e2e-flow-verifier（端到端流量验证器）

E2E Flow Verifier

Product verification skill that drives user flows with the qe-browser fleet skill (Vibium engine), asserts state at each step, and records evidence as a ZIP of screenshots + DOM snapshots.

Activation

/e2e-flow-verifier [flow-name]

Dependency

This skill uses .claude/skills/qe-browser/ for all browser automation. The vibium binary is installed automatically by aqe init. See qe-browser/SKILL.md for the full command reference.

Flow Verification Pattern (qe-browser + batch + assert)

Define the flow as a JSON batch plan and drive it with the qe-browser batch runner:

# flows/{{flow-name}}.json
cat > /tmp/{{flow-name}}.json <<'EOF'
[
  {"action": "go",      "url": "{{base_url}}"},
  {"action": "wait_load"},
  {"action": "fill",    "selector": "[data-testid=email]",    "text": "{{test_user}}"},
  {"action": "fill",    "selector": "[data-testid=password]", "text": "{{test_password}}"},
  {"action": "click",   "selector": "[data-testid=login-btn]"},
  {"action": "wait_url","pattern": "/dashboard"},
  {"action": "assert",  "checks": [
    {"kind": "url_contains",    "text": "/dashboard"},
    {"kind": "selector_visible","selector": "[data-testid=dashboard]"},
    {"kind": "no_console_errors"},
    {"kind": "no_failed_requests"}
  ]},
  {"action": "click",   "selector": "[data-testid={{action_element}}]"},
  {"action": "assert",  "checks": [
    {"kind": "text_visible", "text": "{{expected_text}}"}
  ]}
]
EOF

# Record evidence AND drive the flow
vibium record start --screenshots --snapshots --name "{{flow-name}}"
node .claude/skills/qe-browser/scripts/batch.js --steps "@/tmp/{{flow-name}}.json"
FLOW_EXIT=$?
vibium record stop -o "test-results/{{flow-name}}/evidence.zip"
exit $FLOW_EXIT

The batch runner exits non-zero if any step fails, so CI gates work with a plain $? check.

Evidence Collection

After each flow verification, vibium record produces a ZIP containing:

1. Screenshots — one per action + annotated failure shots
2. DOM snapshots — before/after each interaction
3. Timeline — ordered event log with URLs and timings
4. Console/network logs — captured inline

Use vibium har-export for a separate HAR 1.2 network log if needed.

Asserting backend state alongside UI

The qe-browser assert.js check kinds include network assertions that capture backend calls made from the page:

node .claude/skills/qe-browser/scripts/assert.js --checks '[
  {"kind": "response_status", "url": "/api/{{resource}}", "status": 200},
  {"kind": "request_url_seen", "url": "/api/{{resource}}"}
]'

For API calls that don't originate from the page, run the API check in a separate step (curl + jq, or a dedicated API test skill).

Common Flows to Verify

Reusing auth state across runs

# Once: log in and save state
vibium go "{{base_url}}/login"
vibium fill "[data-testid=email]" "$USERNAME"
vibium fill "[data-testid=password]" "$PASSWORD"
vibium click "[data-testid=login-btn]"
vibium wait url "/dashboard"
vibium storage -o test-results/auth/state.json

# Every subsequent run
vibium storage restore test-results/auth/state.json
vibium go "{{base_url}}/dashboard"

Gotchas

• Re-map after DOM changes — Vibium @e1 refs are invalidated by navigation; use vibium diff map to see what changed and re-map if needed.
• Selectors break on deployment — prefer data-testid over CSS classes.
• Auth tokens expire — use fresh login per run, or rotate storage snapshots.
• Evidence ZIPs grow — keep failure runs only in CI, gc after N days.
• No raw Playwright — this skill used to use @playwright/test. If you need Playwright's network interception (page.route), use it in a separate skill and call it as a step; don't reintroduce the dependency here.