intelli-verse-x Browser Automation
Inspired by curated Browser Automation with playwright-cli patterns, adapted for Content Factory path truth (no phantom src/).
When to use
- Live UI interaction (click, fill, assert, multi-tab)
- Frontend E2E create/extend/debug
- Auth/session UI debugging with screenshots
When NOT to use
| Need | Prefer |
|---|---|
| Scrape / extract page text for research | contentx-firecrawl-validation / Firecrawl MCP |
| Static HTML fact-check | Firecrawl |
| API contract tests | testing skill + pytest |
Prefer project E2E first
Existing CF frontend Playwright setup:
- Config:
frontend/playwright.config.js - Specs:
frontend/e2e/specs/(smoke,login,pipelines,publish,audit) - CI:
.github/workflows/e2e.yml
Extend those specs before inventing new harnesses. Do not invent src/ or src/contentx/.
Core patterns (playwright-cli / Playwright)
1. Navigate → wait for load / networkidle (or role-based ready)
2. Locate → getByRole / getByLabel / getByTestId (avoid brittle CSS)
3. Act → click, fill, select, press
4. Assert → expect visible / URL / text
5. Evidence → screenshot on failure; optional trace
| Action | Guidance |
|---|---|
| Navigate | Prefer absolute app URL from config/env; wait for stable selector |
| Click / fill | Role + name; fill then blur/tab if needed |
| Screenshot | Full page on fail; store under out/ or .working_dir/ |
| Extract | Text/attributes only when UI interaction required; else Firecrawl |
| Multi-tab | context.newPage() / popup listeners; close extras when done |
Flake control
- Prefer auto-waiting locators over fixed
sleep - Retry flaky steps once (max 2 attempts) with fresh navigation
- Isolate shared state (unique user data, no order dependence)
- Soft-timeouts: fail with screenshot + console/network snippet
- Mark known flakes in the spec comment; do not mute asserts permanently
Auth / session caution
- Never commit cookies, storage state, or passwords
- Prefer test accounts / env-injected secrets (K8s secret / local
.env— never print) - Do not automate production admin destructive actions without explicit user APPROVE
- Clear storage between tests when auth leaks cause flakes
Agent checklist
- Confirmed UI interaction needed (not scrape-only)
- Checked
frontend/e2e/specs/for reuse - Locators are role/label based
- Failure path captures screenshot
- No secrets in repo or logs
Related
.cursor/skills/testing/SKILL.md— test pyramid / E2E placement.cursor/skills/contentx-firecrawl-validation/SKILL.md— scrape & fact-check