER call /api/auth/login-email (WAF blocks in tests) import { injectUser1Session, injectVendorSession, injectAdminSession } from './fixtures/auth-helper'; await injectUser1Session(context); // customer await injectVendorSession(context); // vendor await injectAdminSession(context); // admin ## Hard rules for spec writing - waitUntil: 'domcontentloaded' — networkidle hangs on CF Worker + PWA - Selectors: role/label/data-testid — never CSS classes - Never waitForTimeout — use waitForSelector or condition polls - RTL app, he-IL locale, Asia/Jerusalem timezone - Import from @playwright/test, not playwright - Import paths relative to the spec file location ## After writing the spec, run it: [LOCALHOST] pnpm exec playwright test tests/e2e/.spec.ts [LIVE] ALLOW_LIVE_E2E=1 pnpm exec playwright test tests/e2e/.spec.ts --config=playwright.live.config.ts ## Task ## Report back - The spec file path you created - Full test output (pass/fail per test) - Full error + stack for any failures - Any assumptions you made ``` --- ## Phase 2 — Verify (Claude) Read cursor-agent output. Check: - Tests pass? → done (unless behavior looks wrong) - Tests fail? → Phase 3 - Spec missing coverage? → re-instruct Phase 1 with more specific task --- ## Phase 3 — Fix (cursor-agent) Send cursor-agent back with failure: ``` You wrote a Playwright spec for Multideal that is now failing. ## Spec file tests/e2e/.spec.ts ## Failure output ## Fix rules - Read the source component or API route the test exercises before changing anything - Prefer fixing the test (selector, timing, assertion) over app code - If app code is genuinely broken, call it out explicitly — do NOT silently fix both - After fixing, re-run the spec and report results ## Report back - What was wrong - What you changed (test vs app code) - Full test output after fix ``` --- ## Routing | User says | Start at | |-----------|----------| | "write e2e for /vendor/settings" | Phase 0 → 1 | | "fix failing e2e tests" | Phase 1 with existing spec + run instruction | | "tests are flaky" | Phase 3 with flakiness description + spec path | | "add coverage for checkout" | Phase 0 → 1 | --- ## Quick reference | Thing | Path | |-------|-------| | Auth helper | `tests/e2e/fixtures/auth-helper.ts` | | Live guard | `tests/e2e/_live-guard.mjs` | | Run (local) | `pnpm exec playwright test ` | | Run (live) | `ALLOW_LIVE_E2E=1 pnpm exec playwright test --config=playwright.live.config.ts` | | Regression baseline | `MODE=baseline bash tests/e2e/visual-regression/run.sh` | | Regression compare | `bash tests/e2e/visual-regression/run.sh` | | Baseline PNGs | `tests/e2e/visual-regression/visual-regression.spec.ts-snapshots/` | ## Regression Baseline Before dispatching cursor-agent (or any agent) for UI changes, capture a visual baseline against `dev.multi.deal`: ```bash cd apps/web MODE=baseline bash tests/e2e/visual-regression/run.sh ``` After deploy, run compare to detect regressions (default `MODE=compare`): ```bash cd apps/web bash tests/e2e/visual-regression/run.sh ``` Baseline PNGs live in `tests/e2e/visual-regression/visual-regression.spec.ts-snapshots/` (Playwright `toHaveScreenshot`). On failure, actual/diff PNGs are in `test-results/`. Requires `ALLOW_LIVE_E2E=1` (set by `run.sh`). ## Learned Rules ### ctx-execute-language-shell | fired:1 | 2026-06-10 `ctx_execute(language:"bash", ...)` → wrong, rejected: enum has no "bash". Use `language:"shell"`. Prevent: always pass `language:"shell"` when running ca.sh / any shell via ctx_execute (valid enum: javascript|typescript|python|shell|ruby|go|rust|php|perl|r|elixir|csharp).